Build & Sell with Codex (5+ Hour Course) — Transcript
Full transcript
- 0:00I'm about to take you from a complete
- 0:01beginner to a pro AI builder with
- 0:03Codeex. And you don't have to have a
- 0:04technical background at all. My name is
- 0:06Nate. I have no technical background.
- 0:08And I started using AI about two years
- 0:09ago. And in that time, I had an AI
- 0:11consultancy that we scaled to over
- 0:12$100,000 a month. And then I exited that
- 0:14business. And now across my different
- 0:16business units, we generate an average
- 0:17of over $400,000 per month because we
- 0:19all use AI so well. And none of us are
- 0:22actually like coders or anything like
- 0:23that. So in this course, I'm going to
- 0:24show you guys how to install Codex. I'm
- 0:26going to talk about the core concepts.
- 0:27I'm going to talk about how you turn it
- 0:28into your AI operating system and your
- 0:30second brain. We're going to go over
- 0:32things like skills. We're going to build
- 0:33some together. We're going to go over
- 0:34how you create deliverables that are
- 0:36branded and feel like it's coming from
- 0:37you and feel like it's coming from your
- 0:38company. We'll go over images. We'll go
- 0:40over design websites using the browser
- 0:43using Codex to edit videos, using codeex
- 0:44to build automations. We'll even address
- 0:46the whole codeex versus cloud code
- 0:48debate. And by the end, not only will
- 0:49you be a pro AI builder, but you'll also
- 0:51know exactly how you could go off and
- 0:53sell this kind of stuff to clients. So,
- 0:54there are going to be timestamps down
- 0:56below. Feel free to skip around to what
- 0:57interests you. Save this video so you
- 0:58can come back to it for later. And now,
- 1:00let's not waste any time and just get
- 1:01straight into this one. All right, so
- 1:02we're going to go ahead and get started
- 1:03with installation. You're going to open
- 1:04up Google and you're going to type in
- 1:06cloud code. Wait a minute. That was so 3
- 1:08months ago. No, I'm just kidding. But it
- 1:10is funny how Chad GBT also pops up right
- 1:11here. So, you're going to type in Chadbt
- 1:13or Codeex. I'll show you what that looks
- 1:14like. I will just type in codeex
- 1:16install. And then you see right here,
- 1:17ChadBt Codeex because Codex is a product
- 1:20of OpenAI and Chad GBT. So, this is what
- 1:22it looks like. What you're going to want
- 1:23to do is just download this for your
- 1:24operating system. So right here I would
- 1:26click on download for Windows and then I
- 1:27would go through the installer to get
- 1:28the codeex desktop app. Now this
- 1:30actually will be called the chatbt
- 1:32desktop app but we're going to be using
- 1:33codec. Now if you do use VS code already
- 1:35or you use cursor or something like that
- 1:37and you'd rather run codec through the
- 1:38IDE extension or through the CLI then
- 1:41you certainly can do that. But I love
- 1:42the desktop app. I think that they're
- 1:44bringing it in really really nicely with
- 1:45their browser use and their sites and
- 1:47routines and everything like that. And
- 1:49the rest of this course we're going to
- 1:50be using the Codeex desktop app. So, if
- 1:52you want it to all look the same, then
- 1:53just go ahead and download the app real
- 1:54quick. Now, once you're inside here,
- 1:56it's going to look like this. You're
- 1:57going to have two different tabs up top
- 1:58right away. You'll see chat and work,
- 2:00but then on the lefth hand side, you
- 2:01will switch over here from chatbt, which
- 2:03is chat and work to codeex. And if you
- 2:06do have a chatbt subscription already,
- 2:07like you're paying 20 bucks a month or
- 2:08100 or 200, that's the same exact
- 2:10subscription. So, you won't have to make
- 2:12a new account, but if you are currently
- 2:13on a completely free account for OpenAI
- 2:16or for Chatbt, then I would at least get
- 2:17on the $20 a month plan right now. And
- 2:19if you start to blow through your usage,
- 2:21then just go ahead and upgrade. Because
- 2:22the way that this works is you can see
- 2:23in the bottom left, you can see that I
- 2:25have 94% left and I'm currently on the
- 2:27$200 a month plan for Codex. And so if I
- 2:30use this up within the week, then I
- 2:31would just basically have to wait till
- 2:32it resets. Or I could pay per usage API
- 2:35billing, which is just a little bit more
- 2:36expensive than being on the
- 2:37subscription. The subscription of Codeex
- 2:39actually gets you about $14,000 of
- 2:41inference if you use the entire
- 2:43subscription. So, it's it's really
- 2:45really cheap for being on the
- 2:46subscription, paying $200 a month, and
- 2:48I'm getting about $14,000 of usage out
- 2:50of it. And I know that 200 bucks a month
- 2:52or 100 bucks a month may seem expensive
- 2:54relative to subscriptions, but we're not
- 2:56paying for Netflix here. We're not
- 2:57paying for entertainment. We're paying
- 2:58for productivity. We're essentially
- 2:59treating this as the cheapest hire you
- 3:01will ever make. $200 a month for the
- 3:03output that multiple hires would give
- 3:04you if you start to use it right, which
- 3:06is why you're investing time in this
- 3:08course right here. Anyways, before we
- 3:09get into this, I wanted to address one
- 3:11question, which is okay, why would I use
- 3:13codeex over something like work? Because
- 3:15work is kind of marketed as like for
- 3:17knowledge work for nontechnical people,
- 3:19blah blah blah. I think it's good that
- 3:21they're doing that, but ultimately it's
- 3:23one of those things where codecs can do
- 3:25everything that work can do and it can
- 3:27do more. So, why would you not just
- 3:28invest the time learning codecs because
- 3:30it's basically the same, but then one
- 3:32day you might need some of the extended
- 3:33functionality that Codeex has and you
- 3:35can just do it right there because
- 3:36you've already learned Codex. I'm not
- 3:38technical at all. I don't know how to
- 3:39write code, but I've been able to do
- 3:40some very, very incredible things with
- 3:42codecs and with cloud code and tools
- 3:43like that, all with my natural language.
- 3:45And this entire course, you're going to
- 3:46see it's literally just natural
- 3:48language. There's just a few things that
- 3:49you need to understand in order to get
- 3:50the most out of it. So, if you're just
- 3:51starting or if you've been using work
- 3:52for a while, just trust me, go ahead and
- 3:54switch over to Codeex. You're going to
- 3:55thank yourself one day. And trust me,
- 3:57it's not anything more difficult or
- 3:58intimidating at all compared to work.
- 4:00You're just getting more power. Awesome.
- 4:02So, now that that's out of the way,
- 4:04there's a lot of things that we're going
- 4:05to be looking at in this app. And I
- 4:06don't want to just run you through in a
- 4:08really boring way and click on every
- 4:09single thing and explain what it means.
- 4:11I'm going to basically be doing that
- 4:12throughout this entire course as we get
- 4:13familiar with everything. So the first
- 4:15thing that I actually want to do is kind
- 4:16of start with like a 30,000 foot view of
- 4:18everything that's possible in Codeex and
- 4:19how simple it all is when we really
- 4:21bring it together. So we're going to
- 4:22start with me explaining to you all of
- 4:24the core concepts that you actually need
- 4:26to know in order to use Codex and in
- 4:28order to understand what it can do. And
- 4:29I'm going to explain the stuff extremely
- 4:30simply with some examples. So let's hop
- 4:32into this next segment. Today I'm going
- 4:34over the 18 core Codeex concepts that
- 4:36you actually need to know in order to
- 4:37start using it right away and getting
- 4:39value from it. It doesn't matter if
- 4:40you're not technical at all or you've
- 4:41never used Codex before. By the end of
- 4:42this video you'll understand exactly how
- 4:44this thing actually works so that you
- 4:45can start using it right away. So let's
- 4:46not waste any time and just get straight
- 4:48into this one. All right, so I've split
- 4:49these 18 core concepts into four
- 4:50different parts and they get cooler as
- 4:52we continue to go on. So part one we're
- 4:54going to start off here is foundations.
- 4:56So we're going to kick off with concept
- 4:57number one which is projects. Now, a lot
- 5:00of you guys, I'm assuming, have used
- 5:01chat GBT or cloud in the web before
- 5:03where you can talk to an LLM and it can,
- 5:05you know, give you an answer. But the
- 5:06problem you face is that every time
- 5:08you're talking to it or you start a new
- 5:09chat, you have to kind of familiarize it
- 5:11with you and your business and what
- 5:12you're working on that week or that
- 5:14quarter. So, with projects, you can
- 5:15basically keep everything organized into
- 5:17one project so that as you keep working
- 5:19inside of it, it only gets smarter about
- 5:21knowing who you are. So, right here on
- 5:22the lefth hand side, you can see that
- 5:23I've got a few projects going on. I've
- 5:25got Herk 2, I've got Hyperframes, I've
- 5:26got Trading Challenge, I've got AIS
- 5:28demo, and I typically am always just
- 5:29working inside of my Herk 2 project. My
- 5:31Herk 2 project is basically what I call
- 5:33my AI operating system. And what a
- 5:35project is is actually very simple. It
- 5:36is just a collection of folders and
- 5:38files. So right here, this is inside of
- 5:40my file explorer on my computer and it's
- 5:41called Herk 2. And this is exactly the
- 5:43project that I have open in Codeex. When
- 5:45I say right here, I'm inside of Codex,
- 5:48right? This is the file path that I'm
- 5:49working in, which is if I open this up,
- 5:51my Herk 2 project. So what that means is
- 5:53this project has rules about me. It has,
- 5:56you know, credentials. It has all my
- 5:58projects. It has all my YouTube videos.
- 6:00This thing has so much knowledge about
- 6:01me. So now when I come into the chat and
- 6:03I say, "Hey, can you just give me a
- 6:05rundown of what we've been working on
- 6:07this past month?" This thing basically
- 6:08gets familiarized with my project. And
- 6:10then it looks at everything that we've
- 6:12been working on together. It even looks
- 6:13through all of my previous chats that
- 6:15I've had with Codeex. So here you can
- 6:17see that it looked back from the past
- 6:18month and it said, "Here's what we've
- 6:19worked on. We've been improving your
- 6:21Herk 2 and your AISOs. We've been
- 6:23working on YouTube content and strategy,
- 6:25AI model testing and demos, apps and
- 6:27product experiments, websites, branding,
- 6:29business operations. So, this is no
- 6:31longer just an AI chatbot. This is now a
- 6:34co-founder. This is a personal
- 6:35assistant. Which brings us on really
- 6:37nicely into concept number two, which is
- 6:39something called an agents.md.
- 6:41And don't let that MD intimidate you
- 6:43here. If I go over to the right hand
- 6:45side and I open up my files, you'll
- 6:47notice if I scroll down, I have
- 6:48something right here called an
- 6:49agents.mmd. And when I open this up,
- 6:51it's literally just a markdown file.
- 6:53That's what the MD means. And it's just
- 6:55rules. Like this is a very simple thing
- 6:57that you all can read and it's not
- 6:59technical at all. So this is basically
- 7:00like the rules for your project. So you
- 7:03remember inside of my Herk 2 project, I
- 7:04have this agent on MD. Well, if I was
- 7:06inside of my hyperframes editor project,
- 7:08there would be a different agents.md. So
- 7:10this thing basically just sets the
- 7:11ground rules for the project that you're
- 7:12working in. So right here you can see
- 7:13that this one says, "Hey, you are Nate
- 7:15Herk's AI operating system. Your job is
- 7:16to help him spend less time on
- 7:18operations so he can focus on making
- 7:19YouTube videos." And then I give it
- 7:21things like core operating rules for how
- 7:22to work with me. I give it some stuff
- 7:24about how to be safe on my desktop. And
- 7:26then the real bulk of my agent MD is
- 7:28what I call routing. So this is a
- 7:29routing map. It basically means Okay, so
- 7:31yeah, Nate's project is huge. There are
- 7:33hundreds, probably thousands of files
- 7:35and folders inside of this project. So
- 7:37let me just explain to you how this
- 7:39actually works. If you need business
- 7:40stuff, you go to the wiki. If you need
- 7:42corporate structure, you go to the
- 7:44corporate structure section inside the
- 7:45wiki. If you need his voice, you go
- 7:46here. If you need course knowledge, you
- 7:48go here. If you want active projects,
- 7:49you go here. So everything that I'm
- 7:50doing, we're routing it back somewhere
- 7:53so that my agents.mmd file tells this
- 7:56project where everything is so that
- 7:57every time I open up a new chat, like I
- 7:59said, I don't have to reexlain things.
- 8:00And basically the way that this works,
- 8:01if we picture this being a chat thread,
- 8:03you know, you come over here and you
- 8:04say, "Hey, you say, "Hey, Codex, I need
- 8:06you to help me do this." And before
- 8:08Codex actually even like reads this
- 8:10message, what it does is it first reads
- 8:13the agents.mmd. So I'm just going to put
- 8:15a here and that stands for agents.mmd.
- 8:16But codeex will basically read this and
- 8:18then it will read your message and then
- 8:20it will respond with something like, you
- 8:21know, hey, how can I help you today? So
- 8:23the kind of stuff that you want to put
- 8:24inside your agents.mmd is the type of
- 8:26stuff that you want codeex to know about
- 8:28you every single time you're talking to
- 8:30it in that project. And now moving on to
- 8:32concept number three, we have something
- 8:34called the agent loop. So you'll hear
- 8:36about something like codecs or claude
- 8:38code or Hermes agent. These are all
- 8:40called agent harnesses. And these
- 8:42harnesses have their own agent loops.
- 8:44And real quick, if you're wondering
- 8:46about codecs versus like work chatbt
- 8:48work, I always use codecs. The majority
- 8:50of my work is not building products or
- 8:52building software, but I only use codecs
- 8:54because it's just the most powerful. And
- 8:56once you get through this video, you'll
- 8:57realize how easy it is to use. But
- 8:58anyways, the reason why this is called a
- 9:00harness is because they have this thing
- 9:02called the agentic loop, which at the
- 9:03highest level basically just means when
- 9:05you ask the agent a question, it has a
- 9:08bunch of tools at its disposal and it
- 9:10basically thinks and it reasons and it
- 9:12uses tools and then it responds to you
- 9:13and then it thinks and then it reasons
- 9:15and it takes action. It uses tools,
- 9:16responds to you and that's basically the
- 9:18agent loop. And what's really awesome
- 9:19about that is you can see it in action.
- 9:21So right here you can see that this said
- 9:23worked for 25 seconds. And if I open
- 9:24this up, we can basically see what it
- 9:27did, right? Right? So like it hid this
- 9:28response and it gave us our output, but
- 9:30it had a reasoning loop inside. It first
- 9:32said, "Okay, I'll check our recent tasks
- 9:34and project nodes, then I'll pull
- 9:35together the main themes from the past
- 9:36month." And when it did that, it ran all
- 9:39these commands. It ran this, which you
- 9:41don't have to know what this means, but
- 9:42basically what this did is it looked
- 9:43inside something right here called a
- 9:44memory.md. So very similar to the
- 9:46agents.md, but this is about memories
- 9:49rather than just like rules. It then
- 9:51looked at other chats. And so we can
- 9:53basically watch it do things. So, to
- 9:54show you guys a real example of this,
- 9:56hey, could you just check my most recent
- 9:58YouTube video that I uploaded and give
- 10:00me three comments that you thought were
- 10:02funny from that video.
- 10:04Now, as I shoot this off, what I want
- 10:06you guys to pay attention to is what
- 10:07we're watching. So, first it said
- 10:09thinking. Now, it says, "I'll find your
- 10:10latest upload, read through the
- 10:11comments, and pick three that made me
- 10:12laugh." And now we can see what it's
- 10:14doing. So, it's looking through the
- 10:15memory. It's looking at these things.
- 10:17It's running these commands. And these
- 10:19commands are basically right here. It's
- 10:21reading, you know, files to get
- 10:23established and to get oriented with
- 10:24what it needs to do. And so, you can
- 10:26actually learn a lot about how codeex
- 10:28works and how to best work with it by
- 10:30when you ask it a question, you just
- 10:32kind of watch what it does. So, you can
- 10:33see what it had to do was it had to run
- 10:35a command in order to actually go talk
- 10:37to YouTube, look at the data, and pull
- 10:40it back. And now we can see three
- 10:42comments. We have this one, never seen
- 10:43anyone so excited to lose money. Trading
- 10:45with AI is very easy. I just prompt
- 10:47Claude only take profitable trades. And
- 10:49if I want to lose money, I can do it
- 10:50myself faster and better. And what
- 10:52you'll notice is how quick that all
- 10:53happened. Now, I do have my Astra on
- 10:55fast mode, but I will talk about that
- 10:56later because that is a core concept
- 10:58that's coming up. By the way, guys, I've
- 10:59got this completely free SOP for you
- 11:01about getting your first Am. It's going
- 11:03to go over the exact steps that has been
- 11:05proven for hundreds of our AIS Plus
- 11:07members to get their first paid gigs. It
- 11:09goes over the one sentence service pitch
- 11:11that can get you started today. Why your
- 11:12first client should cost you money. The
- 11:14five-minute video that answers, can this
- 11:15person actually deliver before you've
- 11:17actually received any money, what to do
- 11:19when you have zero case studies. There's
- 11:20so many good things in here that are
- 11:22going to help you out. Even if you
- 11:23already do have clients, I would
- 11:25recommend grabbing this because like I
- 11:26said, it's yours completely free. So, if
- 11:28you want to grab this, there's a link
- 11:29for it down in the description. Let's
- 11:30get back to the video. All right, and
- 11:32moving on to the last concept of part
- 11:34one. It's a slash goal. And this is
- 11:37honestly one of my favorite things about
- 11:39AI ever. It's the ability to do
- 11:41something called a slashgoal. And right
- 11:43here you can see that that says set a
- 11:45goal to keep pursuing. Now AI agents are
- 11:47super goal oriented which is super cool.
- 11:49So let's say you have Mr. AI agent right
- 11:51here. And you basically say hey here is
- 11:53your goal. You know this is what you
- 11:54need to do and you give it this
- 11:56objective goal. What happens is it will
- 11:58basically just keep working and keep
- 12:00working and keep working until it
- 12:02actually hits that goal. And it's cool
- 12:03because even if you kind of like give
- 12:05them a roadblock, they are going to be
- 12:08so goal-oriented that they're going to
- 12:09keep retrying different things and
- 12:11they're going to just basically keep
- 12:13going until they're able to succeed and
- 12:15hit that goal for you. Now, obviously
- 12:16the more objective a goal, the easier it
- 12:18is to prove that. So, if you say, "Hey,
- 12:20can you, you know, work on this until
- 12:22you've pulled in 257 sources and then
- 12:25write the report, that's an objective
- 12:27goal, right? 257 sources and written
- 12:29report." But you can also set goals that
- 12:31are a little bit more emotional. Like
- 12:32your goal could maybe be more like, you
- 12:34know, until you're fully satisfied or
- 12:36until you've, you know, verified this
- 12:38over and over and you feel good about
- 12:40it. And in that case, you're still going
- 12:42to, you know, get there and it will
- 12:44still decide when it's done. But as you
- 12:46get more objective with your goal, the
- 12:48goal prompts are just more effective.
- 12:50But let me show you one goal that I
- 12:52actually did that you guys have already
- 12:53been witnessing in this specific example
- 12:55is I went to my Hyperframes project. And
- 12:57hyperframes basically lets your coding
- 12:58agents build HTML and then render it as
- 13:01video. So essentially it creates video.
- 13:03But what I did here is I shot it off
- 13:04a/goal prompt. You can see this one says
- 13:06sent as goal. So basically what I did is
- 13:08I said I need you to help me create some
- 13:09motion graphics for a YouTube video. So
- 13:11all the motion graphics you've seen so
- 13:12far were from this exact prompt right
- 13:15here. I'm going over 18 different Codex
- 13:16concepts. I want, you know, an intro. I
- 13:18want transition cards. I want all this.
- 13:20So I basically defined that I wanted one
- 13:21intro scene and that I wanted 18 cards.
- 13:24And then you can see that it said I
- 13:25finished and I visually checked. meaning
- 13:26it actually took screenshots to make
- 13:28sure everything looked good before it
- 13:29decided to be done. And now we have all
- 13:31these animations. But what's cool about
- 13:33the goal prompts is that models nowadays
- 13:35are getting so good, the AI models are
- 13:37getting so good that it's actually
- 13:39better to give them a goal and then step
- 13:41out of their way. If you really just try
- 13:42to like micromanage them, you're kind of
- 13:44just like muzzling their capabilities.
- 13:46So if the situation isn't super super
- 13:48risky, give them a goal and you can get
- 13:49out of the way. Now, there's one quick
- 13:51thing that I felt the need to address,
- 13:52which is basically the question, "But
- 13:54Nate, what if we've been using cloud
- 13:55code for a long time?" Well, I'm really
- 13:57glad that you asked. It's actually
- 13:58really simple. Really, the main
- 14:00difference is the agents.mmd, thecodex,
- 14:02and the aagents. Because in cloud code,
- 14:05you basically have like your cloud. MD,
- 14:06and you have your cloud. So, when I
- 14:08first started switching over from claude
- 14:09code to codex, all I did was I said,
- 14:11"Hey, codeex, take a look at this herku
- 14:13project. I've been building this up
- 14:15using cloud code for the past couple
- 14:16months. I need you to help me get this
- 14:18codeex ready." And basically what that
- 14:20means is I made a copy of my cloud.mmd
- 14:22and called it agents.mmd. So I have two
- 14:24of those now. One one aent.mmd one
- 14:25claude but they're basically the same
- 14:27file. And then I made a copy of all my
- 14:28cloud skills and put them in the agents
- 14:30folder and I made a copy of all of my
- 14:32like claude settings files and I put
- 14:34them in my agents. And the cool thing is
- 14:36you don't have to do any of that
- 14:37manually. You literally just say hey
- 14:38codeex look at the documentation analyze
- 14:40my project and make this codeex
- 14:42accessible. Make it codeex ready. So
- 14:44anyways, if you were feeling a little
- 14:46bit of doubt about all that kind of
- 14:47stuff, that's the way I did it. And now
- 14:49you have a bunch of local files and
- 14:50folders that you can use on any agent
- 14:52harness whenever. It's very flexible.
- 14:54It's very tool agnostic, and that's what
- 14:55you want to be building at the end of
- 14:56the day. So anyways, had to address that
- 14:58real quick. Let's get back to the video.
- 15:00All right, awesome. So let's move on to
- 15:01part two of these core concepts, which
- 15:03is environments. So the first piece of
- 15:05environments is number five, which is
- 15:08local versus cloud. So there's a couple
- 15:10things to talk about here. The first
- 15:11thing is if someone says, "Oh, is this
- 15:13running locally or are these local
- 15:15files?" All that means is does that
- 15:17exist only on the machine that you're
- 15:18currently using? Whether you're on a
- 15:20MacBook right now or you're on a desktop
- 15:21PC like I am right now. When you say
- 15:23local, it just means it's only
- 15:24accessible by you. So right here, if I
- 15:26open up my files, these are basically
- 15:28local files. Now, they're syncing with
- 15:30one drive when I want them to. So that
- 15:31would kind of push them into the cloud,
- 15:33but otherwise it's completely local. So
- 15:35like if you're working on a word doc on
- 15:36your computer and you forgot to save it
- 15:39and then email it to yourself on so you
- 15:41could work on it on a different laptop
- 15:42that was a problem because it was
- 15:44locally on that other machine. But
- 15:45something like Google Drive is always in
- 15:47the cloud obviously. But this also goes
- 15:48beyond just files. This goes to like
- 15:51things that are produced. So right here
- 15:52you can see that it actually created me
- 15:54this little site for all these motion
- 15:56graphics. So if I open up this site you
- 15:57can see that I can view all of these
- 15:59videos. I can click into the different
- 16:00ones and it threw this together for me
- 16:02really nice and it served this to me on
- 16:04a local host. So right here you can see
- 16:06that this URL is 127.0.0.1
- 16:09and the port of this local host is 8008.
- 16:12Now realistically that's not super
- 16:14important. You don't really need to
- 16:15understand what exactly this means. But
- 16:17this is just a local host. But this is a
- 16:20little confusing because it looks like
- 16:21this is a web address that if I, you
- 16:22know, emailed to you, you could open
- 16:24this up and have the site. But what
- 16:25would happen is this would show on your
- 16:27screen or it would show on your computer
- 16:29as nothing actually lives here. So like
- 16:31if I copied this and I pasted this into
- 16:32a new browser but I changed the number
- 16:34to 7. I don't think I have anything
- 16:37running on port 8087
- 16:40on my local machine even though you
- 16:41might. So nothing actually loads up. And
- 16:43it's quite funny. I remember seeing
- 16:44these these tweets where I think it was
- 16:46a joke but it was like uh someone was
- 16:48pretending they're a beginner and they
- 16:50were like wow cloud code and codeex it's
- 16:52so cool. Look what I built in one day.
- 16:53and then they, you know, attached a
- 16:55bunch of local host addresses and
- 16:57obviously no one else could open that up
- 16:59because they're not on that local
- 17:00device. So anyways, that is the
- 17:01difference between local and cloud. Now
- 17:03the other piece here is that when you're
- 17:04spinning up these new chats, you can
- 17:06kind of have these be either local,
- 17:08meaning right here, you know, I'm
- 17:09working in my local files and folders.
- 17:11And that's great because codeex can
- 17:13actually like edit these files, right?
- 17:15It can delete things, it can make new
- 17:16ones, it can edit them, it can move them
- 17:18around, it can go to my downloads, it
- 17:20can go to my desktop, it can do anything
- 17:21on my computer really. or you could
- 17:23actually have these work in the cloud.
- 17:24And what you'll notice here is when you
- 17:25choose cloud, you no longer have the
- 17:27ability down here to actually choose the
- 17:29model you were using and the permissions
- 17:31because this sets up sort of like a
- 17:33cloud sandbox environment. Now, I'll be
- 17:35honest, I hardly ever do this because
- 17:37when I'm working locally, as long as I
- 17:39keep my machine on, as long as I keep my
- 17:41PC on, it feels pretty cloud-like. And
- 17:44here's what I mean by that. Right here,
- 17:45you can see that I'm mirroring my phone
- 17:46and I'm in the Chad GBT app. And if I go
- 17:48over here and I click on this right here
- 17:50which says remote, this lets me see all
- 17:53of my actual codeex chats that are right
- 17:56here. So if I go to Herk 2 and I go to
- 17:58summarize past month, this is exactly
- 18:00what we were just looking at right here.
- 18:02You can see this is literally the same
- 18:03conversation. It has those three chats
- 18:05and I could come in here and I could say
- 18:06hi and this gets sent in my actual
- 18:08codeex right here on my laptop or sorry
- 18:10on my desktop as well as my phone. So,
- 18:13as long as my machine's on, even though
- 18:14this is running locally and I'm out on a
- 18:16walk, I'm out at dinner. I could even be
- 18:18in a different state and I could control
- 18:20my desktop from my phone right here. So,
- 18:23let's say for some reason I left a file
- 18:24locally on my computer, I could just
- 18:26say, "Hey, can you email that file to
- 18:27me?" And then I could use it on my
- 18:29laptop. So, anyways, just wanted to show
- 18:30off that remote capability. And that's
- 18:32going to bring us into concept number
- 18:34six, which is work trees. Now, honestly,
- 18:36this is something where if you're not
- 18:37really building software or, you know,
- 18:39production apps, this isn't going to
- 18:41matter to you too much. Like I don't
- 18:43hardly ever use work trees, but it is
- 18:45good to know about since you will see
- 18:46these in the interface. So right here
- 18:48you can see that if I was to go to um
- 18:51the local setting, it says that you can
- 18:52work in a new local work tree which says
- 18:55create a copy of Herk 2 to work in
- 18:57parallel. So basically because I'm
- 18:59working in Herk 2, if I was going to do
- 19:00something like risky or if I wanted to
- 19:02try and experiment with something or for
- 19:04example, if I was building software and
- 19:05I wanted to try to change the way that
- 19:07the onboarding flow worked, but I wanted
- 19:08to like not interrupt the main branch,
- 19:11especially if other people on my team
- 19:12are working on that main branch, I could
- 19:14use a work tree to basically create a
- 19:15copy and I could test in there. And then
- 19:18what you can do later is you can merge
- 19:19it back into the main. So it's basically
- 19:21just a duplicate that lets you test. But
- 19:23like I said, there's hardly any times
- 19:24for knowledge work that I ever use work
- 19:26trees. And if I ever am building
- 19:27software and stuff, a lot of times Codex
- 19:29will say, "Hey, by the way, you might
- 19:30want to use this. You might want to do
- 19:31this in a new work tree." So, I wanted
- 19:32to call it out because it's important
- 19:34because you might see things like this
- 19:35master branch and you might see other
- 19:36work trees that might have been spun up
- 19:38automatically. But I will say it's not
- 19:39something that I'm actively thinking
- 19:40about or think that you need to be
- 19:42actively thinking about in order to get
- 19:43the most out of codeex. So, wanted to
- 19:45bring that up. But let's move on to
- 19:46number seven, which is codeex. Now, this
- 19:50is really interesting. We talked about
- 19:51how when you're inside of your project,
- 19:52you have things like an agent.mmd. Now,
- 19:55if I go to my files, there's also
- 19:56something in here called a codeex. So,
- 19:59if I open this up, what you'll see in
- 20:00here is I have some agents and I have
- 20:02some workflows, but I also have a
- 20:04config. ML file, which is basically like
- 20:07some local settings. And there are
- 20:09basically going to be two different
- 20:10places where a codeex file lives. And
- 20:13that's either going to be inside of your
- 20:14project. So like in this example, this
- 20:16is my codeex inside of my Herk 2. But
- 20:19there's also going to be one which is
- 20:21user or global, which means this lives
- 20:23under your actual user. And these are
- 20:25settings that hold things like memories
- 20:27and sessions and automations. These are
- 20:30things that will apply to your global
- 20:32codec. So whenever you're in Codex,
- 20:33regardless of what your project you're
- 20:35actually in, the docex will have
- 20:37personal settings and app data and stuff
- 20:38like that. Now once again, I think this
- 20:40is something it's important to
- 20:41understand where that lives and what it
- 20:43does. But this isn't a folder that I'm
- 20:44actively thinking about and actively
- 20:46having to like maintain. But I did want
- 20:48to call that out because sometimes if
- 20:50you need to have a certain setting
- 20:51applied or something like that, that is
- 20:53where it lives. Now, the cool thing
- 20:54about this is if you're ever confused,
- 20:56you say, "Hey Codeex, what did you do
- 20:58there?" or "Do we need to add this to my
- 21:00codeex or do we need to move this to a
- 21:01different folder?" You can just ask it
- 21:03questions because it can search through
- 21:04its own documentation and understand how
- 21:06to do things like correctly. But I do
- 21:08think that understanding the difference
- 21:09between project scope things and user
- 21:11scope things that's definitely important
- 21:13especially as we get into the agents
- 21:15which is going to come up later. And
- 21:16that is a great segue into part three of
- 21:18the core concepts which is control and
- 21:20customization. So number eight here is
- 21:23about different AI models. So when you
- 21:26are starting a new chat you can choose
- 21:28the model you want to use. You can also
- 21:30change the model in between messages.
- 21:32So, I could shoot off a message on Astra
- 21:33and then I could, you know, change this
- 21:35to something else and then I could shoot
- 21:36off another message and I can keep
- 21:37changing if I want. Now, if you're
- 21:39watching this video way later, maybe the
- 21:40model names look different, but
- 21:41essentially the theory is the same. You
- 21:43have different models to choose from and
- 21:45they all have different prices and
- 21:47strengths. So, like right now, GBD6
- 21:49Astra is just the absolute strongest,
- 21:50but it's going to cost the most. It's
- 21:52going to eat the most of your weekly
- 21:53limit. Right here, you can see that I'm
- 21:54at 100% because it just got reset. But
- 21:57as you chat with Astro, it will drain
- 21:59your weekly usage faster than you know
- 22:015.6 Soul would, which would drain your
- 22:03usage limit faster than 5.6 Tarot would.
- 22:06Now, right here, I just want you to look
- 22:07at the API pricing so I can explain to
- 22:09you how these models are actually
- 22:11billing you. So, they build by tokens,
- 22:13and they usually build by 1 million
- 22:15tokens. Meaning, on Astra, for every 1
- 22:17million input tokens, it will cost you
- 22:20$10. And for every 1 million output
- 22:22tokens, it will cost you 50 bucks. And
- 22:23here you can see the different pricing
- 22:25for these different models. And you can
- 22:26see how it gets cheaper as you go down.
- 22:28Now, this is API pricing, which means if
- 22:30you were building an automation that was
- 22:32programmatically using these models,
- 22:34that's how much it would cost you. But
- 22:36this usage inside of Codex, that's not
- 22:37programmatic. This is usage limit or
- 22:40subscription limit. So, when you pay 20
- 22:42or 100 or 200 bucks a month for Codeex,
- 22:44this is what you're actually your usage
- 22:45limit is eating up is that subscription.
- 22:48And you're getting way more inference.
- 22:49Like, this is a $200 a month
- 22:50subscription. And if I use all of my
- 22:53weekly usage every single week, I'm
- 22:56getting about $14,000 worth of inference
- 22:59every month and I'm only paying 200
- 23:00bucks for that. So the subscriptions are
- 23:02much cheaper than if you were using the
- 23:03API pricing. But if you ever do go over
- 23:05your weekly limit, you'd have to buy
- 23:07extra codeex credits. I don't exactly
- 23:09know why they price it like this, but
- 23:10you know, you can see they just give you
- 23:12codeex credits instead. So let's say I
- 23:14went over my weekly limit. I could go
- 23:16here and I could click on buy credits.
- 23:18This would bring up this screen which
- 23:19lets you buy credits by like the
- 23:21thousands or something like that. And
- 23:23that is what I wanted to show you guys.
- 23:24So like 2500 credits, 5,000 credits,
- 23:2725,000 credits. And this is how it would
- 23:29charge you based on those tokens. Now,
- 23:31if you've never heard of tokens before,
- 23:32I know that might be a little confusing.
- 23:34Think about it as roughly four
- 23:35characters or roughly 3/4s of a word.
- 23:38And there's an OpenAI tokenizer right
- 23:40here where if I paste in some text, it
- 23:42will show you how many tokens this would
- 23:43be. So this paragraph, for example, was
- 23:46about 240 characters or sorry 274
- 23:48characters and about 56 tokens. And it's
- 23:50interesting because sometimes
- 23:51punctuation is a token, right? Like this
- 23:53period is a token, this comma is a
- 23:55token, this word's a token, um this
- 23:57word's a token. And this is basically
- 23:59just kind of breaking down what it might
- 24:00look like for these tokens. But anyways,
- 24:02you can see this was 56. And what else
- 24:04you might notice is that I earlier said
- 24:06input tokens or output tokens. Input
- 24:08tokens are whatever you're feeding in.
- 24:10So your prompts or if it's reading like
- 24:11a PDF or a word doc those are input
- 24:14because they're going into the model and
- 24:15output tokens are anything that it
- 24:17outputs. So even its reasoning here is
- 24:19output tokens. This entire thread is
- 24:21output tokens and that's why the output
- 24:23tokens are more expensive. Okay. So
- 24:25let's move on to number nine which is
- 24:26about effort. So within each model not
- 24:29only do you choose the model but you can
- 24:30also choose the effort level. So right
- 24:32here I can move Astro down to medium or
- 24:34I can move it down to low or I can move
- 24:36it up to extra high or if I go like
- 24:38ultra I have to allow full access
- 24:40because that can get you know pretty
- 24:42autonomous. But anyways the point being
- 24:44as you move up on these effort levels
- 24:46you get more quality like you get more
- 24:49intelligence but you also get more cost.
- 24:52it's more expensive to run it at higher
- 24:53effort levels. And that's why when you
- 24:55see these benchmarks, when there are new
- 24:56models released, you'll see the model,
- 24:58like for example, Cloud Fable 5, you can
- 25:00see this mark was on low and this one
- 25:03was on medium and this one was on high.
- 25:05So, it shows how one model can behave
- 25:07differently and also cost a different
- 25:08amount based on the effort level that it
- 25:10was being used and tested at. Now, it
- 25:12actually is pretty nice to play with
- 25:14these because sometimes Astra High is
- 25:16just so much more powerful than what you
- 25:17need to do. Like if you were using GBD6
- 25:19Astra High to help you write an email,
- 25:21that's overkill and you're paying way
- 25:23more than you should realistically to
- 25:24write an email. I could probably come
- 25:25down here and use 56 Terra and be just
- 25:27fine with that and it's a lot cheaper.
- 25:29So definitely try to play with the model
- 25:30and play with the effort based on your
- 25:32task. But a lot of times if you're just
- 25:33doing a lot of knowledge work, I would
- 25:34just chuck it on sole and I'd maybe just
- 25:36go to medium and just call it a day
- 25:38there. Now the other thing about effort
- 25:39is you can do something called fast
- 25:40mode. This obviously makes the model
- 25:42respond quicker and do things quicker.
- 25:43And it it's 1.5 speed, but it costs you
- 25:45more usage, so it will eat your weekly
- 25:47subscription faster as well. All right,
- 25:50and moving on to number 10, we have
- 25:51permissions. Basically, right down here,
- 25:53you can see that this is orange and it
- 25:54says full access. This is basically
- 25:56Codeex's yolo mode, which means that
- 25:58it's not going to stop and ask you a
- 26:00bunch of permissions and questions. Can
- 26:01I do this? Can I do this? If you go on
- 26:03ask for approval, it's going to always
- 26:05ask to edit external files and always
- 26:07ask to use the internet. So that's if
- 26:08you want to sit here and really sort of
- 26:10like manage it and you're wanting to
- 26:11make sure everything's safe, you know,
- 26:13maybe start off like this so you can
- 26:14just understand, oh, okay, it's asking
- 26:15me if it can run this command or it's
- 26:17asking me if it can open up my browser
- 26:18and then you just get familiar with what
- 26:20it looks like for Codex to actually run
- 26:21that agentic loop. You could also go on
- 26:23approveof for me. So it's only going to
- 26:25ask you if actions are detected as
- 26:27unsafe, like maybe deletes or maybe
- 26:29certain sorts of like API calls or
- 26:31things like that. So that's how these
- 26:32permissions work. And if you really want
- 26:33to get granular about, oh, I want to
- 26:35like make sure these certain actions
- 26:36never happen, you can ask Codex about
- 26:38that and it can work things into those
- 26:40config files like we talked about
- 26:41earlier with the codeex that it can
- 26:43certainly like block things out. But now
- 26:44let's move on to number 11, which I
- 26:46think is my favorite one, which is
- 26:48skills. Skills are so so important.
- 26:50They're basically reusable workflows.
- 26:52They're recipes that let you do
- 26:54something once and then teach Codex how
- 26:56to do it the same way every single time.
- 26:58So now you can just run all these
- 26:59skills. So like I said, a skill is
- 27:01basically just a recipe. So, let's say
- 27:03you come in here and you make, you know,
- 27:05a chocolate chip pancake and you
- 27:06followed a recipe to make that pancake.
- 27:08You would basically say, "Okay, if I
- 27:10ever want to make chocolate chip
- 27:10pancakes again, I'm going to use this
- 27:12recipe because I know that I'm going to
- 27:13get a good result." But let's say that
- 27:15you burnt these pancakes and you said,
- 27:17"Okay, well, what I need to do is make a
- 27:18quick update in this recipe to say, hey,
- 27:20you know, cook it up 30 seconds less on
- 27:22each side." And the next time you run
- 27:23the skill, you basically see, okay, are
- 27:25these better? Are these good? Do I need
- 27:26to make any more feedback in here? Or
- 27:28now, is this skill just the way I like
- 27:30it? And then your agents are able to use
- 27:32these skills. So here's an example where
- 27:34I basically gave it a YouTube video. So
- 27:35you can see that this is literally just
- 27:36a local file path of a YouTube video.
- 27:38And I said that I needed a video
- 27:40description, timestamps, and a LinkedIn
- 27:41post. It went ahead and it ran its
- 27:43agentic loop and it gave me a
- 27:45description, timestamps, and a LinkedIn
- 27:46post. And then I said, "Okay, now I need
- 27:48you to turn this into an X article." And
- 27:49put the draft into my X as a draft. And
- 27:51it knew exactly how to do this with a
- 27:53thumbnail, with a title. It also put
- 27:55screenshots from my actual video and it
- 27:56sprinkled them throughout my whole X
- 27:59article. And this is because it used a
- 28:00bunch of skills to be able to know how
- 28:02to do this. So right here, if I ask it,
- 28:03what skills did you use? It used my
- 28:05YouTube description skill, my timestamp
- 28:07skill, my LinkedIn post skill, my X
- 28:08article skill, my X article from video
- 28:10skill, my format X article skill. It
- 28:12used all of these skills because I've
- 28:13taught it how to do something good once.
- 28:15And then every time I run these skills,
- 28:17I just give feedback. Hey, that was
- 28:18good. You know, I liked this, but I
- 28:20didn't like this. Update the skill. And
- 28:21now what's cool is you can just invoke
- 28:23these either with natural language. It
- 28:24will be smart enough to do it with
- 28:25natural language, or you can do it as a
- 28:27slash command. So, I have one in here
- 28:28that I can call like the audit. And you
- 28:30can see this will run my audit or I have
- 28:32one in here called the 3D brain. And
- 28:34this runs the 3D brain. Man, I think I
- 28:36planned this out so good. This is going
- 28:37to bring us on to number 12, which is
- 28:40agents. So, we talked about codeex. And
- 28:42agents is basically the exact same. It
- 28:45just holds different things. So, right
- 28:46here, you can see that the agents holds
- 28:48your skills. And these once again can
- 28:50either be project level skills or they
- 28:51can be user scope skills. So, they can
- 28:53be global skills. So, if I come into
- 28:55this chat right here, which is inside of
- 28:57my um you know, Herk 2 project, and I go
- 29:00to open up the files, and I go to my
- 29:03agents folder, which is right here, we
- 29:05have a folder called skills, and this is
- 29:07a bunch of skills, as you can see. Now,
- 29:09if I open up an actual skill, so for
- 29:11example, let me go to my YouTube
- 29:13description skill. This is an actual
- 29:15skill.md. So, once again, MD just means
- 29:17markdown file. So, what we have here is
- 29:19some metadata, and this is what Codeex
- 29:20will read to understand, okay, do I need
- 29:22to invoke this skill or not? So, it has
- 29:23the name and it has basically when to
- 29:25use the skill. And then all the skill is
- 29:27is just instructions. It's, hey, for
- 29:29Nate's YouTube descriptions, keep them
- 29:31short, two to four sentences, paragraph
- 29:33form, no bullet points, no m dashes.
- 29:35Here's the process. You read the video
- 29:36outline. You identify the core topic.
- 29:38You write a short paragraph. And that's
- 29:40the skill. It's a very, very simple
- 29:41prompt. But if I go to my ex article
- 29:43one, they can also be way more in-depth.
- 29:44So, this one is way longer. There's a
- 29:46lot of other rules and a lot of other
- 29:48criteria. There's way more steps.
- 29:49There's way more things to do. And also
- 29:51what's cool is skills can reference
- 29:53other things. Skills can reference other
- 29:55skills. They can reference other agents.
- 29:57They can reference Python files or
- 29:59context files. So in this example, this
- 30:01skill references my voice guidelines
- 30:02like my LinkedIn style guide as well as
- 30:04my ex article style guide. But that is
- 30:06where your skills live inside of the
- 30:08agents folder. And then moving on to
- 30:10number 13. Here we have plugins. And
- 30:12this is where a lot of the magic
- 30:13actually happens because you can super
- 30:15easily come over here in the Codex
- 30:17desktop app. You can go to plugins and
- 30:19you can connect a bunch of things. You
- 30:20can see all the stuff that I have
- 30:21installed like Alpaca, Gmail, ClickUp,
- 30:23Clay, Google Drive, GitHub, Canva, and
- 30:26the list goes on and on. You can also
- 30:28search for different plugins. So, let's
- 30:30say you were really interested in
- 30:30connecting something like, I don't know,
- 30:32let's see, do they have a LinkedIn
- 30:34plugin? They have a LinkedIn plugin
- 30:35where we can find the right profession,
- 30:36we can grow our business with ads. We
- 30:38have all these other sorts of plugins,
- 30:39too. So, if you want to try to connect
- 30:41your different apps, just come in here
- 30:43and see if you can connect them using
- 30:44these plugins. Maybe they have
- 30:46Composeio, they don't have Composeio,
- 30:47and if they don't have a plugin here,
- 30:49you can still connect it using Aenv
- 30:51using an API key. But plugins are so
- 30:53much easier because you can just
- 30:54basically sign in once. They've got ones
- 30:56for creativity, for developer tools, for
- 30:58business and operations, for data
- 30:59analytics, communication, and they're
- 31:01basically always growing this plug-in
- 31:03library. All right, so that now brings
- 31:05us on to part four, which is tools and
- 31:08scale. Okay, we've got five left here.
- 31:10Number 14, we have the browser. Now,
- 31:12this is definitely one of my favorite
- 31:14things about Codeex. If I open up the
- 31:16tab over here, we can open up a browser,
- 31:18which is really cool because I can
- 31:19basically just control this. So, like I
- 31:21could go to school.com. And what you'll
- 31:23notice is when I log into school.com
- 31:25over here, I'm just going to zoom out a
- 31:26little bit. You can see right here that
- 31:27I'm already logged into my school
- 31:29account. So, I could manage my different
- 31:30communities, which also means that
- 31:31Codeex can help me manage my communities
- 31:34from here. Because when you save these
- 31:35login inside of the Codeex browser, it
- 31:37then saves them. So, if you say, "Hey,
- 31:38could you go to that account and pull
- 31:40the report for me?" or hey could you go
- 31:42into school and could you make this post
- 31:43for me it can do it through the browser
- 31:45if there's not a plugin or if there's
- 31:47not like an API or MCP server and the
- 31:49browser use of codeex and specifically
- 31:51with GBT6 Astra is the best that I've
- 31:53ever used ever I'll play a clip right
- 31:56here where I asked it to basically open
- 31:57up Canva and paint me and draw me using
- 32:00the tools in there and I gave it just a
- 32:02picture of me and it was able to
- 32:03replicate it in a way that I thought was
- 32:04really really good. So, silly example,
- 32:06but it definitely demonstrates the
- 32:07vision and the browser use capability of
- 32:10this agent. All right. So, number 15, we
- 32:12have sites. So, if you go over here to
- 32:14explore and you click on sites, this is
- 32:15very, very cool because it basically
- 32:17lets you put things out there onto the
- 32:20cloud. Remember earlier we talked about
- 32:22local versus cloud. Codex can build
- 32:23something for you and it can be a local
- 32:25host, but then you say, "Hey, can you
- 32:26just put that on a site real quick so
- 32:27that you could have your team log in or
- 32:28so that you could share it with, you
- 32:30know, other people and make it publicly
- 32:31accessible." Like right here, you can
- 32:32see this London and Paris one is
- 32:34everyone whoever has this URL could go
- 32:36ahead and open this up. Whereas this one
- 32:38is just me. So it would literally have
- 32:40to be logged into my account in order to
- 32:42be able to open up this site. And what
- 32:44else is amazing about that is you can
- 32:45open up the analytics for this site. So
- 32:47you can basically host things. You can
- 32:49connect your own domain and this is
- 32:51basically going to replace something
- 32:52like a Versel or like a I don't know a
- 32:55Squarespace, wherever you wanted to host
- 32:56your sites normally. You can now just do
- 32:57it right here from the Codex desktop app
- 32:59which is pretty cool. You can also even
- 33:00connect a database. So on the back end
- 33:02if you had to store like user login data
- 33:04or you know permissions or settings or
- 33:07you know a database of customer records
- 33:09whatever it is you could actually store
- 33:11that here too and it's really cool. All
- 33:12you need to do to actually use a site is
- 33:14you just basically have to do either say
- 33:16hey can you turn that into a site or you
- 33:17could do slash site and then it will
- 33:19basically know that you can you know
- 33:20host a site or build a site. So great
- 33:22touch here from codeex. All right so
- 33:24number 16 we have sub aents. So, sub
- 33:27agents are really cool because it lets
- 33:29you delegate work out to a bunch of
- 33:31different Codex agents. So, right here
- 33:32in this main chat, you can see that I'm
- 33:34talking to GBT6 Astra, which is a
- 33:36really, really smart and intelligent,
- 33:38but also expensive model. So, let's say
- 33:40I wanted to do a bunch of research, but
- 33:42I didn't want to waste Astra's brain on
- 33:43that. Or let's say I wanted to do like a
- 33:45bunch of different testing and just do a
- 33:46lot of stuff at once. Astra here that
- 33:48I'm talking to can delegate workout to
- 33:50tons of sub aents. Hey, I need you to
- 33:52search through X and search through
- 33:54YouTube and also dig through my own
- 33:55comments. So, just find, you know, what
- 33:56are people talking about right now in
- 33:58the AI space. Is there any big news or
- 34:00drama? I don't want you to do this
- 34:01research though. I want you to basically
- 34:03just delegate a bunch of sub agents out
- 34:04to do the work and I want all of these
- 34:06sub agents to be using 5.6 Terra as the
- 34:09model. So, shoot off those agents and
- 34:11let me know what they find. So, this is
- 34:12really cool because I can specify that I
- 34:14want those sub agents, those little like
- 34:16researchers and workers to use different
- 34:19models. So right here it's using its
- 34:21agentic loop and it's going to spin up
- 34:22these agents. And whenever you see these
- 34:24little colorful things, these are
- 34:26different agents that it's spun up. So
- 34:27right here, this one's called X. And
- 34:29when I open up X, it is actually this
- 34:31sub agent working right here. I can
- 34:33watch this sub agent go through its
- 34:35agentic loop and I can watch it like
- 34:37pull in data and everything like that.
- 34:38So this is the same thing as over here,
- 34:40but this one is just a different model
- 34:42and it's not the main session that
- 34:43you're currently talking to. You can see
- 34:45we have two others that started working.
- 34:46We have this one called YouTube Pulse.
- 34:47And once again, it's going through the
- 34:48same exact agentic loop. We have this
- 34:50one called X drama. We have this one
- 34:52called audience comments. And we can
- 34:54basically watch all of these work, but
- 34:56we don't have to manage them because
- 34:57once these agents are done, the main
- 34:59session here will say, "Okay, cool. All
- 35:01five are done. They reported back to me.
- 35:02Here's what they found." And this is
- 35:04really good for gathering different
- 35:06perspectives or researching out a bunch
- 35:08of things in parallel. And these
- 35:09intelligent models are really, really
- 35:11good at delegation. So sub agents are
- 35:13definitely something that you need to be
- 35:14using. So, while these sub agents are
- 35:16fanned out and they're working, let's
- 35:17move on to number 17, which is scheduled
- 35:20tasks. Scheduled tasks are awesome
- 35:22because this basically lets you have
- 35:24your codeex working autonomously for
- 35:25you. So, on the lefth hand side, if you
- 35:27click on scheduled, you can basically
- 35:28ask codecs to schedule things for you,
- 35:31reminders, updates, or run full
- 35:33sessions. Because all a scheduled task
- 35:35basically is is it injects a prompt into
- 35:37a codec session like this. So, it uses
- 35:39your same local files, it uses your same
- 35:41skills, it uses your model, it uses
- 35:44everything. thing the agentic loop it
- 35:45just does it automatically for you. So
- 35:47in here I you see that I could create
- 35:48one and I could set it up manually. I
- 35:50could have it run locally or I have
- 35:51could have it run on the cloud. I could
- 35:53also choose what project to work in. So
- 35:55Herk 2 or you know Hyperframes editor. I
- 35:57can also choose what chat. So every one
- 36:00of these routines could be a new chat or
- 36:02it could be a new chat for this task or
- 36:04it could be even in an existing thread
- 36:05that we already have going. And then you
- 36:07can basically choose when to schedule
- 36:09it. So, it could be daily, it could be
- 36:10hourly, it could be, you know, weekly,
- 36:12or you could get it really, really
- 36:14custom in here. And the cool thing is
- 36:15you can describe everything in here
- 36:16manually, or you could basically just
- 36:18ask Codex to make one for you. So, for
- 36:21example, I have these like seven or
- 36:23eight different ones that are running
- 36:25for um trading. They're literally
- 36:27trading $10,000 of my real money. And I
- 36:29have this set up as scheduled tasks so
- 36:30that they can basically all work
- 36:32together, talk to each other, check the
- 36:33market, make trades, things like that.
- 36:35And of course, you guessed it, these run
- 36:36in a different project right here called
- 36:39trading challenge. And they all run
- 36:40inside of this challenge thread. You can
- 36:42tell that this has scheduled tasks in it
- 36:43because there's a little clock icon next
- 36:45to that thread. And moving on to the
- 36:48very last concept we have for today,
- 36:49number 18, this is voice mode. Now,
- 36:52everything that you've learned so far,
- 36:54skills, sub aents, projects, agents, all
- 36:57of this can be controlled by using your
- 36:59voice to have a real conversation with
- 37:01Codeex. With your voice, you can manage
- 37:03different codec threads. You can
- 37:04delegate different agents. You can run
- 37:06skills. You can set up schedule tasks.
- 37:07You can build sites. You can do anything
- 37:09here with your voice. And all you have
- 37:11to do is come down to the bottom left
- 37:12and click on start new voice chat. So,
- 37:14forgive me here as I'm awkwardly like
- 37:15hunching down because my headphones
- 37:17don't reach far enough. But watch this.
- 37:19Hey, Codex. So, I need you to help me
- 37:21out real quick inside of my Herk 2
- 37:22project. Can you start a new thread to
- 37:24make a thumbnail? I'm making this
- 37:26YouTube video about 18 CEX concepts. So,
- 37:28if you could make me a thumbnail for my
- 37:30YouTube channel where you see me sitting
- 37:32at a desk working on a laptop and I'm
- 37:34pointing to the right and there's like
- 37:3618 different little AI agents. Um, yeah,
- 37:39just start start with that. And then
- 37:40once you have finished that thumbnail,
- 37:43could you send it to the thread that is
- 37:45currently working inside of my her 2 as
- 37:47well and it is doing research on YouTube
- 37:49and X and um looking at my channel and
- 37:52stuff like that. Could you just send
- 37:53that thumbnail to that thread?
- 37:54>> All right, let me check on that. Okay,
- 37:56I've got your thumbnail brief queued up
- 37:58with you with the laptop pointing toward
- 37:59the 18s using your existing photos for
- 38:02reference. And that found your research
- 38:04thread. So, I'll send the finished
- 38:05thumbnail there. Now, do you guys see
- 38:07how this is all coming together? It
- 38:08started a work tree chat. It was doing
- 38:10the agentic loop. It also created a goal
- 38:13down here. It literally created its own
- 38:14goal because it knew that we had this,
- 38:17you know, task for it and it started
- 38:18working. And you know, the voice chat's
- 38:20still going. So, I don't know what it's
- 38:21going to think of this, but I just
- 38:22wanted to call that out because all of
- 38:24what we just talked about is all coming
- 38:25together because even you can see over
- 38:27here, it started making that actual
- 38:30prompt. Like, it shot off this prompt
- 38:31right here and it started off this new
- 38:33thread. And now this is working in a
- 38:35work tree and it's going to create that
- 38:37thumbnail for us and then it's going to
- 38:38send it over into this actual chat and
- 38:41we'll see that actually happen. But as
- 38:42you keep navigating, like you can go to
- 38:44the web and you can go to, you know,
- 38:46other things and you can keep this voice
- 38:48working. So as you're just working on
- 38:50your laptop, it'll show you what's going
- 38:51on and you can keep talking to it. And
- 38:53you can also do all this on your phone
- 38:54as well because of the whole remote
- 38:56thing. You can use voice mode when
- 38:58you're on your phone, too. Okay, so that
- 38:59finished up and you can see what happens
- 39:01is it actually can send tasks or
- 39:04messages to each other, right? So what
- 39:06happened was this was the original
- 39:07thread that it created to make the image
- 39:09and it made this and then it sent it to
- 39:11this thread and said, "Hey Nate wanted
- 39:12me to send this to you. Here it is." So
- 39:14with just your voice, I know that was a
- 39:16very simple example, but you're able to
- 39:18coordinate a bunch of different codec
- 39:20threads, manage them, multitask while
- 39:22you're on the go, while you're flipping
- 39:23between different apps. Very cool. All
- 39:25right, so now that you guys have learned
- 39:26the core concepts of codecs, we're going
- 39:28to start to actually get in there and
- 39:31play around with some of this stuff. So,
- 39:32what I want you guys to do here is think
- 39:34about the way that we're actually going
- 39:35to start using codecs. Don't think of it
- 39:37as a tool that helps you build like AI
- 39:39workflows. Think of it as your new
- 39:41operating system. Basically, what I mean
- 39:43by that is you should be reaching for
- 39:45codeex whenever you want to do a new
- 39:46task. And that's the whole mindset shift
- 39:48behind becoming AI native, which is
- 39:49actually the book I wrote right here. As
- 39:51you can see, this book is all mindset
- 39:52oriented. It's basically just about how
- 39:54do you talk to AI? How do you use it?
- 39:56And how do you change the default to,
- 39:58oh, okay, I'm just going to open up
- 39:59Google Chrome and open up this new tab
- 40:00and just do the job. the old way, the
- 40:02manual way that I'm used to. How do you
- 40:04shift to, okay, I'm going to open up
- 40:05codecs and I'm going to use this as my
- 40:07operating system. I'm going to figure
- 40:08out how can I build a skill for this?
- 40:10How can I use the browser used to do
- 40:11this? How can I turn everything that I'm
- 40:13doing into faster processes because I've
- 40:16built systems around it? And that
- 40:17doesn't mean we're giving up a lot of
- 40:18the judgment or the control. That still
- 40:20sits with us, the human. It just means
- 40:21that we're able to move faster and we're
- 40:23able to do way more. And the way that
- 40:25that truly happens is when you build
- 40:26your AIOS and your second brain, meaning
- 40:29you're not just opening up a new chat.
- 40:31you know, a new chatbt tab every single
- 40:32time you have to do something. You're
- 40:34building a system where every single
- 40:35time that you talk to it, it gets
- 40:36smarter. Every single time that you
- 40:38build a deliverable, it saves it and it
- 40:40remembers it. You're essentially turning
- 40:41AI from, you know, what we used to call
- 40:43like a personal assistant or like an
- 40:44executive assistant. Not anymore. We're
- 40:46turning this into a co-founder. We're
- 40:47turning this into like a business
- 40:49partner that actually can work with you
- 40:50and keep up with you and remind you of
- 40:53things rather than like forgetting
- 40:54things or hallucinating things. So,
- 40:55we're going to jump into this next
- 40:56section and I'm going to teach you guys
- 40:57how to build your own AI operating
- 40:59system. And just a quick heads up, guys,
- 41:00throughout this course, I'm going to be
- 41:01playing different clips and different
- 41:02little segments that I might have
- 41:04recorded at a different time of day or,
- 41:05you know, last week or something. So, if
- 41:07I look a little bit different in some of
- 41:08these, that's why. But I have very
- 41:10intentionally laid out this entire
- 41:11course in the order that I think makes
- 41:12the most sense to follow it. So,
- 41:14hopefully that'll make sense. But yeah,
- 41:15let's get into the next piece and start
- 41:16building out your AIOS. GBT6 Astra is
- 41:19the most powerful AI model I've ever
- 41:20used. So, I turned it into my AI
- 41:22operating system and my second brain.
- 41:24And this is just a quick visual, but
- 41:25right now what you're seeing is my
- 41:27actual company. This is my second brain.
- 41:29This is everything that I know and
- 41:30everything about my business. And
- 41:32because I have everything actually
- 41:33stored and usable, that means so does
- 41:35GPT6 Astra. So when I say AIOS, what I
- 41:38mean by that is an AI operating system
- 41:39where now I can basically do anything I
- 41:42need to from right inside of this
- 41:43interface with Codeex with GPT6 Astra
- 41:46because now it doesn't feel like I'm
- 41:47just talking to a chatbot. It feels like
- 41:48I'm talking to a co-founder, someone
- 41:49that actually knows everything that's
- 41:51going on in my life and my business. So
- 41:53today I'm going to show you guys how you
- 41:54set up an AI operating system from
- 41:55scratch, but if you already have one
- 41:57going, how you can keep improving it.
- 41:58And I'm going to give away a bunch of
- 41:59free skills that help you set up exactly
- 42:01this and help you set up other things
- 42:02like auditing your AIOS, leveling it up,
- 42:05and just setting up the structure.
- 42:06Right? Now, what you'll notice here is
- 42:07that this isn't just a random dump of
- 42:09knowledge. Right here in blue, this is
- 42:11my business wiki. So it's like AIS
- 42:12coaching, our corporate structure, our
- 42:14Ammation Society Plus. And when I click
- 42:16in, we can see all of these things in
- 42:18these different relationships. I can
- 42:19also see how stale this is, so I might
- 42:21need to update it. and it calls out all
- 42:22these areas where like this number is
- 42:24old. I need to update this. But it also
- 42:26shows me all the different connections
- 42:27where this specific node is linked into
- 42:30every other area of my business. So like
- 42:31I said, that was business wiki. I also
- 42:33have meetings that I've added in here,
- 42:34which takes up a big section. I've got
- 42:36video knowledge. So every single YouTube
- 42:38video I've made, it has all that
- 42:39knowledge and it relates it back to
- 42:41everything in my business and in my
- 42:42meetings. Same thing with the clawed
- 42:44memory right here as you can see. Same
- 42:45thing with codeex memory. I've also got
- 42:47a bunch of projects and then I have
- 42:49skills and agents and this forms
- 42:51together my herk brain and I've even
- 42:53gone as far as to take this exact brain
- 42:55and put it in its own OS dashboard where
- 42:57I can basically have my calendar over
- 42:59here. Today's a Sunday so I don't have
- 43:00many meetings and this is fully
- 43:01functional, fully synced with my
- 43:03calendar. We've got these three stats,
- 43:04YouTube subs, AIS, AIS plus and these
- 43:06are fully synced. I can come in here and
- 43:08manage my brain. I can search for
- 43:10different things. I can look through
- 43:11different things. I also have my
- 43:12communication fully synced. So Slack,
- 43:13ClickUp and email and these are
- 43:15functional. I can respond to things from
- 43:16here. I can also see different meetings
- 43:17that I have coming up as well as past
- 43:19meetings with the AI summaries already
- 43:20generated. And then I can get community
- 43:22pulse. So top YouTube comments, top AIS
- 43:24plus posts, what people are stuck on, as
- 43:25well as industry pulse from things like
- 43:27YouTube and X. And all of this of course
- 43:29was made with GBT6 Astra. But really the
- 43:31idea of an AIO OS or an operating system
- 43:33is my framework that I call the four
- 43:35C's. So context, connections,
- 43:37capabilities, and cadence. And the way
- 43:39that I think about this is I think about
- 43:40these first two things as the second
- 43:42brain piece. So I think of this as your
- 43:44second brain and then over here we have
- 43:46the actual AIOS and this is the
- 43:48capabilities and the cadence. So let me
- 43:49real quick explain what I mean by that.
- 43:51What's the actual difference? So the way
- 43:53that you get this thing to actually feel
- 43:54like it knows you is by giving it
- 43:56context. So I mean things about you and
- 43:57your background and your business and
- 43:59your goals and your priorities and
- 44:01things like that. I think of context as
- 44:03the things that are in your life that
- 44:04really don't change too much. So
- 44:05quarterly goals or yearly goals or what
- 44:08your business does, your avatar pain
- 44:10points, things like that. and the
- 44:11connections. I think of things as the
- 44:12tools that you use every day because the
- 44:14data in there is always relevant and
- 44:16it's always changing every day. So, your
- 44:18email, different data is always in
- 44:19there. Your conversation, Slack,
- 44:21ClickUp, your project management, your
- 44:23financial information, giving your AIOS
- 44:25the ability to know all the context, but
- 44:27also reach for things just in time when
- 44:29it needs it. That's how you build a
- 44:30really, really powerful second brain
- 44:31that helps you actually ask your AIOS
- 44:33questions or say something and rather
- 44:35than getting a vague or generic
- 44:36response, it knows what to do. So, for
- 44:38example, when I asked Codex to make me
- 44:40this AIS sizzle reel, I didn't tell it
- 44:42everything I wanted. I just said, "Hey,
- 44:43make me a sizzle reel for AI Automation
- 44:45Society." And it knew all this context,
- 44:46and it also knew that it could go to my
- 44:48school community and look things up in
- 44:49order to make this. In fact, what you'll
- 44:51notice is it actually went so far as to
- 44:53go to the community, take pictures from
- 44:54it, like take screenshots itself, and
- 44:56add those in here because it knew that
- 44:58this is what I was talking about. So, I
- 45:00have these three little tests that I
- 45:02like to run, which is basically if a
- 45:03teammate messages you with a question
- 45:05and you realize that your AIOS would
- 45:07answer better and faster with the exact
- 45:09sources, then that's a good sign that
- 45:12your AIOS really is where it should be.
- 45:14Another one is context switching
- 45:15reduction. One thing that you should
- 45:17challenge yourself to do after you've
- 45:18gotten this set up is to try to force
- 45:20yourself to do everything from here,
- 45:22from this interface. rather than opening
- 45:24up all those new tabs and doing it the
- 45:25old way. Try doing it the new way of
- 45:27using the AI as your operating system.
- 45:29And then number three is the knowledge
- 45:31leaves your head. So stop trying to
- 45:32remember everything. You don't have to
- 45:34rehearse what you decided last quarter
- 45:36or what your customer said in that
- 45:37meeting because you trust the retrieval
- 45:39of asking your AIOS because it can find
- 45:41it in your Slack threads. It can find it
- 45:42in your email, find it in your meetings,
- 45:44or it just knows it. Because if you
- 45:46start to build new skills and
- 45:47capabilities and you start to put
- 45:49cadence into play, which means agents
- 45:50that run autonomously and automations
- 45:52and things like that, these things are
- 45:54not nearly as valuable without the first
- 45:56two C's because otherwise these skills
- 45:58are very generic. And the automations
- 45:59are just like I said, generic. So these
- 46:01four things really come one after the
- 46:03other, context, connections,
- 46:04capabilities, and cadence. And they all
- 46:05work together very beautifully. And
- 46:07that's how we get our AI operating
- 46:09system. So in order to get started or to
- 46:11scale all this up and get all the
- 46:12resources, I need you to go to my free
- 46:13school community. The link for that is
- 46:14down in the description. Like I said,
- 46:16it's completely free. And you're going
- 46:17to go to classroom. You're going to
- 46:18click on all YouTube resources. And
- 46:19you're going to grab the AISOS resource
- 46:21pack. This thing is going to come with
- 46:23five skills. It'll come with an onboard
- 46:24skill to get you set up. It'll set up
- 46:26your whole project structure, your files
- 46:28and your folders. Then you'll have an
- 46:29audit skill to do frequent checks on
- 46:31your AOS. You'll have a link skill,
- 46:33which helps you route to different files
- 46:35when you need to. You'll have a level
- 46:36up, which you can run after your audits
- 46:38that tell you, hey, this is probably
- 46:39some stuff you should do based on this
- 46:40audit. And then you have the 3D brain
- 46:42skill which helps you turn your
- 46:44knowledge into this actual 3D brain like
- 46:46you guys saw right over here. Now, yes,
- 46:48it is these four things. And when you
- 46:49start to run that resource pack, it'll
- 46:51onboard you and ask you questions and it
- 46:52will start to fill in this information
- 46:54which is great. But then what comes next
- 46:56is something that's super super
- 46:57important and that is called the
- 46:59agents.md.
- 47:01And so if you've been using claude, this
- 47:03is basically the claw. MD, but now this
- 47:04is the one that Codeex is more familiar
- 47:06with reading and it's the exact same
- 47:08thing. So, if you are coming over from
- 47:09Codeex, literally what I would do is I
- 47:11would just copy your claw.md and then
- 47:13just name it agents.mmd. It's the exact
- 47:15same thing. And just to show you guys
- 47:16what mine looks like, if I go into here
- 47:18and I go to my files, you can see that
- 47:20it's very simple. It's you are Nate
- 47:21Herki operating system. Here's your job.
- 47:23Agents.mmd and cloud.mmd contain the
- 47:25same stuff. And then I basically give it
- 47:27a few core operating rules like to be
- 47:29concise and I use bullet points and
- 47:30don't use m dashes and you know use the
- 47:32Oxford comma. But then I go into a
- 47:34routing map and that's the majority of
- 47:35my agents.mmd. I think of it as you know
- 47:38core rules but then I think of it as
- 47:39routing meaning can my agent understand
- 47:43where everything lives because if I open
- 47:45up my Herk 2 project which is just like
- 47:47we said a combination of folders and
- 47:48files and that's all this is displaying
- 47:50is folders and files and it's showing
- 47:52the relationship between all of that but
- 47:53the point I'm trying to make is that
- 47:55this is a lot right there's a lot of
- 47:57folders and files and I can go into my
- 47:58projects and there's a lot more in here
- 47:59and then I can go into things like you
- 48:01know my YouTube videos in here and it's
- 48:03even more but because all of this is
- 48:04laid out in a way that to me is
- 48:06intuitive That also means that my agents
- 48:08can crawl it a little bit better. But
- 48:09the more that I have information in here
- 48:11like routing logic, like, hey, if you
- 48:13need business advice or team stuff or
- 48:15OTAAS, you look in the wiki. If you need
- 48:17corporate structure or entities, you
- 48:18look in this section of the wiki. If you
- 48:20want to look at Nate's voice and style,
- 48:21you look here. If you want to look at
- 48:23course knowledge, you look here. If you
- 48:24want to look X, go Y. If you want to
- 48:25look A, go B. And you can see there's a
- 48:27ton of different rules in here, but
- 48:29what's important is that the agent knows
- 48:30this every single time it loads up so
- 48:32that when you ask for something, it
- 48:34knows where to look. Because that's
- 48:35essentially how the agents.mmd file
- 48:36works. You know, you say, "Hi, codeex."
- 48:39And before it even will read your
- 48:41message, it reads this first. So, it
- 48:42reads this and then it reads your
- 48:43message and then it will respond with
- 48:45like, "Hey, how can I help you today?"
- 48:47You know, that's kind of what it will
- 48:48say. But it always reads this context
- 48:50first, which is why it's so important.
- 48:52All right. So, if you're a complete
- 48:53beginner, how do you actually get set up
- 48:54with all this? Well, let me show you.
- 48:56First thing you're going to want to do
- 48:57is you're going to want to go to
- 48:58somewhere on your computer. Let's just
- 49:00say it's your desktop. And you're going
- 49:01to create a new folder. And so for
- 49:03example, this is where I am using my
- 49:05Herk 2 project. But for now, I'm just
- 49:07going to call this AISOS
- 49:11demo. So you can call this whatever you
- 49:12want. Now once that folder has been
- 49:14created locally, which just means on
- 49:16your computer, you're going to open that
- 49:17up in Codeex. So at this point, if you
- 49:19don't have the Codex desktop app,
- 49:21download that. And then if you aren't on
- 49:23a subscription, just get on
- 49:24subscription. But now what we're going
- 49:25to do is we need to open up this OS. So
- 49:27what that means is you would come in
- 49:28here and you would click on new project.
- 49:30You would create a local project and
- 49:32then you would give it a name and then
- 49:34choose this folder. So this is once
- 49:35again going to be AIS demo and I'm going
- 49:37to open up that folder. If I go to my
- 49:39desktop and I just have to find this
- 49:41folder we just made right here. Select
- 49:43that and then go ahead and click create
- 49:45project. So this now means whenever we
- 49:47are doing things in here, we're working
- 49:49inside of this actual local folder which
- 49:51if I open this up, there's nothing in
- 49:53here right now. But you will see as we
- 49:54start this whole onboarding, this folder
- 49:56will fill up. So, what I'm going to do
- 49:58is I'm going to go to the resource pack
- 49:59and I'm just going to copy this actual
- 50:01URL. And I will now paste that into the
- 50:04chat and say, "Hey, I want you to
- 50:06install this and just go ahead and run
- 50:07the onboard so I can start giving you
- 50:09some information and building my AIOS."
- 50:11So, now that this is installed, it's
- 50:13going to start onboarding you. And you
- 50:14can see that it's going to say, "I will
- 50:15save your answer as you go." So,
- 50:17obviously, you guys should take some
- 50:18time and give this some pretty specific
- 50:20information, but I'm just going to blow
- 50:22through this real quick. So, my name is
- 50:24Nate and I run an AI education community
- 50:27and we have a certification program and
- 50:28we have live events and things like
- 50:30that. And what's going to happen is it's
- 50:31going to start creating folders and
- 50:32files for you. If you click on this
- 50:33button and you go to files, you can see
- 50:35that it's already created a bunch of
- 50:37stuff in here. None of this was in here.
- 50:39If you guys remember, this was all
- 50:40completely empty, but now it's starting
- 50:42to set up the actual structure. But
- 50:43really, the two most important things to
- 50:44call out right now are the context
- 50:46folder. This is where it's going to
- 50:46start to create some markdown files of
- 50:49business context and also the
- 50:50agents.mmd. I remember I showed you guys
- 50:52mine and this is going to start getting
- 50:54built out over time with you guys right
- 50:56in here. Then it asks you to paste some
- 50:58of your recent writing so that it can
- 50:59start to build some skills around how
- 51:00you write. So what I'm going to do real
- 51:02quick is just basically fly through
- 51:03these seven questions and then you guys
- 51:04do this as well and then come back. All
- 51:06right. So now that I've done those seven
- 51:07questions, if I open up a new thread, I
- 51:09can actually say, "Hi, who am I?" And it
- 51:11will check my profile first because now
- 51:14it can actually look at all that and
- 51:15understand that. So if I open up the
- 51:17files once again, we can now see that if
- 51:19I go to my context, we have one called
- 51:22about business, we have one called about
- 51:23me, and we have one called priorities.
- 51:25So if I open this up, you can see that
- 51:26this has information. If I open this one
- 51:28up, this one has information, too. And
- 51:30over time, these will continuously get
- 51:32more filled with context. And this is
- 51:34context, like I said earlier when I was
- 51:36talking about these four C's, context
- 51:37that we want to basically be more
- 51:39evergreen, right? Information that's not
- 51:42changing very often. Whereas these are
- 51:44more the connections where you have
- 51:45information that changes very often. And
- 51:47that's when you're going to start
- 51:48filling it up by looking at things like
- 51:50this. You're going to start looking at
- 51:52um what bookmarks do you have? What
- 51:55desktop apps do you have? What apps do
- 51:56you use the most on the weekly basis?
- 51:58And I like to think about it as far as
- 51:59like revenue, customers, calendar, coms,
- 52:02tasks, meetings, and knowledge. And then
- 52:04once you've taken some time to like list
- 52:06them out here, just say, "Hey, Codeex, I
- 52:08want to connect to school. How can I do
- 52:10that? I want to connect to Stripe. I
- 52:11want to connect to YouTube. how can I do
- 52:12that? And it will tell you, oh, you need
- 52:14to get this API key or you need to use
- 52:15this MCP server or maybe you need to use
- 52:17browser use. And it will help you figure
- 52:18out the best way to connect to all these
- 52:20tools. But what else you'll notice is
- 52:21that it's creating the things that we
- 52:23need in case you ever want to go over to
- 52:24cloud code like acloud folder as well as
- 52:26a cloud.mmd. So it's really helping you
- 52:28make this your own AIOS, your own IP
- 52:31where you're not locked into one
- 52:32ecosystem. So anyways, I just wanted to
- 52:34call that out. But the agents and the
- 52:35cloudmd file are basically identical.
- 52:37Right now we have like the operating
- 52:39system, we have some skills, we have
- 52:41where things live, but right now this is
- 52:42pretty empty. So obviously we need to
- 52:44get this filled up a little bit more.
- 52:46You can also see that in the references
- 52:47section we have a voice. So if I go to
- 52:49my references and I go to the voice MD,
- 52:51we can see that this is stuff about how
- 52:53Nate talks based on the information that
- 52:55I've given it so far. So anyways, now
- 52:57you can see we're starting to get some
- 52:58stuff filled up here. But now what
- 53:00you'll do is once you have connected
- 53:02some things, you're going to go ahead
- 53:03and run the audit. So you can do /
- 53:04audit. This will automatically be
- 53:06installed because you use the resource
- 53:08guide. And this basically checks the
- 53:09AIOS. It looks at the four C's. It gives
- 53:11you a score. And it will also look at
- 53:13your agents MD or your cloud NMD and
- 53:15make sure everything's synced up like I
- 53:17said. And what else is cool about this
- 53:18is this creates another folder in your
- 53:21AIOS called audits. You can see there's
- 53:22nothing here called audits. But what's
- 53:24going to happen is it's going to give
- 53:26you a report and then it's going to
- 53:27store your audit. So every single time
- 53:29you run this, let's say you want to run
- 53:30it every Monday or every two weeks, it
- 53:32will store your scores every time so
- 53:34that you can see how you're actually
- 53:35improving your system, you know, month
- 53:37over month. So I know if you're new to
- 53:39Codeex and everything, this might feel a
- 53:40little bit intimidating, but it's really
- 53:42not too bad. And the cool thing about
- 53:43this is you can just ask questions if
- 53:45you ever get confused. Hey, why'd you do
- 53:46that? Or where should I put this? Or,
- 53:48you know, how do you feel about all
- 53:50this? You can treat this thing like your
- 53:52best friend who's really smart, who's
- 53:53also a mentor rather than just like a
- 53:56chatbot. Okay, but you can now see that
- 53:57this audit is done. And obviously, we
- 53:59scored a 30 out of 100 because we
- 54:01literally just built this and there's
- 54:02not much in here. But it'll show you
- 54:03across those four C's what needs
- 54:05improvement. And you can keep working on
- 54:06that stuff. And over here, you can see
- 54:08that we have a new folder called audits.
- 54:10We have this markdown file which shows
- 54:11us that today, September 7th, you ran
- 54:13this audit. And here was basically the
- 54:15conclusion of that audit so that you can
- 54:17improve on it. Now, what you can do from
- 54:19there is you can go ahead and run
- 54:20another skill which comes in the kit,
- 54:21which is called a level up. So, this one
- 54:23will basically look at this audit report
- 54:24and then it will help suggest areas for
- 54:27you to improve it. It'll help you think
- 54:28about different connections and
- 54:29different ways that you could, you know,
- 54:30build in some more capability and build
- 54:32in some more cadence so that this thing
- 54:34keeps growing and keeps growing. So,
- 54:35here you can see it said, I'd start with
- 54:37turning team updates into one clear
- 54:38action list. It targets your stated
- 54:40bottleneck coordination, taking time
- 54:41away from learning AI and making videos.
- 54:43So, two candidates here, which are team
- 54:46action list and certification
- 54:47improvement backlog. So, I would then
- 54:49think, okay, cool. So how do we actually
- 54:52start driving towards this? You know,
- 54:54what do you need to see? What
- 54:55connections do we need to set up? And
- 54:56then how do we build these automations?
- 54:59And so you can just keep running this
- 55:00loop of auditing, leveling up, building,
- 55:02auditing, leveling up, building. And
- 55:04this is the type of stuff that Astra is
- 55:06so good at because it's so so smart. It
- 55:09really feels like it's got some really
- 55:10incredible general intelligence,
- 55:12especially as you give it more and more
- 55:14data because it's really good at
- 55:16building automations. It can oneshot a
- 55:17lot of this stuff and it will help you
- 55:18build all these routines. So automating
- 55:21some of these little tasks is no
- 55:22problem. The real issue is getting
- 55:25everything from your head into the
- 55:27system because the more you give it, the
- 55:28more context you give it, the better it
- 55:30gets obviously. So yes, it can give you
- 55:31these recommendations, but now what I
- 55:33want you guys to think about is how you
- 55:35can give it more data. So I do have this
- 55:37other skill here called Grill Me. It was
- 55:38inspired by Matt PCO's Grill Me and I
- 55:41will actually put this in the AIS OS
- 55:43kit. So if you install it based on this
- 55:45video, you'll have this here. But I
- 55:46didn't do it right now. So anyways, then
- 55:48you could use the grill me and say, I
- 55:50need you to use this grill me skill to
- 55:52just, you know, find out as much as you
- 55:54can about my current priorities in my
- 55:56business. And what this does is it will
- 55:58basically do similar to the onboard
- 56:00where it asks you questions, but this is
- 56:02a separate skill where it just
- 56:03relentlessly grills you over and over.
- 56:05And what's cool about this is you just
- 56:07keep having it grill you on different
- 56:08topics because every time you start one
- 56:10of these interviews, it will create a
- 56:12new file in your AISOs and it will store
- 56:15all this interview. So every single time
- 56:17that you come in here for 20 minutes, 30
- 56:19minutes, and you just get interviewed by
- 56:20AI, it's going to store all that. So
- 56:22you're just building up your second
- 56:23brain over time, and it's really, really
- 56:25cool. As you can see, it said, I'm going
- 56:26to save every answer in business
- 56:28priorities capture. It created a new
- 56:30folder right here. And now we have this
- 56:31markdown file where as I start to answer
- 56:33these questions, it's going to store
- 56:35everything. So that's something that I
- 56:36would recommend going through with
- 56:37business priorities, your team, you
- 56:39know, your goals, and just like a bunch
- 56:41of different topics about your business
- 56:42and about your life. Now, the next thing
- 56:44you need to do, and it's how you
- 56:45actually get something more like this
- 56:47where you actually have like all of
- 56:48these relationships forming between
- 56:50these different things that you want to
- 56:51put into your AISOs or into your second
- 56:54brain. This gets really important
- 56:55because the relationships is what makes
- 56:57this actually connect to everything and
- 56:59what makes it feel more holistic rather
- 57:00than just like a data dump. So, the way
- 57:02that you get to this point with all
- 57:03these relationships is we're going to
- 57:05use Carpathy's LLM wiki. This basically
- 57:07just has AI crawl through data sources
- 57:09and find those connections for you. So,
- 57:11it's really, really cool. And guess
- 57:13what? It's really easy to use because
- 57:14all you have to do is you can either
- 57:16copy this URL, which I'll put in the
- 57:17description of this video. Or you could
- 57:19just come in here and you could copy
- 57:21this. Like you could literally just go
- 57:22like this, copy it, and then come into
- 57:24your actual AISOs. You can start a new
- 57:26chat and then say, "Hey, based on what
- 57:28you currently know about me, I want you
- 57:30to build me a new LLM wiki, and I want
- 57:32you to do so on all my business contacts
- 57:34and everything using this Carpathy wiki
- 57:37method." And then you can literally just
- 57:39paste in that prompt, and boom, it's
- 57:41going to crawl through everything. It's
- 57:42going to create you a new wiki vault.
- 57:43And you can obviously be more specific
- 57:45like I have one vault for my YouTube
- 57:47videos. I have one vault for my business
- 57:48knowledge. I have one vault for my
- 57:50meeting transcripts. And so I started to
- 57:52separate them out over time. But just to
- 57:54start, you could do, hey, this one's
- 57:55just for business or whatever you need
- 57:56to do. And it will help you find those
- 57:58relationships. And then once you've done
- 58:00a bunch of grillies, once you've built
- 58:01out these L&M wikis, then you're going
- 58:03to go in here and you're going to run
- 58:05that 3D brain skill. And that is the one
- 58:07that's going to take all your knowledge
- 58:08and it's going to turn it into something
- 58:09like this. And this obviously will
- 58:11continue to sync and stay optimized and
- 58:15continue to improve for you. And it will
- 58:16probably build that out into something
- 58:18called a local host. But what's really
- 58:19cool is you could ask it to turn that
- 58:21into a site. As you can see, Codex
- 58:23basically lets you build things and
- 58:24store it on their domain. So then after
- 58:26you build it, you could have it be
- 58:27accessible to different laptops or to
- 58:29your phone or whatever you want. Or if
- 58:30you wanted to keep it local, you can do
- 58:32that as well. And if you also wanted to
- 58:33connect your own custom domain, you can
- 58:35do that, too. So really, the
- 58:36possibilities here are endless. And it's
- 58:37just about doing that loop, like I said,
- 58:39of constantly scaling up your four C's
- 58:42and doing that loop of the audit level
- 58:44up, audit level up. Now, the last thing
- 58:46I wanted to address was the whole idea
- 58:48of GBD6 Astra driving everything, right?
- 58:51Because Astra is an incredibly powerful
- 58:53model, but I also wanted to say so is
- 58:555.6 Soul and so is 5.6 Terra. And for
- 58:58the majority of this knowledge work that
- 59:00you're doing here in your AOS, like
- 59:02asking, oh, can you find this doc or can
- 59:04you help me create this spreadsheet? A
- 59:06lot of that using Astra is probably
- 59:08overkill. Like Astra is one of the most
- 59:10it is the most capable model that I've
- 59:12ever played with. So what I would
- 59:13recommend is build out some of the
- 59:14stuff. You know, get familiar with how
- 59:16this model works. But then when you
- 59:17start to try to, you know, use different
- 59:20skills and when you start to try to
- 59:22connect to different things, I would
- 59:23recommend trying with 5.6 Soul because
- 59:26Astro is going to crawl through your
- 59:28usage limit, which is the weekly limit.
- 59:30It's going to eat at that more because
- 59:32it's a more expensive model than 5.6
- 59:34Soul is. And once you really start to
- 59:36get comfortable here, you can play with
- 59:37different things too, like the effort
- 59:38level. You can make it go on lights,
- 59:40which reduces some of the capability,
- 59:41but it also makes it cheaper. And then
- 59:43you could also go to 5.6 soul, and you
- 59:45could play with the effort level over
- 59:46here as well. There's a lot of cool
- 59:47things to dig into once you really start
- 59:49to connect a lot of stuff to your codeex
- 59:52system. Once again, I just think it's
- 59:53important to call out that all you're
- 59:55building here is you're building files
- 59:57and folders. And what's cool about that
- 59:59is as you build out your files and
- 1:00:00folders, it will grow into something
- 1:00:02that's just amazing. So now I have my
- 1:00:04herk 2 which like I said has everything
- 1:00:06in here. And the cool thing is if I
- 1:00:07wanted to build a Hermes agent I can
- 1:00:09connect that to this. If I go back to
- 1:00:10cloud code I connect it to this.
- 1:00:12Whatever is the new tool, whatever is
- 1:00:14the new AI model, I don't have to worry
- 1:00:16about switching over. I don't have to
- 1:00:17worry about being vendor locked because
- 1:00:19I own all of this stuff right here and
- 1:00:21every agent can crawl through it and use
- 1:00:22it. And that's really what we're
- 1:00:23building here. We're not just building a
- 1:00:25codec skill or a codeex hub or a little
- 1:00:28like 3D brain. We're building IP. We are
- 1:00:30building a knowledge base that is going
- 1:00:32to be the most valuable thing because
- 1:00:34now we can work through it because
- 1:00:36there's so much context and knowledge in
- 1:00:37here. So I really want you guys to
- 1:00:38seriously get in there and start doing a
- 1:00:40bunch of grill me and just building up
- 1:00:42your routing and your context. So
- 1:00:44remember free school community linked in
- 1:00:45the description. You'll find all the
- 1:00:46resources that you need right in here.
- 1:00:48So you've got the foundation set up.
- 1:00:50Obviously like I mentioned this is never
- 1:00:51a finished product. I've been building
- 1:00:52my own AIOS for about 9 months now and
- 1:00:55every day I make a change to it. I
- 1:00:56always make improvements. I always clean
- 1:00:58things up. And that will always just be
- 1:01:00the case. You're always going to keep
- 1:01:01improving that thing. But obviously, you
- 1:01:02can build things along the way that help
- 1:01:04you just become way more fast, way more
- 1:01:06efficient, way more powerful. And those
- 1:01:09are called skills. So, let's get into
- 1:01:10this next section where I teach you guys
- 1:01:12how to build skills better than
- 1:01:13basically anyone out there. And this is
- 1:01:14my proven skill building and skill using
- 1:01:17method. So, let's hop right in. Today,
- 1:01:19I've got this proven six-step process
- 1:01:21for building codec skills better than
- 1:01:2299% of people. So, let's not waste any
- 1:01:24time and just get straight into this
- 1:01:25one. All right. All right, so these are
- 1:01:26the six steps that I'm going to go over.
- 1:01:28I'm going to explain each of these and
- 1:01:29tell you why they're so important and
- 1:01:31then we're going to get into Codex and
- 1:01:32actually try some of the stuff out. And
- 1:01:33real quick before we get started, I'm
- 1:01:34going to talk about what actually is a
- 1:01:36skill. So, if you already know what a
- 1:01:37skill is, then just go ahead and skip
- 1:01:38past this part. But for those of you who
- 1:01:39need a refresher, here we go. So, I'm
- 1:01:41going to use my chocolate chip pancake
- 1:01:42analogy. Let's say we have this chef and
- 1:01:45this chef makes the most amazing
- 1:01:46chocolate chip pancakes and you want to
- 1:01:48know how to make those. This is
- 1:01:49essentially the output that you're
- 1:01:50looking for. The way that you would be
- 1:01:52able to copy this chef is by looking at
- 1:01:54the recipe that the chef used or the
- 1:01:56recipe that the chef made. So the chef
- 1:01:58gives you the recipe, this is you. And
- 1:02:00now you're able to pretty much make the
- 1:02:02exact same output, the really, you know,
- 1:02:04popular famous chocolate chip pancakes
- 1:02:06because you followed the recipe. So this
- 1:02:08recipe is basically the skill. This is
- 1:02:10essentially the skill.md file, which
- 1:02:12means a markdown file. It's just a
- 1:02:14simple language. It just means that
- 1:02:15inside of the skill file there are like
- 1:02:17pound signs and asterisks to indicate
- 1:02:19like bullet points and headers and
- 1:02:20things like that. So it's it's just
- 1:02:22natural language. But then the agent is
- 1:02:24basically able to take this skill file
- 1:02:26and just use it so that if you say,
- 1:02:27"Hey, Mr. AI agent, you know, make me
- 1:02:29those chocolate chip pancakes." It
- 1:02:30wouldn't have to be like, "Okay, well,
- 1:02:32let me just do some research on how to
- 1:02:33make them, and I don't know exactly what
- 1:02:34kind of pancakes Nate wants. I don't
- 1:02:36know how big they should be." So what
- 1:02:37I'll do is I'll just follow this skill.
- 1:02:39I'll follow the recipe. And now I get
- 1:02:40the same output that Nate's looking for
- 1:02:42and I get it consistently the same every
- 1:02:44single time because it's all documented
- 1:02:47for me right here. And these can be
- 1:02:48really simple. It can be a simple prompt
- 1:02:49like, hey, you know, help me turn this
- 1:02:51email into something that's more
- 1:02:53professional or something that sounds
- 1:02:54like me and maybe that's like my Nate
- 1:02:55email skill. But it could also be
- 1:02:57complicated processes like doing
- 1:02:58research on the market and analyzing,
- 1:03:01you know, 50 stocks and telling you
- 1:03:03which one to buy. So really, it's
- 1:03:04whenever you want to basically like
- 1:03:05codify some sort of process that you do
- 1:03:08and so that you can delegate that
- 1:03:09process to an agent and then you turn it
- 1:03:11into a skill and now your agent can use
- 1:03:13those skills. All right, so now that
- 1:03:14that is out of the way, let's start with
- 1:03:16number one up here where we have reverse
- 1:03:19engineer. So the whole idea with reverse
- 1:03:21engineering your skills is basically
- 1:03:22that you want to start with an output.
- 1:03:25You want to start with what is the
- 1:03:27definition of done? What are you
- 1:03:28actually looking for? Because let's say
- 1:03:29you ask your agent here for chicken, for
- 1:03:32example, and you actually want like
- 1:03:35chicken parmesan on a bed of pasta, but
- 1:03:38because you just said chicken, the agent
- 1:03:40might interpret that a little bit
- 1:03:41differently and make you a chicken
- 1:03:42sandwich. And then next time you ask for
- 1:03:44chicken, it might make you, you know,
- 1:03:45chicken thighs. It doesn't actually know
- 1:03:47specifically what you want. So if you
- 1:03:49start with an output and you have, you
- 1:03:51know, essentially this chicken parmesan
- 1:03:52and you say, "Okay, let's reverse
- 1:03:53engineer this food and see what went
- 1:03:55into it, how long we cooked it, how did
- 1:03:57we get here?" And that's how you build
- 1:03:59the recipe. So, for example, if you
- 1:04:00wanted to build a skill for the end of
- 1:04:02the week report, you've got certain
- 1:04:04columns in your Excel sheet. You've got
- 1:04:05certain calculations that were made.
- 1:04:07It's way easier to give the agent that
- 1:04:09Excel sheet to give it the final
- 1:04:10deliverable and say, "Hey, this is an
- 1:04:12output that is really good and this is
- 1:04:14what I want to build a skill for so that
- 1:04:16you understand how to take some raw
- 1:04:18input and turn it into this output that
- 1:04:20I have already told you that I like and
- 1:04:22this is what we're looking for every
- 1:04:23time." And then it can basically walk
- 1:04:24you backwards through that process.
- 1:04:25Okay, what data did you look at? where
- 1:04:27did you get it from? How did you
- 1:04:28calculate it? Where did you format it?
- 1:04:30You answer those questions and then you
- 1:04:31have a version of a skill that already
- 1:04:33knows sort of like the north star that
- 1:04:35it's building towards. I think it's so
- 1:04:36much easier to run the process, get the
- 1:04:39output and say, "Okay, let's turn that
- 1:04:40into a skill." Rather than saying, "Hey,
- 1:04:41build me a skill for building a YouTube
- 1:04:43dashboard." And then you might have this
- 1:04:46vision and your agent might have a
- 1:04:47completely different vision and then
- 1:04:48you're going to get frustrated when it
- 1:04:50delivers you a chicken sandwich and you
- 1:04:51actually wanted chicken parm. Now
- 1:04:53remember, all of these concepts I'm
- 1:04:54going to, you know, bring back together
- 1:04:55when we actually hop into Codex and I
- 1:04:57show you some stuff. But let's just keep
- 1:04:58moving down the list because I think
- 1:04:59this foundation is important. So number
- 1:05:01two, we have one specific job and one
- 1:05:04specific trigger. So remember how I said
- 1:05:05that these skills are markdown files.
- 1:05:08Skill.md. What happens in the skill.md
- 1:05:10is you've got a YAML description. It's
- 1:05:13called YAML front matter. And then
- 1:05:14you've got the actual skill
- 1:05:15instructions, which is the meat of the
- 1:05:17skill. So right here is one of the
- 1:05:18skills that I'm going to keep coming
- 1:05:20back to in this video because it's it's
- 1:05:21one of my favorite skills. And you can
- 1:05:23see up here this metadata, this is the
- 1:05:25YAML front matter. So if I view the
- 1:05:26source, it looks like this. It's
- 1:05:28actually just separated by these three
- 1:05:30dashes. So this would be the YAML front
- 1:05:31matter. And then everything below it
- 1:05:33would be the actual instructions of the
- 1:05:35skill. Now this is markdown. And when I
- 1:05:37said markdown is very simple. It is. You
- 1:05:39can see here are two pound signs. And
- 1:05:41that just indicates a header. We can see
- 1:05:43that we've got these little dashes which
- 1:05:45are bullet points. And so when markdown
- 1:05:47is actually rendered, it looks more
- 1:05:48something like this. And you can see
- 1:05:49it's just formatted better. So this is
- 1:05:51the way that I like to look at it. But
- 1:05:52if you click on view source inside of
- 1:05:54the Codex app, you can see the raw
- 1:05:55markdown file. But anyways, the metadata
- 1:05:58up here is the YAML front matter and
- 1:06:00that tells us what is the name of the
- 1:06:01skill and when do you explicitly use it.
- 1:06:04And there are other fields that can be
- 1:06:05populated in here like in this one we
- 1:06:06have an argument hint. And basically
- 1:06:08this X article skill is used when I ask
- 1:06:11my agent to turn an YouTube video into a
- 1:06:14long form X post. And so sometimes it'll
- 1:06:16feed in this argument which is my
- 1:06:18YouTube video URL that it will take, it
- 1:06:20will download, it will transcribe it, it
- 1:06:22will take screenshots of it, all that
- 1:06:23sort of, you know, all the stuff. And
- 1:06:25then once the agent has basically said,
- 1:06:26"Okay, cool. Nate wants to turn this
- 1:06:28YouTube video into an X article. So,
- 1:06:30okay, cool. I found this skill that I
- 1:06:31need to use. Now that I know this is the
- 1:06:32skill I'm going to invoke, let me read
- 1:06:35the entire description of the skill. So
- 1:06:37now I understand exactly what to
- 1:06:38actually do." And you can see this skill
- 1:06:40is pretty long. There's a lot of
- 1:06:41instructions in here. And this wasn't
- 1:06:43like me on one try turning this into a
- 1:06:45skill. This is probably a skill that's
- 1:06:47been iterated on 25 or more times. And
- 1:06:50we'll talk more about that in a bit. But
- 1:06:51the whole idea is that you want a skill
- 1:06:53to do one very specific job. You don't
- 1:06:54want to have a skill that's like run the
- 1:06:56marketing team. You want to have a skill
- 1:06:57that breaks that marketing team, all of
- 1:06:59those processes into individual steps.
- 1:07:01So in my book right up here, becoming AI
- 1:07:03native, I talk about this and I call it
- 1:07:05the function breakdown. I call it the
- 1:07:07the tree. Basically, your job is a tree
- 1:07:09and you've got different trunks with
- 1:07:11different like, you know, bullet points
- 1:07:12in your job description and then each of
- 1:07:13those trunks have branches and each of
- 1:07:15those branches have ultimately leaves
- 1:07:17and you want to turn all of those little
- 1:07:18leaves or those little tasks into skills
- 1:07:20because you can chain skills together
- 1:07:22over time. But the point being, if you
- 1:07:24have a skill with one very specific job,
- 1:07:27you can give it one very specific
- 1:07:28trigger, which means that your agents
- 1:07:30are going to be able to execute them
- 1:07:31more consistently, invoke them
- 1:07:32automatically. It just helps you
- 1:07:34separate out what your agents are doing
- 1:07:36in a much better way than if you have a
- 1:07:38skill that's like 30 pages long and it's
- 1:07:40supposed to do so many different things.
- 1:07:41So think about breaking down processes
- 1:07:43at the task level and turning each of
- 1:07:45those tasks into skills. All right,
- 1:07:47moving on to number three. We have
- 1:07:49thinking about the freedom level. So
- 1:07:51what I mean by this is when you're
- 1:07:53automating things, you basically have to
- 1:07:55figure out is this a deterministic
- 1:07:57automation or a non-deterministic
- 1:07:59automation? Basically, is this something
- 1:08:01that's predictable and we know exactly
- 1:08:02what comes in and we know exactly what
- 1:08:04happens and we know exactly what comes
- 1:08:05out or is it more of a non-deterministic
- 1:08:07AI agent, more of a nondeterministic
- 1:08:10black box where we know sort of what the
- 1:08:12input's going to look like. We sort of
- 1:08:14know what's going to happen inside the
- 1:08:15process, but we do kind of know like
- 1:08:17what the output should be. We still know
- 1:08:18the definition of good and the
- 1:08:20definition of done. You can think about
- 1:08:21it like, can I automate this using
- 1:08:23rules? Hard rules, hard logic. Is x
- 1:08:25greater than 10? Is x equal to 500? or
- 1:08:28is it something that needs a lot of
- 1:08:30judgment? Analyze this. Turn this into
- 1:08:32an email. Turn this into a spreadsheet.
- 1:08:34And it's really important to think about
- 1:08:35where does your skill live on that
- 1:08:37spectrum? Because if you have a very
- 1:08:39deterministic skill, like for example,
- 1:08:41it just has to take a Excel sheet and it
- 1:08:44just has to populate the cells into the
- 1:08:46CRM or something and that's basically
- 1:08:48just data processing transferring.
- 1:08:49That's a very deterministic process. And
- 1:08:52if you write the skill in a very
- 1:08:53nondeterministic way, there's a lot more
- 1:08:56room for error. There's a lot more room
- 1:08:57for AI to interpret something wrong and
- 1:09:00do something wrong. And that's the case
- 1:09:01where you want the skill to be like step
- 1:09:02one do this, step two, do this. Exactly,
- 1:09:05exactly. Exactly. Do this. It's very
- 1:09:06specific. But if you've got something
- 1:09:08nondeterministic, like for example, my X
- 1:09:10article skill, we don't want to be so
- 1:09:12specific. We want the AI to be able to
- 1:09:13use its judgment and to be able to use
- 1:09:15research and thinking to generate a
- 1:09:18unique output because every YouTube
- 1:09:20video is different, which means every X
- 1:09:21article is going to be different, which
- 1:09:22means all the screenshots that it needs
- 1:09:23to take are going to be different. I
- 1:09:24couldn't say something like screenshot
- 1:09:26the video at the 1 minute mark and then
- 1:09:27the two-minute mark and then the 5minute
- 1:09:29mark and put those into the article at
- 1:09:31line 400, line 600, line 700 because
- 1:09:33then the articles would come out feeling
- 1:09:35generic and they probably wouldn't even
- 1:09:36make sense and it just wouldn't be a
- 1:09:38very good output. So think about the
- 1:09:39freedom level of the process and when
- 1:09:41you're doing this manually, do you
- 1:09:43follow the same set of instructions
- 1:09:44every time or are you constantly using
- 1:09:46judgment and are you constantly doing
- 1:09:48different things within the process to
- 1:09:50ultimately get to that definition of
- 1:09:52good? By the way, guys, I've got this
- 1:09:53completely free SOP for you about
- 1:09:55getting your first a automation client.
- 1:09:57It's going to go over the exact steps
- 1:09:58that has been proven for hundreds of our
- 1:10:00AIS Plus members to get their first paid
- 1:10:02gigs. It goes over the one sentence
- 1:10:04service pitch that can get you started
- 1:10:05today, why your first client should cost
- 1:10:07you money, the 5-minute video that
- 1:10:08answers, can this person actually
- 1:10:09deliver before you've actually received
- 1:10:11any money, what to do when you have zero
- 1:10:13case studies. There's so many good
- 1:10:15things in here that are going to help
- 1:10:16you out. Even if you already do have
- 1:10:17clients, I would recommend grabbing this
- 1:10:19because, like I said, it's yours
- 1:10:21completely free. So, if you want to grab
- 1:10:22this, there's a link for it down in the
- 1:10:23description. Let's get back to the
- 1:10:24video. All right, moving on to number
- 1:10:26four, we have verification. And this is
- 1:10:28probably the most important piece of
- 1:10:32building skills. So, the whole idea of
- 1:10:34building a skill is that now you can
- 1:10:36trust that your agent's going to give
- 1:10:37you an output more consistently that
- 1:10:39meets your standard. But what you want
- 1:10:41to do in there is you don't want to be
- 1:10:43constantly being handed the agents V1.
- 1:10:46You want to be handed an output that's
- 1:10:48already good to go that you can give it
- 1:10:49a quick skim, approve it all, and then
- 1:10:51shoot it off. And the cool thing about
- 1:10:53agents building stuff to ultimately
- 1:10:54achieve some sort of northstar is that
- 1:10:57you can also have agents check its own
- 1:10:59work, verify its own work, or you know,
- 1:11:01other agents verify different agents
- 1:11:03work. So every single skill that I build
- 1:11:04works in some sort of verification loop,
- 1:11:06meaning the agent who builds the thing,
- 1:11:09delivers this output, and then we either
- 1:11:10have the same agent or different sub
- 1:11:12agents come in and verify that output
- 1:11:14and then provide feedback. And this
- 1:11:16verification loop can go on multiple
- 1:11:17times. Sometimes it's just, hey, here's
- 1:11:19the V2 and it's approved. Here's the V7.
- 1:11:21I've even had agents verify over and
- 1:11:23over and give me like a V15 or 16. And
- 1:11:26that basically just ensures that your
- 1:11:27time isn't being wasted verifying and
- 1:11:29sending feedback when you can have
- 1:11:30agents do that for you. Now, when it
- 1:11:32comes to verification, there is a
- 1:11:33difference between objective checks and
- 1:11:35subjective checks. So, let me explain
- 1:11:36what I mean by that. An objective check
- 1:11:38is something that can literally be
- 1:11:39proven. Sort of like the
- 1:11:40non-deterministic versus deterministic
- 1:11:42thing. Objective means okay um I wanted
- 1:11:45you to do research and then the
- 1:11:46verification was to pull in 500 sources
- 1:11:49cut that down to the best 250 and then
- 1:11:52every single fact that's in the article
- 1:11:54I want that to be double checked by
- 1:11:56second agent and that's objective we can
- 1:11:57literally prove 500 sources cut down to
- 1:11:59250 every single fact double checked by
- 1:12:01a different agent that's objective that
- 1:12:03is a rule does x equal y is a greater
- 1:12:05than b but then you have some things
- 1:12:07that are subjective like verifying the
- 1:12:09video right if it's it's creating a
- 1:12:10video for you or if it's creating a
- 1:12:11website for you. A lot of these
- 1:12:13subjective checks are more like, oh, you
- 1:12:16know, does everything look good? Is
- 1:12:17everything inbounds? Is it loading in a
- 1:12:19certain way? When you don't have an
- 1:12:21actual hard metric to align it to, you
- 1:12:23have to sort of be a little bit more
- 1:12:24subjective on your verification. And
- 1:12:26what you're thinking about doing is
- 1:12:27turning that agent into an LLM as a
- 1:12:30judge, which basically means, okay, so
- 1:12:31if AI is going to be applying its
- 1:12:34approval, and it has to use judgment to
- 1:12:36approve, then how can you tell the agent
- 1:12:38exactly what you're looking for? What
- 1:12:39does good typically look like or feel
- 1:12:42like even though there's not a hard
- 1:12:43rule? So, for example, with the X
- 1:12:44article skill, I've got some objective
- 1:12:46checks in there, like there are 10
- 1:12:47screenshots from the video. But then
- 1:12:49there's other subjective checks like how
- 1:12:51do you make it flow in a way that makes
- 1:12:53sense? How do you make it sound like
- 1:12:54Nate? How do you make sure that the
- 1:12:56images that are put into the the
- 1:12:58article, the screenshots, are cropped?
- 1:13:00There's other subjective checks that we
- 1:13:02have to let it do things where it like
- 1:13:04opens up the app and it scrolls through
- 1:13:05and it reads through and it screenshots
- 1:13:07and it verifies. But a lot of those
- 1:13:08checks, like I said, are more
- 1:13:09subjective. And so that's where
- 1:13:11iterating on these skills takes a lot of
- 1:13:13time because you're constantly running
- 1:13:14the skill, analyzing the output, giving
- 1:13:16feedback, running the skill again,
- 1:13:17analyzing the output, giving feedback.
- 1:13:19And you're basically just like training
- 1:13:20the skill where every single time you
- 1:13:22use it, it gets better. But the thing
- 1:13:23about verification is there's always a
- 1:13:25way. You might be thinking to yourself,
- 1:13:26well, I don't really know how I would
- 1:13:27have agents verify that process. Agents
- 1:13:29can do anything. They can use your
- 1:13:31computer. They can use your browser.
- 1:13:32They can look at things. They can listen
- 1:13:34to things. They can do anything. So just
- 1:13:35think about it like this. If you
- 1:13:36assigned that task to a human, what
- 1:13:38would you do to basically give it the
- 1:13:40stamp of approval? And then just explain
- 1:13:42that to the agent and that's your
- 1:13:44verification. That doesn't mean it's
- 1:13:45going to always come out 100%, but it's
- 1:13:47going to be more like 95% rather than
- 1:13:49giving you something that started off at
- 1:13:50like 75 or 80. Okay. And moving on to
- 1:13:52number five, we've got walk it down. And
- 1:13:54when I say walk it down, I mean walking
- 1:13:56it down the model list. So whether
- 1:13:58you're using Cloud Code or Codeex, we
- 1:14:00all know that different models have
- 1:14:02different strengths. And we all know
- 1:14:03that models that have the strongest
- 1:14:05strengths are the most expensive. So for
- 1:14:07example, Astra is more expensive than
- 1:14:08Soul, but more capable than Soul. Soul
- 1:14:10is more expensive than Terra, but more
- 1:14:12capable than Terra. So the idea is you
- 1:14:14build a skill and you're getting this
- 1:14:15good output. And maybe you were testing
- 1:14:17that with Astra. Okay, well let's run
- 1:14:19that same skill on Soul. Are we getting
- 1:14:20the same result? If we're getting the
- 1:14:22same result and it's cheaper, okay,
- 1:14:24let's test it on Terra now. Are we
- 1:14:25getting the same result? Is it cheaper?
- 1:14:27Okay, let's test it on Luna. There's no
- 1:14:28reason to be running a skill with Astra
- 1:14:30if you could run it with Luna and get
- 1:14:32the exact same output. But it is
- 1:14:34important to test it because not every
- 1:14:35skill fits that. I've had some skills
- 1:14:37where I run them on Haiku or Luna
- 1:14:39because they're just super simple. For
- 1:14:40example, when I give my agent a YouTube
- 1:14:42video and I say, "Hey, I need a YouTube
- 1:14:43video description and timestamps that I
- 1:14:45can do with Luna, and doing it with
- 1:14:47Astra is overkill. But for this ex
- 1:14:49article thing, I typically like to use
- 1:14:50at least Soul or Astra because it just
- 1:14:52has a better understanding of
- 1:14:54screenshotting things and putting it in
- 1:14:56the right spot and cropping it down and
- 1:14:57even blurring things out. It's so much
- 1:14:59better with the browser use with Soul
- 1:15:01and Astra. And then once you've landed
- 1:15:02on a model that you like, you can take
- 1:15:04that one step further and you can walk
- 1:15:06it down the effort level. So start it
- 1:15:08off on maybe high and if it's not good
- 1:15:10enough, then move it up a little bit.
- 1:15:11And if it is good, then move it down to
- 1:15:12medium and just keep finding basically
- 1:15:14the simplest model or the lightest and
- 1:15:17cheapest model that still executes at
- 1:15:19the level of quality that you're looking
- 1:15:20for. And then moving on here to number
- 1:15:21six, we have the bike method, which is
- 1:15:24I've kind of already alluded to it
- 1:15:25multiple times throughout this video so
- 1:15:26far, but really it's just the idea to me
- 1:15:28that your skill is never done. Every
- 1:15:30single time you run the skill, you're
- 1:15:32going to improve it. You're going to
- 1:15:33say, "Hey, here's what I really liked.
- 1:15:34Here's what I didn't like. Update the
- 1:15:36skill so that next time it's better."
- 1:15:37Almost every single time I run a skill,
- 1:15:39I give it feedback and I tell it to
- 1:15:41update. And if you think about this,
- 1:15:42like you're teaching a kid to ride a
- 1:15:43bike, that's what a skill really should
- 1:15:45feel like. You start off very cautious.
- 1:15:47You're watching everything. You're
- 1:15:48guiding them. You you've got your hand
- 1:15:50on the the the steering wheel. Steering
- 1:15:52wheel, I meant like handlebars. And
- 1:15:54you're you're right there. And then what
- 1:15:55happens is you give feedback. You say
- 1:15:56like, "Okay, that was good, but you
- 1:15:58know, you were leaning a bit too far to
- 1:15:59the left. Make sure your weight's more
- 1:16:00in the center." And then you keep going.
- 1:16:02Eventually, you take off the training
- 1:16:03wheels because the skill is getting
- 1:16:04better and you're getting more trust in
- 1:16:05the skill. and then you give more
- 1:16:07feedback again and then eventually you
- 1:16:08take off the elbow pads, right? And then
- 1:16:10you keep giving feedback. But that
- 1:16:11doesn't mean that you just take off the
- 1:16:12kid's helmet and let them bike on the
- 1:16:14highway. You're not going to do
- 1:16:15something reckless like that. You're
- 1:16:16also not just going to shoot them off
- 1:16:18down the road and then go inside and
- 1:16:19take a nap. You're still going to be in
- 1:16:21in some way watching or having certain
- 1:16:24guardrails in place in order to make you
- 1:16:25feel comfortable and in order for you to
- 1:16:27have more trust in the actual skill
- 1:16:28itself. And I truly believe in the idea
- 1:16:30that there's no such thing as a finished
- 1:16:31product unless your skill is super super
- 1:16:33deterministic and it runs perfect every
- 1:16:34time and you don't have to worry about
- 1:16:36it. But when the skills are more on the
- 1:16:37judgment side of the spectrum, you know,
- 1:16:39as we talked about up here, the more
- 1:16:41your skills live over here, the less I
- 1:16:44believe that you can actually have a
- 1:16:45finished skill because like I said,
- 1:16:46every time things might happen a little
- 1:16:47bit differently and every time your
- 1:16:49process might change or new models might
- 1:16:50drop and you're just constantly going to
- 1:16:52be giving feedback every time you use
- 1:16:53it. Okay, so let's hop into Codeex here
- 1:16:55and we can take a look at how this
- 1:16:57actually works. Now when you are inside
- 1:16:59of your project and if you're inside of
- 1:17:02your AIOS which is usually what I'm
- 1:17:03building in the codec skills will live
- 1:17:05in a folder called the aagents and then
- 1:17:08inside of the aagents there's a skill or
- 1:17:09sorry not a skill there's another folder
- 1:17:10calledskills and if you're doing this in
- 1:17:12cloud code the skills live in aclaude
- 1:17:15and then in a folder called skills now
- 1:17:17these can be transferable I basically
- 1:17:18just tell codeex or claude hey see all
- 1:17:21the skills in there and bring them over
- 1:17:22here too so just I have duplicates of
- 1:17:24them but that's where they actually live
- 1:17:26inside of your um project unless you
- 1:17:29have them at a global level which just
- 1:17:31means they're living locally across all
- 1:17:33of your codeex projects rather than just
- 1:17:35inside of your AOS. So that's where they
- 1:17:36live and you can always just say hey you
- 1:17:38know Codex I have this skill I don't
- 1:17:40remember where it is could you show me
- 1:17:41the file path and it will find it and
- 1:17:42you can move it if you want but anyways
- 1:17:44that's where they live and you can see I
- 1:17:45actually just ran this ex article skill
- 1:17:47on a video so it says your ex article
- 1:17:49draft is saved. I'll click into it and I
- 1:17:51actually did obviously already publish
- 1:17:53this one but I didn't even have to
- 1:17:54change anything. I basically just
- 1:17:55clicked in here. I read it all and I
- 1:17:57published it. And you can see that as
- 1:17:59it's going through, it already formatted
- 1:18:00everything and it used the video. Of
- 1:18:02course, it took screenshots of things
- 1:18:03and it like put them in the right spot
- 1:18:06to line up with what was actually being
- 1:18:08said in the article. And it even added
- 1:18:10this little um I don't know what you
- 1:18:11want to call it, like a spotlight. It
- 1:18:13added that. I didn't actually do that,
- 1:18:15but it will basically crop things out
- 1:18:17and it will add things like that into my
- 1:18:19actual article because I've told it in
- 1:18:21the skill to do things like that. And
- 1:18:22you can see that for this specific
- 1:18:24video, this ran on GBD6 Astra. So, what
- 1:18:27I'm going to do real quick is I'm going
- 1:18:28to walk this down. I'm going to give
- 1:18:30this same prompt to Soul and Terra and
- 1:18:33tell it to make the X article and it's
- 1:18:34going to use the skill. And we'll see
- 1:18:36how those outputs compare to Astra's
- 1:18:38output because this is a pretty complex
- 1:18:39skill. It has to transcribe the video.
- 1:18:41It has to open up the article. It has to
- 1:18:43make the thumbnail. It has to format
- 1:18:45everything and it has to screenshot
- 1:18:47everything and put it in and then
- 1:18:48actually like drag it around. So there's
- 1:18:49a lot of browser use and there is a ton
- 1:18:51of judgment inside of this. Like I said,
- 1:18:53this is not a deterministic skill. So
- 1:18:55you can see right now I have Soul,
- 1:18:57Terra, and Luna all running. It's been
- 1:18:59about 10 minutes, so I'll just check
- 1:19:00with you guys when this is done. But I
- 1:19:02just wanted to show you how all of them
- 1:19:03because of the skill, because of the
- 1:19:04verification, are viewing images. You
- 1:19:06can see how many times Luna has viewed
- 1:19:08these images here because it's taking
- 1:19:10screenshots and it's trying to
- 1:19:11understand, okay, which of these should
- 1:19:12be worked into the article and where and
- 1:19:14why. Same thing down here with Soul.
- 1:19:17Obviously, this one is a super super
- 1:19:18visualheavy sort of skill, but because
- 1:19:21I've worked that in there, all of them
- 1:19:22are viewing images no matter what. And
- 1:19:24so, it'll be really interesting to see
- 1:19:25how they decide to place them and, you
- 1:19:27know, what images they actually choose
- 1:19:28to put in there because X limits you to
- 1:19:30a certain amount of images per article.
- 1:19:32So, I'm going to let these run. We'll
- 1:19:33compare the results and we'll kind of
- 1:19:35talk about how walking it down has
- 1:19:37changed my perspective on this specific
- 1:19:39skill. All right, so you can see now
- 1:19:40that we have Luna and Terra have
- 1:19:42finished up. Soul is still working. So,
- 1:19:44we'll start reviewing these two and then
- 1:19:45we'll hop over to Soul. Hopefully, it'll
- 1:19:47be done by the time we're looking
- 1:19:48through these. So, Luna took 28 minutes
- 1:19:50and 33 seconds. Let's open up the draft
- 1:19:52and see what we are working with. So, we
- 1:19:55have um the thumbnail in there. We have
- 1:19:57I designed a one person million-dollar
- 1:19:58business with Claude. Um the title is
- 1:20:01the same as like the opening line, which
- 1:20:02I don't love. There's also this weird
- 1:20:04spacing was right there. I don't know if
- 1:20:06it tried to like center it, but I do
- 1:20:08notice that. Anyways, we have sort of
- 1:20:10like the TLDDR. We have these different
- 1:20:12headers here. I'm yet to see an image.
- 1:20:15Here comes the first screenshot. So,
- 1:20:16this is talking about um the three
- 1:20:18different types of ideas to compare. And
- 1:20:20in the screenshot, we do see all three.
- 1:20:22So, that matches up pretty well. Um we
- 1:20:24see agent report card as well as down
- 1:20:26here. So, really what I'm checking for
- 1:20:28right now is that it all sounds like me.
- 1:20:31So, this is kind of trained on like my
- 1:20:32LinkedIn style guide and my YouTube
- 1:20:34style guide and like the way that I
- 1:20:35speak, but I'm also making sure that the
- 1:20:37images look good. So, the test suite is
- 1:20:39a golden data set. We know what the
- 1:20:41correct answer or correct action should
- 1:20:43be. And we see an evaluation right here.
- 1:20:45The first run scored 88. That's what we
- 1:20:46see. Um, approved the new policy. Same
- 1:20:5016 tests. Ran it again. That looks
- 1:20:52decent. I clicked create report and we
- 1:20:56don't really see the report here. So,
- 1:20:59that's a little bit off, I would say,
- 1:21:01but it's not too bad. I'm actually
- 1:21:02impressed that Luna is doing it this
- 1:21:04well. This is customer support. This is
- 1:21:08kind of just the home dashboard. And
- 1:21:10then down here we see searching for
- 1:21:12companies using clay. It did do a nice
- 1:21:14little spotlight animation here when
- 1:21:16we're looking at the actual pricing. And
- 1:21:17then we've got the three Ps down here.
- 1:21:19Okay, so this is not too bad. We'll have
- 1:21:21to compare, but I would say that I was
- 1:21:22impressed by Luna's ability to
- 1:21:24screenshot, add spotlights, put it in
- 1:21:26the right spots. So that is Luna. That
- 1:21:28also took 28 minutes and 30 seconds.
- 1:21:30Okay, so we have Terara. And by the way,
- 1:21:32all of these are running on high. So we
- 1:21:33have Terara here. took 26 minutes. So
- 1:21:35about 2 minutes, 2 and a half minutes
- 1:21:37faster than Luna. Let's open up this
- 1:21:39draft. See what we got. So same exact um
- 1:21:43thumbnail. We have an image right away.
- 1:21:45How I built a $1 million AI business
- 1:21:48model with Claude. This image isn't
- 1:21:50great. Like to start off the article
- 1:21:51like that, that's not a very strong way
- 1:21:54to start it off cuz it looks like it's
- 1:21:55cut off like that. So that's bad. Um we
- 1:21:57go through the three filters.
- 1:22:00Oage
- 1:22:01doesn't look very good either. I would
- 1:22:03already say right now that um Luna did a
- 1:22:06better job. I mean, this isn't a bad
- 1:22:08Also, this is a zoomed in. So, it took a
- 1:22:10screenshot of the video and then it
- 1:22:11zoomed in on the score going from 88 to
- 1:22:1394. So, that's not terrible. Read
- 1:22:15complaint check company fit recommend
- 1:22:18small trial. What you'll notice here is
- 1:22:21it actually, if I can open this up, it
- 1:22:23redacted sensitive details. So, it put
- 1:22:25this block over it and said sensitive
- 1:22:28details redacted. So, I think that
- 1:22:29that's pretty cool. I like when it does
- 1:22:30things like that. It did a little
- 1:22:32spotlight effect here. It redacted more
- 1:22:34sensitive details in this one. As you
- 1:22:35can see, it did it again here. So, it's
- 1:22:37it did it again here. It It shows so
- 1:22:39many times when it needed to redact
- 1:22:41sensitive details. It did it again. It
- 1:22:42did it again. Okay, but this is
- 1:22:44obviously an issue. It didn't verify
- 1:22:45this good enough. Like, you can see that
- 1:22:47there's literally four images in a row
- 1:22:49down here. This one looks bad. I think
- 1:22:51that, you know, you you you have to use
- 1:22:52the browser to like drag these around.
- 1:22:54And I think that it just gave up or it
- 1:22:56stopped and thought it was done. So, so
- 1:22:59far I would say that Luna did a better
- 1:23:00job than Terara. And Terra was, you
- 1:23:02know, it's a more capable model
- 1:23:03apparently than Luna. It's also more
- 1:23:04expensive. So, right now, if I was
- 1:23:06choosing between these, I would choose
- 1:23:08Luna. But now, Soul just finished up and
- 1:23:10Soul took 38 minutes. So, normally when
- 1:23:12I do run this, I run this on Astra. I
- 1:23:13usually run this on Astra low, and I run
- 1:23:16it on fast mode to get it done faster.
- 1:23:18But, let's see. This did it. Let me open
- 1:23:20this up. Okay, so here is uh Soul's
- 1:23:23version. I built a $1 million business
- 1:23:25with Claude. We have the same sort of
- 1:23:27like TLDDR, the three filters. We have a
- 1:23:29picture of the three filters right here.
- 1:23:30So, that's not bad. Choosing the
- 1:23:32product. Okay, it's zoomed in here.
- 1:23:34These are all zoomed in, by the way. So,
- 1:23:35it's it's it's using more, I guess,
- 1:23:37vision intelligence to choose what to
- 1:23:39actually display in the image. You can
- 1:23:41see it redacted here the demo receipt.
- 1:23:43So, not too bad. It zoomed in a little
- 1:23:46bit here on the report, although we
- 1:23:48don't see the score 88. What it what it
- 1:23:49highlighted here was 14 tests pass and
- 1:23:52two failed instead of showing us the 88.
- 1:23:54So, just pointing that out. It also
- 1:23:56zoomed in down here on some analysis of
- 1:23:59that report. It did a little spotlight
- 1:24:01effect here to show the 94. It redacted
- 1:24:04more of these receipts. It did another
- 1:24:05spotlight here. It shows sales and
- 1:24:08support without losing control. So,
- 1:24:09that's not terrible. We've got this
- 1:24:11customer support image as well. We've
- 1:24:13got more internal source ID redacted.
- 1:24:15Another spotlight effect on the money.
- 1:24:17And yeah, so this one isn't too bad
- 1:24:20either. I would say honestly like I'm
- 1:24:22wondering if Luna's was better. And
- 1:24:24that's the thing, like these are so
- 1:24:25different and sometimes I have tutorial
- 1:24:26videos, sometimes I have more of this
- 1:24:28style where they're edited. So, it's
- 1:24:29very hard to just say, "Okay, I've run
- 1:24:31one test and I know that now I'm going
- 1:24:32to go with Luna instead of like Astra or
- 1:24:34something for for a specific example."
- 1:24:36Now, I will tell you guys, I've done
- 1:24:37this multiple times on multiple
- 1:24:38different YouTube videos and Astra just
- 1:24:40seems to be the best. So, I do stick
- 1:24:41with Astra on low, but I do think it's
- 1:24:43worth running them through a few
- 1:24:45different examples and and
- 1:24:46understanding. And that's why I said
- 1:24:47that there's really no such thing as a
- 1:24:48finished product, especially the more
- 1:24:50non-deterministic you get on that whole
- 1:24:52spectrum. And what you can see about all
- 1:24:54of these with Luna, it gave us a QA
- 1:24:56report, it gave us a visual plan, and it
- 1:24:58gave us selected frames. Terra gave us a
- 1:25:01QA report as well. And Soul gave us a QA
- 1:25:03report. And like I said, that's the most
- 1:25:05important part is just them proving that
- 1:25:07they've verified stuff. So, it verified
- 1:25:09the title, the body, 76 non-mpty text
- 1:25:12blocks, um 11 inline screenshots, seven
- 1:25:14H2 headings, visual review and privacy,
- 1:25:17editorial coverage, artifacts. It's
- 1:25:20proving to me that it's done. And it's
- 1:25:21proving to me that it did its
- 1:25:22verification. And trust me, if it didn't
- 1:25:23do this QA report and it didn't do the
- 1:25:25verification, this skill would be so
- 1:25:27much worse as far as actually getting it
- 1:25:29to place the screenshots in the right
- 1:25:31spot. So, for this specific example of
- 1:25:32the skill, let me talk about how I did
- 1:25:34all of these things. I took an article
- 1:25:36that I did manually and I did a few of
- 1:25:37these manually where I literally went
- 1:25:38through the YouTube video, screenshotted
- 1:25:40things, put it in the right spot and I
- 1:25:41gave it some of those to look at. So,
- 1:25:42first was like, hey, analyze all these
- 1:25:44outputs. I want to turn this into a
- 1:25:45skill. So, help me understand like
- 1:25:47what's good about these and and how do
- 1:25:49you choose the screenshots and how do
- 1:25:50you place them and things like that. So,
- 1:25:51I reverse engineered and then obviously
- 1:25:53this is one very specific job. It was to
- 1:25:55turn a YouTube video into an X article.
- 1:25:57It's not just to write X articles in
- 1:25:58general. We've got I've got a different
- 1:25:59skill for that. This is literally to
- 1:26:01turn a YouTube video into an X article,
- 1:26:03which means the trigger is very specific
- 1:26:05as well. So here, all I said was turn
- 1:26:07this YouTube video into an X article.
- 1:26:08And it automatically was able to invoke
- 1:26:10the X article skill as you can see. But
- 1:26:12if I wanted to do a slash command and do
- 1:26:15X article, I can invoke the skill like
- 1:26:17that as well. But typically, if I say,
- 1:26:18hey, I want this YouTube video to be
- 1:26:20turned into an X article, it just
- 1:26:21invokes the skill automatically because
- 1:26:23it's one specific use case and it's one
- 1:26:24specific trigger. We talked about the
- 1:26:25freedom level in here a lot. This is
- 1:26:27obviously a super super non-eterministic
- 1:26:28process. We talked about the
- 1:26:30verification. We did the walking it down
- 1:26:32live. And as far as the bike method,
- 1:26:33like I said, guys, I would basically
- 1:26:35come into here and I would look at what
- 1:26:36Tara did that I didn't like. And I would
- 1:26:38basically be like, "Okay, Tara, so in
- 1:26:40your article, you had some really weird
- 1:26:42screenshots in the beginning. It looked
- 1:26:44like things were cropped out." The whole
- 1:26:45idea is that people can look at the
- 1:26:46screenshot and they won't need context
- 1:26:48of what happened a few seconds before
- 1:26:50that frame or a few seconds after. And
- 1:26:52you have multiple examples in here like
- 1:26:53the first screenshot you did as well as
- 1:26:55the second screenshot that you did that
- 1:26:56just don't look very good. You also have
- 1:26:58things at the end like these random four
- 1:27:00images that don't feel like they're
- 1:27:01placed right. So I feel like your
- 1:27:03verification wasn't good enough, but
- 1:27:04also the third the third image in this
- 1:27:07set of or sorry the second image in this
- 1:27:09set of three at the bottom. It's like
- 1:27:10halfway through an animation. So that
- 1:27:12was first of all a bad screenshot to
- 1:27:14take and second of all a bad place to
- 1:27:15put it. So that's some of the feedback
- 1:27:17that I have for you. Now if you could
- 1:27:19analyze this, understand what I mean by
- 1:27:20this and then update the skill so that
- 1:27:22you don't make these mistakes in the
- 1:27:23future and so that you can highlight
- 1:27:24some of the important things that I just
- 1:27:26mentioned blah blah blah. And then I
- 1:27:27would just shoot that off and it would
- 1:27:28go through, it would look at the
- 1:27:29feedback and it would take all of that
- 1:27:31information and it would update the
- 1:27:32skill. And that is how you start to sort
- 1:27:34of see that bike method coming into play
- 1:27:36every single run, give good feedback and
- 1:27:38also give bad feedback and tell it to
- 1:27:40update the skill. Otherwise, all of
- 1:27:42those changes that you have to make
- 1:27:43manually, you're going to have to make
- 1:27:44manually again the next time. So anyways
- 1:27:46guys, those were the six tips. Those
- 1:27:48were all of them ran on a real example
- 1:27:49of one of the skills that I use pretty
- 1:27:51much every day. Every time I make a
- 1:27:52YouTube video. So I appreciate you guys
- 1:27:54watching through all that. And also
- 1:27:56guys, real quick, just wanted to tell
- 1:27:57you about my second channel where I just
- 1:27:58kind of like document these things that
- 1:28:00I'm doing at AI events or these
- 1:28:02different trips I'm taking to meet up
- 1:28:03with AI creators or speaking at things.
- 1:28:04And if you want to check out sort of
- 1:28:06like what else I'm into besides just
- 1:28:07like making tutorials, then head over to
- 1:28:09the second channel. I'll put the link in
- 1:28:11the description and I hope to see you
- 1:28:12guys over there. I hope you enjoy this
- 1:28:13type of content as well. All right, now
- 1:28:14that you guys understand how skills work
- 1:28:16and how you use them, let's talk about a
- 1:28:18few specific examples of how you can
- 1:28:20start to build skills that are very,
- 1:28:21very specific to you and they feel like
- 1:28:24they're coming from you and they feel
- 1:28:24like they're coming from your business
- 1:28:26when you're creating deliverables or and
- 1:28:28or when you want your team to create
- 1:28:30deliverables that sound like you guys
- 1:28:31and that they all look the same and
- 1:28:33things like that. So, let's hop into
- 1:28:34this next section. Okay, so we've built
- 1:28:36our AIOS, we've got that all set up,
- 1:28:38we've talked about skills, now let's
- 1:28:39start to talk about some specific use
- 1:28:41cases of using your business and skills.
- 1:28:44So, what I wanted to talk about here is
- 1:28:45how we can start to create some branded
- 1:28:47deliverables like sheets or docs or
- 1:28:49slideshows or, you know, even videos and
- 1:28:51websites and all that kind of good
- 1:28:53stuff. But one thing that you want to
- 1:28:55get set up is kind of like your brand
- 1:28:57guidelines, your brand rules, logos,
- 1:28:59color schemes, typography, things of
- 1:29:01that nature. So, right here we're inside
- 1:29:03of the AIS OS that we have set up. And
- 1:29:05you can see that I'm said, hey, what's
- 1:29:07my company in this AIOS? It's up at AAI.
- 1:29:08It's a Automation Society. And if I go
- 1:29:10in here, we have all of our files and
- 1:29:12folders set up. So, if you are ever in a
- 1:29:14situation where you want to add some
- 1:29:16context or you want to do something with
- 1:29:17your AIS or sorry, not your ASOS, your
- 1:29:20AIOS, and you don't really know exactly
- 1:29:22where things should live and you feel
- 1:29:23like you're going to do something wrong.
- 1:29:24The first thing I want to say is I don't
- 1:29:25think there is such thing as wrong.
- 1:29:27There's not like one universal standard
- 1:29:28because it's AI. AI is able to search
- 1:29:30through things and understand your rules
- 1:29:31and every single person's AIOS is going
- 1:29:33to be different, but no one's is like
- 1:29:35wrong. The only way it's wrong is if
- 1:29:36it's going to hallucinate things or if
- 1:29:38it can't find things and if there's way
- 1:29:40too much data, things like that. So, if
- 1:29:42you're noticing that it's lying or that
- 1:29:43it can't find things or that it's making
- 1:29:45up details, that's where you probably
- 1:29:47did something wrong and you just need to
- 1:29:49sort of run an audit, analyze things,
- 1:29:50and clean some things up. And once
- 1:29:51again, you can do that with natural
- 1:29:52language. But let's say I wanted to add
- 1:29:54some like logos and some color schemes
- 1:29:55and things like that. I would say, "Hey,
- 1:29:58so I want to add some logos. I want to
- 1:29:59give you some more brand guidelines so
- 1:30:01that in the future when we're making
- 1:30:02things like docs or any deliverables,
- 1:30:04you can make sure it feels branded to
- 1:30:06me." meaning like my voice but also you
- 1:30:09know physical styling when it comes to
- 1:30:11these deliverables. So I want to have
- 1:30:13some sort of folder inside of the AIOS
- 1:30:15called brand guidelines. Where should we
- 1:30:17put that based on our current folder
- 1:30:19structure
- 1:30:22and we will go ahead and shoot that off.
- 1:30:23So it made the suggestion and it
- 1:30:25suggested what we could put in there and
- 1:30:26I'm just going to say yep go ahead and
- 1:30:27set that up. And now you can see that it
- 1:30:29created that folder right here brand
- 1:30:30assets and we have some things in there
- 1:30:32like a few markdown files some spaces
- 1:30:34for logos and things. And so I opened up
- 1:30:36this in my actual folder right here. You
- 1:30:38can see this is the U brand assets and I
- 1:30:41clicked into logos. I clicked into uppai
- 1:30:43and then I opened up another file
- 1:30:44explorer where I actually just want to
- 1:30:45drag in some logos. So I'm going to copy
- 1:30:47these three and I'm going to copy them
- 1:30:51and I'm just going to come and paste
- 1:30:52them right in here. So now we have three
- 1:30:53upi logos. And then I could also do the
- 1:30:55same thing. I'll copy these four and I
- 1:30:58will paste these into the AI automation
- 1:31:00society side over here. So now we just
- 1:31:01have a few logos set up. And now you can
- 1:31:03see that those are showing up here in
- 1:31:05our actual AIOS. Now what I would
- 1:31:07recommend doing from here is yes, it's
- 1:31:09good to have logos in there. But I
- 1:31:10probably want to brainstorm a little bit
- 1:31:12on what are our brand guidelines. If you
- 1:31:15already have that, like if you already
- 1:31:16have your colors and your typography,
- 1:31:18like the fonts that you use, then just
- 1:31:19go ahead and give that information to
- 1:31:21codeex right here and it'll put it in
- 1:31:23there. But if you need to brainstorm on
- 1:31:24it, then just brainstorm on it a little
- 1:31:26bit. And what else you can do is have it
- 1:31:27generate you a brand guidelines doc. So
- 1:31:30watch this. You can see that I just
- 1:31:31dropped in some logos for UPAI. Now, in
- 1:31:34there, I have a few different logos.
- 1:31:35I've got like a white one and a light
- 1:31:36blue one. The light blue one is the main
- 1:31:39logo. That's like our primary logo, and
- 1:31:40that's what I want you to always add on
- 1:31:42our deliverables. Like in the docs, I
- 1:31:44want you to have it as a header. In the
- 1:31:45slides, I want you to have it like in
- 1:31:46the bottom left or something. And the
- 1:31:48colors are pretty much that same way.
- 1:31:50You know, we like to use white and we
- 1:31:52like to use that light blue. And I
- 1:31:54always want our deliverables to feel
- 1:31:56professional and they should feel
- 1:31:58trusted. And so we really like to
- 1:31:59utilize that light blue and those light
- 1:32:01colors. I don't exactly know what type
- 1:32:03of font that I like to use, but I don't
- 1:32:05want it to feel like too techy. I want
- 1:32:06it to feel very modern and clean. So
- 1:32:08could you just go ahead and generate me
- 1:32:10a brand guidelines sheet, you know, like
- 1:32:12an actual image that shows um the way
- 1:32:15that we like to have our buttons, the
- 1:32:16way that we like to have um logos,
- 1:32:18colors, things like that. And that's
- 1:32:20actually how I've made all of these
- 1:32:22different images that I have where we
- 1:32:23see like our brand guidelines with, you
- 1:32:25know, Nate Herk. It has like these
- 1:32:27colors. it has these sorts of buttons
- 1:32:28and all this stuff like that so that as
- 1:32:30we're generating new things in the
- 1:32:31future, I can always just point my AI
- 1:32:33towards that file and that's like the
- 1:32:35source of truth. And I like to usually
- 1:32:36have one that's a markdown file so it's
- 1:32:38easier for the AI to read and you can
- 1:32:40see the actual hex codes and everything
- 1:32:41like that and the font names, but then I
- 1:32:43also like to have one that's like an
- 1:32:44image in case you want to send that to
- 1:32:45your team or send that to a sponsor or
- 1:32:48something like that. And so while that
- 1:32:49is making that, I'll show you guys some
- 1:32:50other examples and I'll also talk about
- 1:32:51like why this is so important. And real
- 1:32:53quick, you can see that inside of the
- 1:32:54readme of the brand assets folder, here
- 1:32:56is what it generated. It says, "Hey,
- 1:32:58this folder is the visual source of
- 1:32:59truth for UPAI and AIS deliverables. It
- 1:33:01keeps approved assets and usage rules
- 1:33:03findable, so future work uses the brand
- 1:33:04instead of guessing for any document,
- 1:33:06PDF, presentation, report, whatever.
- 1:33:08Read this, read this, read here, go look
- 1:33:11at the templates, go look at the logos,
- 1:33:13do all this." And so it basically is
- 1:33:14just creating this read me so that your
- 1:33:16agent always understands why does this
- 1:33:17folder exist and where should it look
- 1:33:19for different things like fonts and
- 1:33:20templates and examples and colors and
- 1:33:22whatever else you want to give it in
- 1:33:24here. So you can see right now it's
- 1:33:25using its generate image tool to
- 1:33:26actually make that for us. And then we
- 1:33:27can turn that into a markdown file
- 1:33:28that's a little bit more structured. And
- 1:33:30there you go. You can see that this came
- 1:33:31out super super quick for us. I can
- 1:33:32click into this. We see we have the
- 1:33:34primary logo. We have the reverse. We
- 1:33:36have a primary action button. We have a
- 1:33:37secondary button. We have things about
- 1:33:39placements like docs and like slides. We
- 1:33:41have typography that it shows. We have a
- 1:33:43color palette right here as well with
- 1:33:44different actual hex codes that we could
- 1:33:46put into Canva or to whatever we need to
- 1:33:48put them in. And obviously I could mark
- 1:33:49this up and I could make comments. If I
- 1:33:51don't like this and I wanted to comment
- 1:33:52on that or if I wanted to change
- 1:33:53anything. That's what's so cool about
- 1:33:54the desktop app is you can do so many
- 1:33:56things right from inside here. You can
- 1:33:58even remove backgrounds. So you can play
- 1:34:00around with this until you get it the
- 1:34:01way you want it and then say, "Hey, can
- 1:34:02you also turn that into, you know, like
- 1:34:03a markdown version for me or a text
- 1:34:05version for me so that different agents
- 1:34:07can read that easier." But look how easy
- 1:34:08I was able to turn one logo into a
- 1:34:10formal brand guidelines doc. And now
- 1:34:12that you have that, what you can do is
- 1:34:13you can generate deliverables with this
- 1:34:15in mind or websites or videos or
- 1:34:17whatever it is. But also you work that
- 1:34:19into the skills. And now I have skills
- 1:34:21like this. This is a student resource
- 1:34:23guide skill. So whenever I have a
- 1:34:24YouTube video or I read an article or I
- 1:34:25just want to do a brain dump and I say,
- 1:34:26"Hey, can you turn this into a student
- 1:34:28resource guide?" It does all the stuff.
- 1:34:30It puts the AIS header up here. It uses
- 1:34:32this font. It uses these sorts of colors
- 1:34:34and headers. It links my YouTube channel
- 1:34:35at the top. You can see this link right
- 1:34:37here to my YouTube channel. It puts
- 1:34:38these horizontal bars. It does all of
- 1:34:40this in the way that I love it. And then
- 1:34:41at the very bottom, what it does is it
- 1:34:43also links people to AI automation
- 1:34:45society plus. And this is because I gave
- 1:34:47it the skill with my brand guidelines
- 1:34:49and with also, you know, of course, my
- 1:34:51rules for building the student resource
- 1:34:52guide. I also have one for writing up at
- 1:34:54AAI memos. It has a UP at AAI header up
- 1:34:56here. It has this sort of like um I I
- 1:34:59guess I was about to say YAML front
- 1:35:01matter as if this is a skill. Whatever
- 1:35:02this whatever you call this, like a
- 1:35:04little header. Once again, it has like
- 1:35:06different colors. It has these the way
- 1:35:07that I structure this. It even has the
- 1:35:09footer here, the confidential internal
- 1:35:10UPAI communication footer. It has all of
- 1:35:13this now because once again, I gave it
- 1:35:14brand guidelines, but also I told it how
- 1:35:16I like to write our internal memos. It
- 1:35:18also works with things like sheets. This
- 1:35:20is one that I run whenever I want to do
- 1:35:21like a YouTube report, whether it's
- 1:35:22quarterly or whether it's threequarters.
- 1:35:24Look at all this stuff. It has colors on
- 1:35:26the tabs down here, so I can easily
- 1:35:27navigate between all these tabs. It has,
- 1:35:29you know, colors inside of this. You can
- 1:35:31see that the font isn't just the basic
- 1:35:33font. It's using Monsterat for the font
- 1:35:34because that's on our brand guideline
- 1:35:36doc. It put the logo up here. It has big
- 1:35:38hero text up here. And then as I go
- 1:35:40through, I'm getting data visualized for
- 1:35:42me as well. I'm getting charts. I'm
- 1:35:43getting different things like this
- 1:35:44because this is the way that I told it.
- 1:35:45I like to see it. And not only is it
- 1:35:47giving me all of this data with all of
- 1:35:49these links and all of these, you know,
- 1:35:50interesting insights, but the whole
- 1:35:52thing feels branded to me. It feels like
- 1:35:55something I would send to my team and
- 1:35:56they'd be like, "Yeah, this is an upai
- 1:35:57sheet." And that's what I want you guys
- 1:35:59to realize is that every single time
- 1:36:00that you're generating outputs with your
- 1:36:02AIOS, you can have it be exactly the way
- 1:36:05you want it. And you can give the brand
- 1:36:06guidelines and you can give the skills
- 1:36:08to your team. So now everyone is
- 1:36:09generating things that look exactly the
- 1:36:11same. So that it all feels like it's
- 1:36:13coming from your brand voice so that it
- 1:36:14all looks consistent. And then when you
- 1:36:16guys start to push out things like
- 1:36:17articles or blog posts or LinkedIn
- 1:36:18posts, it always also has the brand feel
- 1:36:21as well. And it doesn't just have to be
- 1:36:22Google Sheets. It can be PDFs or HTML
- 1:36:24docs or whatever you want. But here's
- 1:36:26another example with a slideshow. This
- 1:36:28is a slide deck where we can see we have
- 1:36:30the branding up here. Once again, this
- 1:36:32all just feels consistent because it's
- 1:36:33using the same color schemes and it's
- 1:36:35building it in a way that feels like
- 1:36:36it's an UPAI doc. And obviously, there's
- 1:36:38some other things that I put into the
- 1:36:39skill here like having these sorts of
- 1:36:40visualizations, having, you know, swim
- 1:36:42lanes or um calendars or having actual
- 1:36:45data being visualized here with charts
- 1:36:46and things like that. And what's cool
- 1:36:47about this one is it also generated this
- 1:36:50Excel sheet or Google sheet to go with
- 1:36:51it because it had to actually generate
- 1:36:53these visualizations. And it did this in
- 1:36:55Google Sheet and then it brought all
- 1:36:57these over and put them into the actual
- 1:36:58slide deck that it built which I was
- 1:37:00just on the other tab for. And you can
- 1:37:02see that was actually right here. It was
- 1:37:03the mock data slide deck. We have all of
- 1:37:05these different things with the
- 1:37:06branding. It also created um a different
- 1:37:09workbook which was the Google sheet to
- 1:37:11make all of those charts. And like I
- 1:37:13said for all of these we talked about
- 1:37:14how with skills we like to sort of
- 1:37:16reverse engineer them. So what I did was
- 1:37:18I basically used codeex to help me get
- 1:37:20to a place where I liked an output. So
- 1:37:22let's say the student resource guide. I
- 1:37:24went back and forth. Now I have an
- 1:37:25output that I like and now I turn this
- 1:37:27into a skill. And in that skill I
- 1:37:28basically say, "Hey, by the way, include
- 1:37:30all of these things from the brand
- 1:37:32assets. Include the header up top.
- 1:37:33Include the logo down there. Include
- 1:37:34these colors, whatever." And the reason
- 1:37:36I really like to do it inside of Google
- 1:37:38is because we work out of Google Drive.
- 1:37:39So it also knows when it's creating
- 1:37:41deliverables to put it in the share
- 1:37:42drive, to put it inside of the education
- 1:37:45folder or the marketing folder or the
- 1:37:46AIS folder or wherever it needs to go.
- 1:37:48It knows to just throw it right there so
- 1:37:50that my team can access it
- 1:37:51automatically. Because when you're
- 1:37:53building things locally like this um
- 1:37:54HTML report for example, there's not
- 1:37:56really a problem with that. It's just
- 1:37:57less editable because now if I wanted to
- 1:37:59edit this, I would have to manipulate
- 1:38:00the code and I would have to say, "Hey,
- 1:38:02you know, Codex, can you change this
- 1:38:03line or can you make this a different
- 1:38:05color?" Whereas now in here, if I wanted
- 1:38:07to make a quick edit, I could just come
- 1:38:08in here and make a quick edit. And then
- 1:38:10I can also send this to my team right
- 1:38:11away. With this, I would have to like
- 1:38:13download this and send it to them over
- 1:38:14ClickUp or email, they'd have to
- 1:38:15download it, open it up. If they wanted
- 1:38:17to make a change, they'd have to make a
- 1:38:18change and then send it back to me. So,
- 1:38:20I just like to work with live cloud
- 1:38:22data, you know, cloud deliverables like
- 1:38:23Google Sheets, Google Docs. I really
- 1:38:25just like the Google ecosystem,
- 1:38:26honestly. And it's pretty cool because
- 1:38:28you can obviously just come into your
- 1:38:29codecs, you can go up here to the
- 1:38:31plugins, and you can just connect the
- 1:38:33Google Drive plugin, and it lets you do
- 1:38:34pretty much everything through Drive,
- 1:38:36Docs, Sheets, or Slides. And I know this
- 1:38:38is starting to sound like a Google ad,
- 1:38:39but trust me, it's not. But here is
- 1:38:41another example of an HTML skill. And
- 1:38:44this is what I call an HTML explainer.
- 1:38:46So whether this is a deliverable or
- 1:38:47whether this is just something that I
- 1:38:48want, whenever I want to turn a concept
- 1:38:50into something that's a bit more visual
- 1:38:52and in a way that explains it to me or
- 1:38:53to my students or to my team, I use this
- 1:38:56HTML explainer skill. And this one isn't
- 1:38:58as branded, but you can still see that
- 1:38:59in the top left, we have the logo. And
- 1:39:01you can still see that it kind of fits
- 1:39:02the color scheme a little bit. But the
- 1:39:04whole point of this one is that this
- 1:39:05creates HTML. So it's explaining some
- 1:39:07sort of concept to me, and it does it in
- 1:39:09this way where it's very visual. It's
- 1:39:11not very wordy. It's very easy to
- 1:39:12understand and it just lays it out as if
- 1:39:14it was trying to teach me a lesson. And
- 1:39:16it, like I said, it uses visualizations
- 1:39:18really well because that's the way that
- 1:39:20I like to learn. I like to actually see
- 1:39:21examples and I like to see flowcharts
- 1:39:23and little symbols and things like that.
- 1:39:25So, this is an HTML explainer skill. So,
- 1:39:27whenever I need something explained to
- 1:39:29me or I want to explain something to my
- 1:39:30team, I just go ahead and use this
- 1:39:32skill. And this also created a different
- 1:39:33version for us. One with like a
- 1:39:34transparent background and then one like
- 1:39:36this. So, we could send that around and
- 1:39:38like I said, make some tweaks. And now
- 1:39:40you have some brand guidelines. And that
- 1:39:42was just a visual brand guideline doc.
- 1:39:44This is something else that you can
- 1:39:45start to work in is like how does your
- 1:39:46company actually talk? Do you guys have
- 1:39:48slogans? Do you guys have mods? Do you
- 1:39:50guys have big hero stats that you always
- 1:39:52want your whole team to know when
- 1:39:54they're writing emails or deliverables?
- 1:39:56For us, we have this thing called an AI
- 1:39:58phrase kill list, which is things that I
- 1:39:59hate that AI says a lot. And I've got a
- 1:40:01ton of things on here and this is shared
- 1:40:03with my team so that then when they're
- 1:40:05generating emails or content or you know
- 1:40:07copy I don't want them to have their
- 1:40:09output sounding like AI. So this is one
- 1:40:12of the other things that we work into
- 1:40:13our like team brand guidelines not just
- 1:40:15from a visual perspective but also from
- 1:40:16like a what does our voice sound like?
- 1:40:19Who is our avatar we're speaking to? All
- 1:40:20of those things that you want everyone
- 1:40:21on the team to have sort of like a
- 1:40:23unified view on. That's also stuff that
- 1:40:25you can put into your AIOS. You can turn
- 1:40:27into skills and then you can share with
- 1:40:28your team. So, I would encourage you to
- 1:40:30think about stuff like that. How do you
- 1:40:31know that something came from your
- 1:40:33brand? And more importantly, what kind
- 1:40:34of stuff do you want to make sure
- 1:40:35doesn't come from your brand? So, for
- 1:40:37me, that was things like m dashes and
- 1:40:39things like saying, you know, and
- 1:40:41honestly, and honestly, it's not just X,
- 1:40:43it's Y, not a tool, not a feature, a
- 1:40:45revolution. And speaking of visual
- 1:40:47elements of a brand, we know that codecs
- 1:40:50can make us images because chatgbt image
- 1:40:522.5 or whatever the image model is from
- 1:40:54OpenAI by the time you're watching this,
- 1:40:56it's inside of codec. So if I ever say,
- 1:40:58"Hey, make me a picture, make me a
- 1:41:00thumbnail," blah blah blah. It can do it
- 1:41:01from here, which is amazing. And I
- 1:41:03really think OpenAI's image model is the
- 1:41:05best right now. But what it can't do
- 1:41:06right now is generate video or it can't
- 1:41:09use other AI models. If you wanted to do
- 1:41:10a different, you know, maybe Gemini's
- 1:41:12model is really good, nano banana, or
- 1:41:13maybe you want to do like a sea dance
- 1:41:14video or a cling video. Well, let me
- 1:41:17talk about next how can you connect
- 1:41:19codeex to a different tool that will let
- 1:41:21you generate images and videos with tons
- 1:41:23of different models. So let's jump into
- 1:41:25the next segment. One of the great
- 1:41:26things about Codex is that right inside
- 1:41:28of it, it can make images for you. As
- 1:41:29you can see right here, I gave it these
- 1:41:31three images and I said, "Hey, I need a
- 1:41:32YouTube thumbnail." I described what I
- 1:41:33want and then I got it right here. But
- 1:41:35if I asked it to turn this into an
- 1:41:37animated video, it wouldn't be able to
- 1:41:38do it with its native capabilities. It
- 1:41:40would have to find some other form. It
- 1:41:42would have to build code. Like here, it
- 1:41:43wants to use hyperframes. So, typically
- 1:41:44what you'd do is you'd connect it to
- 1:41:46something like Higsfield that has a
- 1:41:47bunch of video and image models like
- 1:41:49Cance 2.5 and all these other ones like
- 1:41:52Cling and Omni Flash and Miniax. The
- 1:41:55problem with Higsfield was that it was
- 1:41:56way more expensive because you had to
- 1:41:58get on a subscription. If we look at the
- 1:41:59monthly, if you wanted to use Seedance
- 1:42:012.5, you had to be on a 60 bucks a month
- 1:42:03plan and you still had a limited number
- 1:42:05of credits. But Higfield just actually
- 1:42:07dropped an API which means that you can
- 1:42:09actually just pay per usage. Meaning you
- 1:42:11can connect Codeex to Higsfield API and
- 1:42:14then you're not actually paying every
- 1:42:15single month. You're only paying for the
- 1:42:17times that you actually make requests.
- 1:42:19So today I'm just going to show you real
- 1:42:20quick how easy this is to set up. So
- 1:42:21let's get right into it. So this is a
- 1:42:23completely separate account. What you're
- 1:42:24going to do is you're going to go to
- 1:42:25Google and you're going to type in
- 1:42:26Higsfield API. And then you're pretty
- 1:42:29much going to click on this first link
- 1:42:30right here. And it's going to take you
- 1:42:31to this API dashboard where you'll be
- 1:42:32able to create your own key and create
- 1:42:34your own account. And when you sign up,
- 1:42:36you're going to get a ton of discounts
- 1:42:37because they just dropped this, which is
- 1:42:3915% off sale models, up to 50% off your
- 1:42:42favorites, and you get 15 bucks right
- 1:42:43away in your account to play with. So
- 1:42:45create an account, and then when you're
- 1:42:46inside of that, it's going to look like
- 1:42:47this. What you're going to want to do is
- 1:42:48you're going to go to either the quick
- 1:42:50start on this right hand side and you
- 1:42:52can copy this prompt if you want, but
- 1:42:53you don't have to. It's it's a lot
- 1:42:55easier than that. But you do have to
- 1:42:56make an API key. So you'll click on
- 1:42:58create API key. You'll type in a name.
- 1:42:59So I'll just call this one demo and
- 1:43:01you'll create your key. And then all you
- 1:43:02have to do is copy this key right here
- 1:43:04and then hop into CEX and I'll show you
- 1:43:06what to do next. Real quick before we
- 1:43:07hop into CEX, I just wanted to show you
- 1:43:09guys how to make sure you take full
- 1:43:10advantage of this launch because this is
- 1:43:13a 7-day limited offer. So, these things
- 1:43:15up here that we see, 15% off all sale
- 1:43:17models. That is just when you sign up,
- 1:43:19you'll already have that discount. In
- 1:43:21order to get up to 50% off your picks,
- 1:43:23you have to come into the pricing
- 1:43:24section and it says choose three models
- 1:43:26with max discount up to 50% off. And
- 1:43:28then you would just come in here and
- 1:43:29choose two video models and one image
- 1:43:30model. And you can see all of the prices
- 1:43:32over here. As you can see, with 30% off,
- 1:43:34we would be getting it for this price
- 1:43:36compared to the original current price.
- 1:43:38And then the last one is getting 15 free
- 1:43:40dollars to play with on your account.
- 1:43:42All you have to do here is connect a
- 1:43:43credit card and use an eligible business
- 1:43:46domain email. And then once you've
- 1:43:48claimed all of that stuff, you now have
- 1:43:49that API key. You now have everything
- 1:43:51set up and you can run 30 generations at
- 1:43:53a time. So let's hop back into codeex
- 1:43:54and I can show you how to get this all
- 1:43:56set up. So you're going to open up a new
- 1:43:57chat and you're just going to say, "Hey,
- 1:43:59inside of myv file, can you just create
- 1:44:02me a placeholder for Higsfield API? I'm
- 1:44:04going to drop you in a Higsfield API
- 1:44:06key." And then you can just figure out
- 1:44:07how that all works so we can start
- 1:44:09running some image and video
- 1:44:10generations. and then just shoot that
- 1:44:12off and it will tell you exactly where
- 1:44:13to paste in your key. You don't want to
- 1:44:15paste it right in here into the chat
- 1:44:16thread. You want to paste it into thev
- 1:44:18file. So you can see I already have that
- 1:44:19in there. So it says, "Hey, that already
- 1:44:21exists." But if it doesn't, it'll give
- 1:44:22you a placeholder. And then all you have
- 1:44:23to do is ask for the file path that you
- 1:44:25can actually open this up. So I could
- 1:44:26rightclick this, go to open with, and I
- 1:44:28can just open this with whatever I want.
- 1:44:30I'll choose default app. That opens up
- 1:44:31my notes where I can go ahead and see
- 1:44:33right here it added Higsfield API key.
- 1:44:35And then you will just go ahead and
- 1:44:36paste in your API key right there. And
- 1:44:38then make sure you save that before you
- 1:44:39close out of it. and then just go,
- 1:44:41"Okay, cool. I put in that API key. Can
- 1:44:43you see what models are available for me
- 1:44:45through that Higsfield API?" And then
- 1:44:47just go ahead and shoot off a test video
- 1:44:49generation where there is a man sitting
- 1:44:51at a laptop using Higsfield and using
- 1:44:54codecs to generate videos. But you guys
- 1:44:57won't have to worry about that at all
- 1:44:58because I have this skill right here
- 1:44:59called Higsfield API generations. And it
- 1:45:01has all these defaults in here. It has
- 1:45:03all of the settings. It has all of the
- 1:45:05different things that you might need or
- 1:45:06you would have your agent have to
- 1:45:08normally go out and do research on.
- 1:45:09Everything's right here. So basically
- 1:45:10all you have to do is you've put in your
- 1:45:12Higsfield API key into your EMV. And now
- 1:45:14you can see it's basically going to use
- 1:45:15that whenever you need an image or a
- 1:45:16video and it knows exactly how to work
- 1:45:18with Higsfield API. It knows about all
- 1:45:20of the different functionality and
- 1:45:22options that it has in there. And now
- 1:45:24it's just generating me a bunch of
- 1:45:25different images and videos that you can
- 1:45:26see right here on screen with just using
- 1:45:29natural language. So pretty cool stuff.
- 1:45:30If you want to grab this skill, it's
- 1:45:31completely free in my free school
- 1:45:32community. The link for that is down in
- 1:45:34the description. Just head into the free
- 1:45:35school community. You're going to go to
- 1:45:36classroom and click on all YouTube
- 1:45:38resources and there will be a huge
- 1:45:39database with a bunch of resources. You
- 1:45:41should be able to find it right in
- 1:45:42there. And that skill is also going to
- 1:45:43tell you exactly how much each of your
- 1:45:45generations cost you so that you don't
- 1:45:47have to be constantly going back into
- 1:45:48your Higsfield dashboard to look at the
- 1:45:49analytics. You will just get that right
- 1:45:51there in your chat thread. So, while
- 1:45:53that's running, I'll just go over here
- 1:45:54and show you real quick what actually
- 1:45:55this stuff can look like. So, we have
- 1:45:57Higsfield API benchmark out of all of
- 1:46:00these videos that I generated. So, I'll
- 1:46:02show you each one like how much it
- 1:46:04costed or sorry, cost. My goodness, how
- 1:46:07dare I. You can see that the total spend
- 1:46:08for these was $7.40 and we were using
- 1:46:11models like Cance 2.5. And once again,
- 1:46:14this would have been only available to
- 1:46:15us on a 60 bucks a month plan. And then
- 1:46:18you're just sitting there paying 60
- 1:46:19bucks a month even if you don't generate
- 1:46:20anything that month. So, anyways, we
- 1:46:22have some video outputs here. You can
- 1:46:24see this is Cance 2.5. This one has no
- 1:46:26audio, but you can generate these with
- 1:46:27audio as well. So, that's a super super
- 1:46:29clean shot. Same exact prompt went off
- 1:46:31to cling 3.0. And this one has I don't
- 1:46:34know those letters look a little bit
- 1:46:35different compared to that image, but
- 1:46:37they have similar sort of like
- 1:46:38background. The letters in this one are
- 1:46:40also off. So, it's really good to get in
- 1:46:42here and shoot off the same prompt to a
- 1:46:44bunch of different AI models and see
- 1:46:46which one comes back best. In this case,
- 1:46:48Nova came through only as Nova in
- 1:46:50Cedance 2.5. So, in this case, it was
- 1:46:52the best. But, of course, with Cedance
- 1:46:542.5, you are going to be paying a little
- 1:46:55bit more, but once again, it's only pay
- 1:46:57per usage. So here, this one was 1.6
- 1:47:00bucks. This one was only about 20 cents,
- 1:47:02and this one was only about 50. But they
- 1:47:04also can generate sound. So let me play
- 1:47:06you guys a few examples of some UGC
- 1:47:08content. And we'll take a look at the
- 1:47:09cost as well.
- 1:47:11>> I used to skip breakfast. Now I grab
- 1:47:14Nova protein coffee. 20 g of protein,
- 1:47:18real coffee, and I'm out the door.
- 1:47:21>> So this was Cedance 2.5, and it was
- 1:47:23$3.23.
- 1:47:25Here is Cling 3.0.
- 1:47:27I used to skip breakfast. Now I grab
- 1:47:30Nova protein coffee. 20 gram of protein,
- 1:47:34real coffee, and I'm out the door.
- 1:47:37>> And as far as realism, I think Cling 3.0
- 1:47:39did a really, really good job here with
- 1:47:41this character, you know, selfie style
- 1:47:43UGC for only 60s, whereas this one you
- 1:47:46wouldn't say. I don't know. It's like
- 1:47:47it's not noticeably better as far as
- 1:47:48like makes a huge difference. But in
- 1:47:51general, I mean, Cance 2.5 is in my
- 1:47:53opinion the leading video generation
- 1:47:54model right here. But we also have Mini
- 1:47:56Max and this one was a $110.
- 1:47:59>> I used to skip breakfast. Now I grab
- 1:48:01Nova protein coffee. 20 grams of
- 1:48:04protein. Real coffee.
- 1:48:06>> Anyways, I think that this one was
- 1:48:08clearly the worst out of just these
- 1:48:10three very, very simple videos. But it's
- 1:48:12important, like I said, shoot off the
- 1:48:13same prompts at different models. See
- 1:48:14how the cost differ. See how the quality
- 1:48:16differs. See what they're good at.
- 1:48:17because Cance 2.5 might be the best at
- 1:48:20like people and product shots, but maybe
- 1:48:22Miniaax or Cling is better at different
- 1:48:24sort of like cartoon animations or
- 1:48:25something like that. And then of course
- 1:48:26you can use Codex image generation. GBT
- 1:48:28image 2.5 is really really good. It's
- 1:48:30probably my favorite, but there are also
- 1:48:32other image models you can try out like
- 1:48:34Higfield Marketing Studio here or Grom
- 1:48:37Imagine 2.0, Soul 2.0, and these are
- 1:48:40obviously much cheaper than generating
- 1:48:41videos. But I did take this exact same
- 1:48:43prompt and put it into codecs. And this
- 1:48:45is what we got here. So if you like this
- 1:48:46better then maybe you want to just stick
- 1:48:48with GBD2.5 and then you also have the
- 1:48:50benefit of this just counting towards
- 1:48:52your codec subscription or your chatbt
- 1:48:55subscription rather than paying
- 1:48:56separately per usage for that. Anyways,
- 1:48:58hopping back here into codeex. What's
- 1:49:00cool about this is Codex will go ahead
- 1:49:02and do all the research. So you don't
- 1:49:03have to think about the API endpoints.
- 1:49:04You don't have to give it anything else.
- 1:49:06You just tell it to figure it all out
- 1:49:07and it will. And so that's why back in
- 1:49:09here you can copy this prompt if you
- 1:49:10want or you can just say like figure it
- 1:49:12out because you can see this prompt is
- 1:49:13literally pointing it to the actual
- 1:49:15Higsfield API documentation. You can see
- 1:49:18it says your key authenticated
- 1:49:19successfully. The live catalog returned
- 1:49:2176 endpoints. 18 are text to video, 44
- 1:49:24are imagetovideo, 11 are text to image
- 1:49:26and plus there's a few other routes. And
- 1:49:28it's going to use Cance 2.0 for this
- 1:49:29first run. The request was accepted with
- 1:49:31this ID and now it's basically just
- 1:49:33going to keep pulling on that request
- 1:49:35until the generation is done. And
- 1:49:37polling just means it's going to check
- 1:49:38in and then if it's not done, it's going
- 1:49:39to wait and it's going to check in
- 1:49:40again. If it's not done, it's going to
- 1:49:41wait and it's going to keep pulling or
- 1:49:43checking until that video is actually
- 1:49:45done. And then it will download the MP4
- 1:49:47locally. And so, while this is finishing
- 1:49:48up, let me just show you what else is
- 1:49:49available inside of the Higfield API. If
- 1:49:52you go to pricing right now, it says
- 1:49:54that you can get up to 50% off. And this
- 1:49:55is only for the next 7 days. So, if
- 1:49:57you're watching this early, go here and
- 1:49:59you can choose two video and one image
- 1:50:01models to get 50% off on. So, you'll
- 1:50:04literally just come in here and it will
- 1:50:05say, "Hey, what are your models you want
- 1:50:06to choose?" I personally chose Cance 2.5
- 1:50:09and Clink 3.0. And then for the image
- 1:50:11model, I chose the marketing studio
- 1:50:12image. But then you can just go ahead
- 1:50:14and look at all of the models in here.
- 1:50:15And you can see the different modes that
- 1:50:16they run in like standard or turbo or,
- 1:50:19you know, 4K, whatever it is. And you
- 1:50:21can see all of the information about all
- 1:50:22these different types of models and
- 1:50:24their pricing. And then the analytics
- 1:50:25will basically just show you how much
- 1:50:26you've spent. Um what is your request
- 1:50:29looking like? What are the models you
- 1:50:31like to use and things like that. So
- 1:50:32just a pretty little analytics dashboard
- 1:50:34for you. But you can see we have this
- 1:50:36video back. Now, this is already looking
- 1:50:37a little bit weird with Cance 2.0
- 1:50:39because the laptop is facing the wrong
- 1:50:41way, but let's give this a quick play.
- 1:50:48Now, I'm really not quite sure if
- 1:50:49anybody uses a laptop like that, but
- 1:50:50either way, this was Cance 2.0 and there
- 1:50:52is a better model out there now. But
- 1:50:54what's important to realize is that the
- 1:50:55model isn't just going to run perfectly
- 1:50:56for you. You have to be able to prompt
- 1:50:58it in the right way and you will start
- 1:51:00to build skills around how you like to
- 1:51:01prompt image generations and video
- 1:51:03generations. And that's where this gets
- 1:51:05really powerful is when you work in all
- 1:51:06of these different skills and you can
- 1:51:08add things too like your brand
- 1:51:09guidelines or a consistent character
- 1:51:11that you want to have and then over time
- 1:51:13your ability to just inside of codecs
- 1:51:15generate these different videos and
- 1:51:16image assets is going to get better and
- 1:51:19better. Now I did want to end off this
- 1:51:21video talking about is this actually
- 1:51:22cheaper? And the answer is yes and no.
- 1:51:25It all depends on the break even point
- 1:51:26because now that you're being charged
- 1:51:28per usage rather than based on your
- 1:51:29credits that reset every month, there's
- 1:51:31a break even point. And if you're paying
- 1:51:33per usage to the point where you're
- 1:51:35going over that break even point, then
- 1:51:37this is probably where you'd want to be
- 1:51:38on a subscription. But if you're only
- 1:51:40infrequently generating videos and
- 1:51:41images on a monthly basis and you're
- 1:51:43below that break even point, then it's
- 1:51:46much cheaper obviously to be on the pay
- 1:51:48per usage billing, especially when
- 1:51:50you're trying to figure out like what is
- 1:51:51your cadence, then starting with pay per
- 1:51:53usage is a lot safer than just
- 1:51:55committing to a monthly subscription.
- 1:51:57And so I've run the math here and on the
- 1:51:58plus and ultra plans within Hicksfield,
- 1:52:02this is what it would kind of look like
- 1:52:03as far as those break even points. So if
- 1:52:05you're on the plus monthly, it's about
- 1:52:07fewer than 19 clips. And if you're on
- 1:52:09the plus annual, then it's roughly 16
- 1:52:11clips. And if you're below that, then
- 1:52:12the API makes more sense. If you're on
- 1:52:14ultra, it looks like 50 on monthly or 39
- 1:52:17on annual. So ultimately, these are the
- 1:52:20three kind of questions I wanted to
- 1:52:21answer here. If you're maxing out your
- 1:52:22subscription plan every single month
- 1:52:24consistently, then you probably just
- 1:52:25want to stay on a subscription. If
- 1:52:27you're only generating things
- 1:52:28occasionally, like maybe some months
- 1:52:29you'll run, you know, 15 videos, some
- 1:52:31months you won't do any at all, then API
- 1:52:33will win. And if you need programmatic
- 1:52:35workflows or you want to bake this API
- 1:52:37into your own, I don't know, your own
- 1:52:39website or SAS product or your own
- 1:52:40internal tools or whatever it is, then
- 1:52:42the API is going to win there as well.
- 1:52:44So, it's always hard in the AI space or
- 1:52:46in the tech space to just come out of
- 1:52:47the gate and say, "Yes, this is
- 1:52:48objectively cheaper or yes, this is
- 1:52:49objectively better." It's always very
- 1:52:51situational. So hopefully this all kind
- 1:52:53of makes sense and you can kind of use
- 1:52:54this to figure out in your specific
- 1:52:56situation what is probably the right
- 1:52:58path for you. And what's really cool is
- 1:52:59when you get all that dialed in, you can
- 1:53:01do all this in bulk. So we've actually
- 1:53:02done things where we've generated assets
- 1:53:04for like our events or for you know ads
- 1:53:06or whatever it is and we basically just
- 1:53:08use codeex or claw to help us brainstorm
- 1:53:10all the briefs and all the angles and
- 1:53:11then we just shoot off a bunch of
- 1:53:13workers to generate all those assets and
- 1:53:14then you wake up and everything is done.
- 1:53:16Now, what's really cool is what you can
- 1:53:17actually do with Codeex as far as
- 1:53:19building websites when you actually have
- 1:53:21learned now how to generate other videos
- 1:53:23and other images to put on those
- 1:53:24websites. So, we're going to hop into
- 1:53:25this next section where I kind of show
- 1:53:26you guys what's possible with Astra and
- 1:53:28with Codeex when it comes to designing
- 1:53:30things like websites. We have a new AI
- 1:53:33design king and it's not even close.
- 1:53:35Take a look at some of these outputs
- 1:53:36I've been getting with Astra. Look how
- 1:53:37cool this scroll is. How about this
- 1:53:39example with GLO? I think that this is
- 1:53:41just absolutely incredible. I think Asha
- 1:53:43does a really good job at making things
- 1:53:443D, dynamic, immersive without making it
- 1:53:47overwhelming because this is still a
- 1:53:48very clean hero section. Now, this one
- 1:53:50for perk form is a little bit more out
- 1:53:51there, but it's very impressive what it
- 1:53:53was able to do, but I still think it's
- 1:53:54pretty clean. And we have other types of
- 1:53:56animations here. Look at these cans come
- 1:53:57in. I love how it was able to create
- 1:53:59these in 3D. Now, I just put out a video
- 1:54:01that I said Fable 5.1 was the design
- 1:54:03king because it helped me create
- 1:54:04something like this, which I loved. But
- 1:54:06Astra is so much better with it so far.
- 1:54:08Now, all of these examples that we just
- 1:54:10looked at and that we're going to keep
- 1:54:11looking at were oneshot prompts. Think
- 1:54:13about that. If I took the time to
- 1:54:14actually sit down and iterate on these,
- 1:54:17they would be so much better. This was
- 1:54:18Astra's first pass at an AI Automation
- 1:54:20Society website. You can see it tried to
- 1:54:21keep the same vibe. So, right now, the
- 1:54:23one that Fable built, I think, is the
- 1:54:25winner because I've iterated on this
- 1:54:27like 15 times, but as a first shot, I
- 1:54:29think that this is really, really good.
- 1:54:31It's on brand and like I said, it has
- 1:54:33these animations and these interesting,
- 1:54:35you know, like immersive parallax
- 1:54:37things, but it's just it's not too much.
- 1:54:39I've been seeing you guys talk about in
- 1:54:40the comments how, oh, I don't think
- 1:54:41these sites convert. And I think you're
- 1:54:43right, but it depends on the three Ps
- 1:54:45that I talk about, the pain, the person,
- 1:54:47the promise. But I love how subtle this
- 1:54:49is, but instantly I think that this
- 1:54:50feels premium. You feel the depth and
- 1:54:52it's not too overwhelming, and it still
- 1:54:53fits the brand that you know that you
- 1:54:56have here. It still fits the messaging
- 1:54:57that you're trying to get across. even
- 1:54:58something like a luxury watch brand. I
- 1:55:00think that even just this little element
- 1:55:02right here that lets us sort of scroll
- 1:55:03through really makes it, you know, it
- 1:55:06sets it apart from other types of
- 1:55:08websites in this industry because it's
- 1:55:10not just about the hero section. I think
- 1:55:11that the way that all of these sites are
- 1:55:13designed are very on brand. It's not
- 1:55:15like one super specific template that
- 1:55:16we're following every single time. It's
- 1:55:18different each time. So, I know that
- 1:55:20I've just been jumping all around with
- 1:55:22examples. I just wanted to show you guys
- 1:55:23some of the different things that it's
- 1:55:25been able to do and some of the
- 1:55:26different ways that it uses its creative
- 1:55:28freedom and taste because I basically
- 1:55:30just told it to make me a bunch of
- 1:55:31websites. Now, the one thing that all of
- 1:55:33these that I just showed you have in
- 1:55:34common is that they used my scroll craft
- 1:55:36skill, which if you guys don't have
- 1:55:37that, join my free school community
- 1:55:39linked in the description. That's where
- 1:55:40you can get the skill, and I'm
- 1:55:41constantly updating that thing as I find
- 1:55:43out more about AI design. Not only does
- 1:55:45that skill understand things like
- 1:55:46layering, as you can see here, and
- 1:55:47scrolling, but it also understands
- 1:55:49things like typography and spacing and
- 1:55:52all of those other elements that you
- 1:55:54need to make the website actually feel
- 1:55:55good as you continue to scroll through.
- 1:55:57But I think that's been one of the
- 1:55:58biggest things that I've realized lately
- 1:55:59is the layering adds so much depth
- 1:56:02because you can clearly see how the text
- 1:56:03is on a different layer than the
- 1:56:05background image, which is on a
- 1:56:06different layer than the bike, which is
- 1:56:07on a different layer than the rocks
- 1:56:09right here. And that's what makes it
- 1:56:10feel super immersive without being over
- 1:56:12the top, at least in my opinion. because
- 1:56:13at the end of the day, all of this stuff
- 1:56:15is subjective and it's all about taste.
- 1:56:16But like I said, my scrollcraft skill
- 1:56:18has all that knowledge in there. And
- 1:56:19what you've noticed about all of these
- 1:56:20different websites I've shown you so far
- 1:56:22is that they're all different types of
- 1:56:23businesses. We had a content agency, we
- 1:56:25had physical products, this one is a
- 1:56:27consulting firm, and all of these have
- 1:56:29different vibes. They have different
- 1:56:31ways that the sections are actually
- 1:56:32structured. This one has like a sidebar
- 1:56:34over here. And we haven't seen that yet
- 1:56:35on the other ones. Now, on top of
- 1:56:37website design, I'm also talking about
- 1:56:38things like motion graphics and design
- 1:56:40using something like hyperframes. So,
- 1:56:42let me play this example for you and pay
- 1:56:43attention to the 3D elements and also
- 1:56:45pay attention to the music and how it's
- 1:56:46synced to what's going on.
- 1:56:50[music]
- 1:56:57[music]
- 1:57:08>> [music]
- 1:57:20>> and I'm not going to play this whole
- 1:57:21thing, but it also created me a vertical
- 1:57:23version. So, I'll play a few seconds.
- 1:57:25[music]
- 1:57:33Now, those obviously aren't anything
- 1:57:34crazy. they're not going to go viral.
- 1:57:36But if you consider that that was just
- 1:57:37two prompts and took me, you know, 20 to
- 1:57:3930 minutes, how much time would that
- 1:57:41have taken and how much would that have
- 1:57:43cost you if you paid a human editor to
- 1:57:45do that? So much more time and money.
- 1:57:47And here's an example that I think is
- 1:57:48even crazier because what I had to do
- 1:57:50here is for AIS Live, we have this
- 1:57:52folder where we have like over 100
- 1:57:54gigabytes. You can see 152 GB of footage
- 1:57:57to look through. And all I did was give
- 1:57:58it the link and I said, "Hey, create me
- 1:58:00like a sizzle reel recap that's a minute
- 1:58:01long from this event." And look what I
- 1:58:03got. logged off yesterday just buzzing.
- 1:58:05I had goosebumps. I was so excited and I
- 1:58:07had so much fun. [music]
- 1:58:15[music]
- 1:58:20[music]
- 1:58:26Cost doesn't justify price. Price
- 1:58:28justifies cost. [music]
- 1:58:34What is the value of what you know
- 1:58:37[music]
- 1:58:37and what you know how to do?
- 1:58:43This culture that we have here at AIS is
- 1:58:46is just really special.
- 1:58:53[music]
- 1:59:02So, the only thing wrong here is this
- 1:59:03domain. That's not the real domain. But
- 1:59:05I am just so impressed by this. And I
- 1:59:07know some of you guys might look at that
- 1:59:08and be like, uh, I saw some editing
- 1:59:09mistakes or eh, whatever. But the music,
- 1:59:13what it had to think through all of
- 1:59:14that. It had to look through all that
- 1:59:16footage and basically figure out how to
- 1:59:17tell a story out of it. And I thought it
- 1:59:18did a really good job at that. It also
- 1:59:20had to look through all of our logos and
- 1:59:22our designs and our branding and our
- 1:59:24images and our assets for that. It found
- 1:59:26everything and it pulled it all in. And
- 1:59:28I was just like honestly for 35 minutes
- 1:59:30is how long it took. So so impressed
- 1:59:32because I've tried to do something like
- 1:59:34that with Soul. I've tried to do
- 1:59:35something like that with Fable 5.1 and
- 1:59:36it did not go nearly as well. this thing
- 1:59:39just has the understanding and it's so
- 1:59:41pragmatic and logical about it that I am
- 1:59:43just like seriously it's it's it's
- 1:59:45another huge leap forward to me now when
- 1:59:47we talk about actually killing AI design
- 1:59:49slop or having things that really go up
- 1:59:50to the next tier of course it's using
- 1:59:52things like skills that you might find
- 1:59:53on GitHub like mine or like other
- 1:59:55people's but what I think is the most
- 1:59:57important thing is inspiration it's
- 1:59:59really hard to just give an AI model a
- 2:00:00task like make me a website here are the
- 2:00:02brand guidelines and have it just
- 2:00:03actually create you something that
- 2:00:04doesn't look very AI generated so what
- 2:00:06you have to do is you have to have
- 2:00:07inspiration and you can find that in the
- 2:00:08form of something like godly.design
- 2:00:10where you can find actual websites to
- 2:00:12take inspiration from or you can find
- 2:00:13individual components like the hero
- 2:00:15sections or the CTAs or the footers and
- 2:00:18you can find these components. You can
- 2:00:19do the exact same thing on something
- 2:00:20like 21st.dev. There are so many amazing
- 2:00:23things here. Like I said, hero sections,
- 2:00:25different components for the UI, you can
- 2:00:27find forms, AI chats, you can find
- 2:00:29buttons. Using all of these things and
- 2:00:31giving these code blocks or giving these
- 2:00:33website URLs to Astra makes them so much
- 2:00:36better because now I could come into any
- 2:00:37one of these 10 sites or the other ones
- 2:00:39that we had generated and just be like,
- 2:00:41you know what, I really like this, you
- 2:00:43know, um, button for example right here,
- 2:00:45but I would like this to be changed out
- 2:00:46with, you know, this button that I found
- 2:00:48on 21st. Because I feel like it's more
- 2:00:50on brand or I like the, you know,
- 2:00:51dynamic element around it. You can also
- 2:00:53come to something like awards with three
- 2:00:55W's, awards.com that has a ton of
- 2:00:57award-winning sites. And you can also
- 2:00:59come in here and organize by category,
- 2:01:01which I think is really important
- 2:01:02because an e-commerce site probably
- 2:01:03needs to look different than, you know,
- 2:01:05a restaurant and hotel site, which
- 2:01:06probably needs to look different than a
- 2:01:07fashion site. Because yes, there's one
- 2:01:09element of AI slop where it's like, you
- 2:01:11know, you've got those certain types of
- 2:01:12buttons and the certain type of fonts
- 2:01:13and the certain gradient in the
- 2:01:15background. But the other element of AI
- 2:01:16slop is when it feels like you're
- 2:01:19looking at something that is meant for
- 2:01:20someone else. If you're a 70-year-old
- 2:01:22man and you're looking for medication on
- 2:01:24your website and you're scrolling
- 2:01:25through a 3D world, you're going to be
- 2:01:27like, "This is the worst thing I've ever
- 2:01:28seen." But if you're a teenage dude and
- 2:01:29you're trying to buy new headphones and
- 2:01:31the headphones on the site are spinning
- 2:01:32and you can see inside of it and it's
- 2:01:34like transforming. You're going to think
- 2:01:35that's sick. And that's just the truth.
- 2:01:36So think about the pain, think about the
- 2:01:38person, think about the promise, and
- 2:01:40think about inspiration sites. And then
- 2:01:42you just get out of Astra's way. You
- 2:01:43give it your brand guidelines, you give
- 2:01:45it the copy, you give it scrollcraft
- 2:01:47skills so you can create all of this
- 2:01:48sorts of cool stuff. And then you watch
- 2:01:51the site or you scroll through it. You
- 2:01:52say, "Uh, I don't like this. Fix that. I
- 2:01:55don't like this. Fix that." And then you
- 2:01:56just keep iterating. And that's how it's
- 2:01:58done. So CEX is probably the best
- 2:02:00browser use agent I've ever played with
- 2:02:03ever. And browser use is super powerful
- 2:02:05for many things, which is what I'm about
- 2:02:06to get into the next video about. But
- 2:02:09basically, it just can control your
- 2:02:11computer. Not only can it control your
- 2:02:12computer, but it can also control
- 2:02:13browser tabs and it can save sessions
- 2:02:15and things like that. So, if you can't
- 2:02:17use an API for some reason, maybe the
- 2:02:19website you want to use or the service
- 2:02:21you want to use doesn't have one, you
- 2:02:22can use browser use. And like I said,
- 2:02:24Codeex is just the best at it, at least
- 2:02:26right now. So, I'm going to hop into the
- 2:02:28next video with you guys about how you
- 2:02:29actually can take advantage of Codex's
- 2:02:30browser use. Codex's browser use is
- 2:02:32probably the best that I've ever tried
- 2:02:34before. It can literally do anything on
- 2:02:35a browser or on your actual local
- 2:02:38computer. And if you need to be signed
- 2:02:39in for something, you just sign in once
- 2:02:41and Codex will save it. So, the
- 2:02:42possibilities are endless. You can
- 2:02:43actually automate anything. Now, let me
- 2:02:45show you guys how to set it up, how easy
- 2:02:46it is, and a few good use cases. All
- 2:02:48right, so I'm using the Codeex desktop
- 2:02:50app, which is just honestly the main way
- 2:02:51that I like to drive Codeex. It's super
- 2:02:53super slick, and we get the browser to
- 2:02:54pop up on the right hand side, which
- 2:02:56I'll show you guys in a sec. So,
- 2:02:57luckily, we live in a world now where
- 2:02:59most tools that we need to use have an
- 2:03:01API. So, it's much easier to just
- 2:03:02connect codecs through the API, or if we
- 2:03:05just go to the plugins, there's hundreds
- 2:03:06of plugins in here. But every once in a
- 2:03:08while, we don't have an API, and you can
- 2:03:09do that with browser use. But before I
- 2:03:11show you an automation like that, let's
- 2:03:12just take a look at this example where
- 2:03:14you might be building a website. You
- 2:03:15might be building some sort of app and
- 2:03:17you need a browser, an agent to actually
- 2:03:19test through things. If you guys watched
- 2:03:21a recent video I did on like building an
- 2:03:23AI SAS in one day, then this was
- 2:03:25actually from that demo where I had a
- 2:03:27browser agent get spun up and test the
- 2:03:29heck out of our app. And it did 85
- 2:03:31focused automation checks with real
- 2:03:32browser workflow testing, which means it
- 2:03:34was clicking around, it was typing
- 2:03:36things, it was hitting buttons, and it
- 2:03:37was trying to break the app. And it
- 2:03:38found a ton of bugs. and then we were
- 2:03:40able to just fix them. So, in this
- 2:03:41example, I have a super simple website
- 2:03:43spun up here on the local host. It's a
- 2:03:44form submission as you can see, and
- 2:03:46there's a few issues before we actually
- 2:03:47go ahead and start testing this thing.
- 2:03:50You can see like this spacing is weird.
- 2:03:52This thing is overlapping. And so, what
- 2:03:54you can do is you can click on this
- 2:03:54annotate button, and that lets you
- 2:03:56choose very specific elements to
- 2:03:58actually change. So, I can click right
- 2:04:00here on this phone number box, and I
- 2:04:01could describe, hey, this is kind of
- 2:04:03like overlapping. It's out of bounds.
- 2:04:05It's covering up the company box. So, I
- 2:04:07need you to fix this. And then when you
- 2:04:08shoot that off, it actually goes over
- 2:04:09here into your chat. And then you can
- 2:04:11make another annotation like, okay, so
- 2:04:12this whole box isn't really aligned on
- 2:04:14the same y-axis as number one and number
- 2:04:17three. So you need to fix that as well.
- 2:04:18And then those annotations are saved
- 2:04:19over here and we can just go ahead and
- 2:04:21shoot them off. So that's kind of like a
- 2:04:22lovable style or like a claw design
- 2:04:24style thing that you can do with the
- 2:04:25inapp browser in Codeex. What else is
- 2:04:28really cool about this is that you can
- 2:04:29actually just drive this thing. So you
- 2:04:30can go to Google, you can search for
- 2:04:32Grand Canyon images, you can go to all,
- 2:04:34you can go to different links if you
- 2:04:35want. You can use the browser right here
- 2:04:37while you have Codex open. But of
- 2:04:39course, Codex can also just take control
- 2:04:42of it. Now, we can see right here is my
- 2:04:44mouse, but also this is what Codeex's
- 2:04:46mouse looks like. And that's how you
- 2:04:47know it's now sort of getting ready to
- 2:04:49take control of the browser. Okay, so
- 2:04:51now that those changes have been made,
- 2:04:52let's actually start testing out our UI.
- 2:04:54I want to see things that maybe me as a
- 2:04:56human, I would maybe go through here and
- 2:04:58test like five or six edge cases, but we
- 2:05:00can see how Codeex can do it way quicker
- 2:05:02and way more. Okay, so you have the UI
- 2:05:04open. I want to make sure that there are
- 2:05:06no bugs. I want to make sure that this
- 2:05:07thing can't break. So, what I want you
- 2:05:09to do is use your browser use and just
- 2:05:11test the heck out of this thing. Try
- 2:05:12different things. Try your best to break
- 2:05:14it and let me know what bugs you find
- 2:05:15and what things we need to fix. And so,
- 2:05:17by the way, while this is running,
- 2:05:18there's a difference between browser use
- 2:05:20being headless or headed. Headless means
- 2:05:22it's running in the background. So, you
- 2:05:23can keep working in the foreground and
- 2:05:25you won't actually see it going on, but
- 2:05:26it can run a lot of things in the
- 2:05:28background. But, if you want it to be
- 2:05:29headed, that's basically where you're
- 2:05:30able to see it. So, we're probably going
- 2:05:31to be able to see some things right
- 2:05:32here. But if you wanted this to just
- 2:05:34work for a few hours for you headless in
- 2:05:36the background, you could ask it to do
- 2:05:37so and then it wouldn't disturb whatever
- 2:05:39work you're doing. So if you ever hear
- 2:05:40that terminology, that's what it means.
- 2:05:41So there you go. You can see that the
- 2:05:42mouse is moving around right now. It
- 2:05:44tried to click continue and then it says
- 2:05:46first name is required. Enter a valid
- 2:05:48email address and enter a 10-digit phone
- 2:05:50number.
- 2:05:52And now you can see it just entered in
- 2:05:53some information and it was able to hit
- 2:05:54continue. And it's just going to keep
- 2:05:56going through this QA check. We're
- 2:05:57validating that some of these fields are
- 2:05:59required options and that the continue
- 2:06:00button won't actually let it move on
- 2:06:02without filling those in. And by the
- 2:06:04way, the reason you're seeing some of
- 2:06:05these UI bugs is because I asked this
- 2:06:08chat to like make there be a few UI
- 2:06:10things to fix because I wanted to show
- 2:06:11you guys the annotations. So just ignore
- 2:06:13that. Mainly what I'm trying to show you
- 2:06:14guys now is the actual browser use QA
- 2:06:17function. So here you can see as it's
- 2:06:18clicking around, as it's trying to
- 2:06:19navigate this UI, it already said the
- 2:06:22core click path mostly works, but I've
- 2:06:24confirmed two real data integrity
- 2:06:25problems already, which is that
- 2:06:27obviously invalid contact data does get
- 2:06:29through and changing the country code to
- 2:06:31UK resets to US when you return to edit.
- 2:06:33I'm now testing keyboard submission and
- 2:06:35restart behavior, which often exposes
- 2:06:37bypasses the click path misses. So think
- 2:06:39about it like this. No matter what you
- 2:06:41get, you're going to want to test it.
- 2:06:42But would you rather have something that
- 2:06:44has been not tested at all, or would you
- 2:06:46rather have something that has been
- 2:06:47tested 100 times already by an AI agent?
- 2:06:50Especially if you tell the agent to try
- 2:06:51to break it, it's going to find some
- 2:06:53things that need to be fixed. No matter
- 2:06:54how good you are at trying to predict
- 2:06:56edge cases, when you push an app into
- 2:06:58production and users start using it,
- 2:07:00people are weird. People are
- 2:07:01unpredictable, so they're going to do
- 2:07:02things. And so this agent is able to
- 2:07:04hopefully simulate weird humans and just
- 2:07:06find things that you might not have
- 2:07:08thought of. You can see here, it even
- 2:07:09switched this to mobile view to still
- 2:07:10make sure that everything's working. And
- 2:07:12look at this. In mobile, it found a few
- 2:07:13failures in responsiveness. So that
- 2:07:15isn't something that I yet had tested.
- 2:07:17Okay, so this thing finished up now. And
- 2:07:19you can see that we have a bunch of
- 2:07:20things that passed, but a bunch of
- 2:07:22things that also failed. So it was able
- 2:07:24to find all of these bugs because it was
- 2:07:25testing the UI. And it didn't even do it
- 2:07:27that long. If we wanted to make this
- 2:07:28even harder, we could do a /go goal and
- 2:07:30we could say, "Hey, don't stop until
- 2:07:32you've tried a 100 different unique
- 2:07:33scenarios of, you know, UI edge cases
- 2:07:35and then come back to me." But it still
- 2:07:37found a lot of things that were broken.
- 2:07:38And now all we'd have to do is say,
- 2:07:39"Okay, cool. go implement those changes,
- 2:07:41go fix all those bugs. Okay, so let me
- 2:07:43show you guys another example. This is
- 2:07:44when you maybe have some sort of
- 2:07:46platform where you want to be able to
- 2:07:47like download statements or something,
- 2:07:49but they don't have an API. So you can
- 2:07:51actually interact with it with Codex's
- 2:07:53browser use because it basically uses
- 2:07:55vision and then AI to analyze like what
- 2:07:57it's looking at and where to click. And
- 2:07:58I would I just want to show you how easy
- 2:07:59it actually is to build these. All I
- 2:08:01said was, hey, you know, I want to build
- 2:08:02this skill blah blah blah, what would be
- 2:08:04helpful? And you could record a quick
- 2:08:06loom. You could send screenshots with
- 2:08:08arrows, but you actually don't have to.
- 2:08:09you can basically just describe pretty
- 2:08:11clearly what you need to do. So, what
- 2:08:13happened here is it opened up Relay. I
- 2:08:14signed in and it saved that session.
- 2:08:16Now, I said, "Okay, in the relay
- 2:08:17dashboard, you're signed in. What I need
- 2:08:19you to do is go to the dashboard, click
- 2:08:20on accounts, find the statements, um,
- 2:08:23click on these two accounts, download
- 2:08:24them as CSVs, and then put them into
- 2:08:26this folder." And it was able to do that
- 2:08:27on one shot. I didn't have to give it
- 2:08:29any feedback and it downloaded them and
- 2:08:31saved that whole flow as a skill. Now,
- 2:08:33this is super helpful because now I
- 2:08:35could at the end of the month just set
- 2:08:36this up as a scheduled routine because I
- 2:08:38could set a scheduled routine here and
- 2:08:40say, "Hey, use this skill on the most
- 2:08:42recent, you know, month statement in
- 2:08:44Relay." And for the most part, Codex is
- 2:08:46going to save your sessions when you're
- 2:08:48logged in. However, there are some
- 2:08:49limitations like the fact that Relay is
- 2:08:51going to pretty much boot you out
- 2:08:53because it's, you know, it's a banking
- 2:08:55platform. It's going to be secure and
- 2:08:57boot you out. So, look at this here. I
- 2:08:59asked it to do this again and it said,
- 2:09:00"Hey, relay is open at the login screen,
- 2:09:01but you have to sign in again." Now,
- 2:09:03what's really interesting that you can
- 2:09:04do is you can actually open up your
- 2:09:06settings. You can have a password
- 2:09:07manager inside of Codeex and you can see
- 2:09:09this goes to your browser and now you
- 2:09:11can go to import them so that Codeex can
- 2:09:14actually fill that out for you. And that
- 2:09:15way, you're not putting it into the chat
- 2:09:16history or anything like that. You're
- 2:09:18saving it more securely. So, all you
- 2:09:19have to do for that is create a CSV file
- 2:09:21that looks like this. I've got the name,
- 2:09:22the URL, the username, the password, and
- 2:09:24any notes, which in this case are empty.
- 2:09:26I imported that here right here as a
- 2:09:28file, a CSV file. And now we have
- 2:09:30relay.com password. Go ahead and try
- 2:09:32this again in the password manager for
- 2:09:35the inapp codeex browser. You should be
- 2:09:37able to log in now and then run the
- 2:09:39skill on June. All right. So over here
- 2:09:40now you can see the browser has taken
- 2:09:42control and it is going to put in my
- 2:09:44email and my password. Now if this asks
- 2:09:46me for a two-factor authentication code
- 2:09:48for my phone or something, then I'm
- 2:09:49obviously not going to be able to do
- 2:09:50that. I'd have to do that manually. But
- 2:09:51hey, if I was on a Mac and my text went
- 2:09:54there, and I used computer use to do
- 2:09:55that, technically, you would be able to
- 2:09:57automate that whole thing. But I would
- 2:09:59say you probably want to be careful
- 2:10:00about stuff like that. But as you can
- 2:10:01see, it was able to pull the passwords
- 2:10:03from my password manager, paste those
- 2:10:04in, and now it should be able to log in
- 2:10:06to relay for me. You can see here that
- 2:10:07the mouse has moved over to accounts.
- 2:10:10It's then going over to statements, and
- 2:10:12it's going to be able to download these
- 2:10:13two statements for us. So, you can see
- 2:10:14it's pretty cool that it was able to
- 2:10:16figure all this out. Remember, I've only
- 2:10:17ran this skill once, and all I did was
- 2:10:19instruct it with natural language to do
- 2:10:21this. Now, realistically, is that the
- 2:10:23way you should do it, especially with
- 2:10:24something like a bank account? No. What
- 2:10:26you should do is have it do the skill.
- 2:10:29You sit here and watch it and have it do
- 2:10:31that like at least 10 times where you're
- 2:10:33very closely watching it to make sure
- 2:10:35nothing goes wrong and you're making
- 2:10:36sure the script in the skill is super
- 2:10:39strict because at the end of the day,
- 2:10:40it's using vision, it's using AI, and
- 2:10:42that's how it's deciding what to click
- 2:10:43and what to do. So really the way that I
- 2:10:44think about it is when you want to
- 2:10:47automate something, your first call
- 2:10:48should always be to go for an API. Like
- 2:10:51that's what you want to do. It's the
- 2:10:52fastest. It's usually the cheapest. It's
- 2:10:54usually just the most consistent. Now,
- 2:10:56if there's no API, that's where you
- 2:10:58would maybe say, "Okay, cool. Well, now
- 2:10:59let me see if I could use like a
- 2:11:01deterministic, you know, like macros
- 2:11:04script, right? And that's where you'd
- 2:11:05basically be able to have Yes, you know,
- 2:11:07it'd be more of a browser use because
- 2:11:09you have a mouse that's clicking around,
- 2:11:11but it's not like your mouse needs
- 2:11:12reasoning. Your mouse doesn't need to
- 2:11:14look at the screen, figure out where to
- 2:11:15click, and based on the next UI that
- 2:11:17pops up, where do I click again? This is
- 2:11:19basically more like, okay, I have to
- 2:11:21click on this pixel, then this pixel,
- 2:11:24then this pixel every single time.
- 2:11:26That's something that you could do for
- 2:11:27much cheaper and much safer by just
- 2:11:30programming a macro. And that's not that
- 2:11:32hard either. You just ask Claude or
- 2:11:33Codex, hey, help me figure out how to
- 2:11:34make like a deterministic macro script
- 2:11:36to just click on these pixels in that
- 2:11:38order every single time. But if you
- 2:11:40can't do that every single time because
- 2:11:41there is an element of vision or
- 2:11:43reasoning, then you need to come in here
- 2:11:44and you need to do something like some
- 2:11:46browser use. And luckily, Codex has your
- 2:11:48back. I've tested a lot of different
- 2:11:50apps and different CLIs and different
- 2:11:52things for browser use. And so far,
- 2:11:53Codex is just really solid. So, as you
- 2:11:55can see now in my downloads, this is the
- 2:11:57statements that it just downloaded as a
- 2:11:58zip file and it basically took these and
- 2:12:00it put it into my Herk 2 project. If I
- 2:12:02go in here and I scroll down to
- 2:12:04statements and I go to um probably June.
- 2:12:07Yes, June. You can see right here, it
- 2:12:08put those in here in the correct spot.
- 2:12:11Okay, so that was an example where we
- 2:12:13needed to save a password somewhere,
- 2:12:15right? But I've done this with things
- 2:12:17like school. I've done this with things
- 2:12:18like X. I've even done this with things
- 2:12:19like Instagram that don't make you
- 2:12:21constantly resign in because it saves
- 2:12:23that session. So, let me show you
- 2:12:24another example. I'm going to go over
- 2:12:26here to YouTube and I'm going to grab
- 2:12:27the URL from the video that I uploaded
- 2:12:29earlier today. I'm going to go back into
- 2:12:31Codeex, paste in that URL, and say, I
- 2:12:33need you to turn this YouTube video into
- 2:12:35an X article for me. and then use
- 2:12:37browser use to go ahead and um format it
- 2:12:40and put it into an X article as a draft
- 2:12:43and then I will go ahead and review it.
- 2:12:44So, I know some of you guys are probably
- 2:12:45thinking, why don't you just use X's API
- 2:12:47for that? You can, but it's a little bit
- 2:12:50weird with the formatting. It's also
- 2:12:51weird with sometimes adding things like
- 2:12:53media or thumbnails for some reason, at
- 2:12:55least with my testing that I used to do.
- 2:12:57And it's also it charges you, right?
- 2:12:59Because you have to pay X per usage in
- 2:13:02order to do something like that. And so
- 2:13:03because this is just part of my codec
- 2:13:05subscription, I can just run this skill
- 2:13:07and it will go ahead and format
- 2:13:08everything for me and take care of it. I
- 2:13:10even have this skill worked out where it
- 2:13:11can go ahead and open up my YouTube
- 2:13:12videos and it can take screenshots of
- 2:13:14specific moments and put those into the
- 2:13:16article too where it makes sense. So
- 2:13:18obviously this is a bit of a loaded
- 2:13:19process because it has to do a few steps
- 2:13:20before it actually has the article ready
- 2:13:22to put into X. So while that's
- 2:13:24happening, let's just open up a new chat
- 2:13:25and let's see if our X session is there.
- 2:13:28Can you open up a browser? I want to see
- 2:13:30it right here in the app and just go to
- 2:13:32X and scroll around a little bit and
- 2:13:34show me if there's anything new that has
- 2:13:35just come out um like in the AI space.
- 2:13:38Just scroll on my X feed. And so because
- 2:13:39this was a brand new session, we can
- 2:13:41actually see if X is already signed in
- 2:13:43and if it's able to interact with
- 2:13:44things.
- 2:13:48Cool. So it just opened up X. Let me get
- 2:13:50my mouse out of here. You can see
- 2:13:52clearly that this is already signed in
- 2:13:53as me and my feed is visible and now it
- 2:13:55is going ahead and it's starting to
- 2:13:57scroll on the X page. So just start to
- 2:13:59think about already how could you create
- 2:14:01some sort of skills out of this and then
- 2:14:03schedule these as tasks. So maybe every
- 2:14:05morning if you wanted it to scroll
- 2:14:06through Instagram and things like that
- 2:14:08or scroll through your own profile or
- 2:14:10mark things as red or unread, you could
- 2:14:12totally go ahead and do that no problem
- 2:14:14because you can set up the schedule
- 2:14:15tasks right in here and use browser use
- 2:14:17skills that you've already set up. So
- 2:14:19you can see the two things that I
- 2:14:20already noticed was Grockbot and Grock
- 2:14:214.6. So this is an interesting case
- 2:14:24though because there is an API but using
- 2:14:26the browser use is actually cheaper. So
- 2:14:28that's why in this case we decided to do
- 2:14:30it and test it out. Now what about
- 2:14:32computer use that you guys have also
- 2:14:33probably heard. These are basically the
- 2:14:35same thing. It's still going to use
- 2:14:36vision. It's still going to use AI. But
- 2:14:38the difference here is that the browser
- 2:14:39is basically things that are on the
- 2:14:41internet or things that you would type
- 2:14:43into a browser like Chrome. But the
- 2:14:45computer use are things that might be on
- 2:14:46your actual computer. So [snorts]
- 2:14:48changing your computer settings, opening
- 2:14:49up desktop apps of things. So, let's go
- 2:14:52back into codeex. While we're waiting
- 2:14:53for this X article to finish up. Let's
- 2:14:54see how this is going. It is still
- 2:14:56breaking down and turning it into an
- 2:14:58article. Let's see if we can do some
- 2:14:59computer use. So, one thing you will
- 2:15:01have to do though is you'll have to go
- 2:15:02to plugins and you'll have to actually
- 2:15:03install computer use, which is just as
- 2:15:05simple as typing in computer use and
- 2:15:06hitting install like literally 2
- 2:15:08seconds. And then, hey Codex, I need you
- 2:15:10to use computer use to open up my
- 2:15:12desktop app for Glido. And I just want
- 2:15:14you to go to the settings and turn on
- 2:15:16the ability for it to allow and use like
- 2:15:18Bluetooth devices. So, just go ahead and
- 2:15:20get that set up for me. And so, this is
- 2:15:21a little bit different, right? Because
- 2:15:22it's not going to open up your inapp
- 2:15:24browser. It's going to actually kind of
- 2:15:26show your screen being taken over. It
- 2:15:29said, "I'll stop if Windows asks for an
- 2:15:31admin password or presents a security
- 2:15:33sensitive permission that it can't
- 2:15:34safely interpret." It shouldn't run into
- 2:15:35any of that, but it's good that it sort
- 2:15:37of has that guardrail baked in.
- 2:15:38Interesting. Okay, so it said Bluetooth
- 2:15:41access is a privacy permission, so the
- 2:15:42computer use won't let me actually do
- 2:15:44that, but what it said it's going to do
- 2:15:45is open up Glido. As you can see, I
- 2:15:47didn't open that up. It opened up Glido
- 2:15:48for me. And it doesn't look like on my
- 2:15:50recording you guys can actually see this
- 2:15:51right now, but my screen kind of has
- 2:15:53like a big blue hue around the edges. So
- 2:15:56you will clearly know if Codeex is using
- 2:15:59your computer.
- 2:16:05And there you go. You can see that it
- 2:16:06actually turned on the include Bluetooth
- 2:16:08devices button. And I think it got
- 2:16:09confused because I think at first it
- 2:16:11thought that I was messing with my
- 2:16:12actual computer settings rather than
- 2:16:14like an inapp settings. So it thought
- 2:16:15that that was a security thing. But it
- 2:16:17looks like it figured that out on the
- 2:16:18way. And I think the computer use thing
- 2:16:20is pretty interesting because you can
- 2:16:21remote into your codecs from your phone.
- 2:16:24So I can control all of these sessions
- 2:16:26on my phone. And what I could do is on
- 2:16:28my phone I could say, "Hey, computer
- 2:16:29use, go ahead and change my settings or
- 2:16:31go ahead and do something on a desktop
- 2:16:33app that's on my PC that I'm actually
- 2:16:35not sitting in front of." And then you
- 2:16:36could control your computer from your
- 2:16:38phone. So anyways, we're back in the
- 2:16:40session where it's creating that X
- 2:16:41article. You can see that it is actually
- 2:16:43creating some thumbnails for the actual
- 2:16:44article. So hopefully we're getting
- 2:16:46close to the point where it's going to
- 2:16:47actually open up X and start to put
- 2:16:49stuff in there. But I'll check in with
- 2:16:50you guys when we get there. Okay, so it
- 2:16:52just decided to open up X. You can see
- 2:16:54that it's going to the drafts in my
- 2:16:56signed X and then it's going to will go
- 2:16:58ahead and format this new article and
- 2:17:00then we'll take a look. It's clicking on
- 2:17:01the plus button. Man, I think it even
- 2:17:03just like resized the page a little bit.
- 2:17:05It was getting a little picky. In goes
- 2:17:06the title and then it just pasted in the
- 2:17:09entire content of the article. And now
- 2:17:11it's uploading this thumbnail which I
- 2:17:13don't think looks very good. that part
- 2:17:14of the skill has not been refined at
- 2:17:16all, but it's still cool that it was
- 2:17:17able to make that thumbnail and then go
- 2:17:19ahead and put it in there. So, I
- 2:17:20literally just have to make sure that it
- 2:17:22looks okay and I like it and I can
- 2:17:23publish it. Now, I know what a lot of
- 2:17:24you guys are probably thinking when it
- 2:17:26comes to this browser stuff. It's like,
- 2:17:28what's the point, Nate? Like, where's
- 2:17:29the value? And honestly, I understand
- 2:17:30where you're coming from. And that's why
- 2:17:31I wanted to show you guys this piece
- 2:17:33where it's like pretty much API will
- 2:17:35handle 95% of your automation things
- 2:17:37that you need. And then a lot of the
- 2:17:39browser use opportunities might just be
- 2:17:40a script, but then when you need it, if
- 2:17:42you have this in your back pocket, it's
- 2:17:44good. And you also might think, okay,
- 2:17:45cool, but like how much time does this
- 2:17:46really save? This probably saves me like
- 2:17:48maybe 5 to 10 minutes, like
- 2:17:50realistically. But what's cool about it
- 2:17:52is that I can basically have this
- 2:17:54pipeline now where I have a bunch of
- 2:17:55these browser use skills. And so when I
- 2:17:57need to do something, I just open up
- 2:17:58codeex, I shoot off that skill, and I
- 2:18:00can go back to the other task that I was
- 2:18:01working on. It kind of just like helps
- 2:18:02you do more without context switching as
- 2:18:05much. And as you start to stack more and
- 2:18:07more of these pipelines and combine them
- 2:18:08together, you can get some really cool
- 2:18:10things to happen when you start to chain
- 2:18:11together these skills. And look, this
- 2:18:13one actually did it. It basically did
- 2:18:14the thing where I said it opened up the
- 2:18:16YouTube video. It took screenshots of
- 2:18:18it, and now it's going ahead and it's
- 2:18:19planting some screenshots from my video
- 2:18:21in the article. So, if that doesn't
- 2:18:23prove the capabilities of this thing
- 2:18:24being intelligent, take screenshots,
- 2:18:26look at the page, know where to click,
- 2:18:27know what to do, then I don't know what
- 2:18:29else will. Besides the fact that I've
- 2:18:30had this thing play chess, I've had it
- 2:18:31play games. I've had it set a world
- 2:18:33record on Tetris, it can do a lot of
- 2:18:35cool things.
- 2:18:37Look at this. You can see that it even
- 2:18:39took the image and drugg it all the way
- 2:18:41down to where it actually fits in the
- 2:18:43article. I haven't actually seen it do
- 2:18:44that before live. That was pretty cool.
- 2:18:46All right, I hope you guys are enjoying
- 2:18:47all of these use cases. This next one is
- 2:18:49personally one of my favorites. I used
- 2:18:51Codex to edit the intro for this video.
- 2:18:53I use it to edit probably this course
- 2:18:55that you're watching. I use it a lot and
- 2:18:56my team also uses it a lot for courses
- 2:18:58and for ads and for creative content.
- 2:19:01So, it is one of the things that truly
- 2:19:03helps us move really fast because
- 2:19:04editing is no longer the bottleneck when
- 2:19:06you can use a tool like Codeex,
- 2:19:08especially if you've never edited a
- 2:19:09video before and you don't have video
- 2:19:11editing experience. So, let's hop into
- 2:19:13this next section about how you can use
- 2:19:14Codeex to help you edit some videos.
- 2:19:16Stop prompting Claude. Andre Garpathy
- 2:19:18thinks there's a much better way to work
- 2:19:19with AI and his method has three layers.
- 2:19:21Layer one is the spec. Instead of giving
- 2:19:23Claude a task and hoping it understands
- 2:19:24you, work with it to create a detailed
- 2:19:26spec first. Have Claude interview you
- 2:19:28about what you're actually trying to
- 2:19:29achieve. Then break the work into
- 2:19:31smaller checkpoints before it starts. So
- 2:19:33Nvidia just made building with AI
- 2:19:34basically free. They're giving
- 2:19:35developers free API access to over 80 AI
- 2:19:38models. Here's how you use it. Go to
- 2:19:40Nvidia, choose the model you want, and
- 2:19:41generate an API key. [music] You can
- 2:19:43choose from models like Kimmy, GLM,
- 2:19:44Deepseek, and loads more. So, you want
- 2:19:46to start a real business, but you can't
- 2:19:47afford people for content, marketing,
- 2:19:49sales, and all the other types of work
- 2:19:50yet because you can now install AI
- 2:19:52agents for free. One to help create
- 2:19:54content, one to help follow up with
- 2:19:55leads, and other types of agents to help
- 2:19:57handle the jobs that your first
- 2:19:58employees normally would do for you.
- 2:20:00Isn't it crazy that with just my natural
- 2:20:02language, I can get all of these crazy
- 2:20:03motion graphics right here, and I can
- 2:20:05get all of these insane animations over
- 2:20:06here as well. I can have subtitles
- 2:20:08appearing at the bottom, like you see. I
- 2:20:09can take my face and I can put it in the
- 2:20:11bottom right corner, or I can bring it
- 2:20:12up to the top left corner. And I can
- 2:20:14even play you guys clips of a bunch of
- 2:20:15my different YouTube videos right here
- 2:20:17in this sort of 3D fashion. So by the
- 2:20:19end of this video, you will know exactly
- 2:20:20how to actually do this, even if you've
- 2:20:22never really used codec before or you've
- 2:20:24never edited videos before. So I don't
- 2:20:26want to waste any of your time. Let's
- 2:20:27just get straight into this video. All
- 2:20:28right, so I hope you guys are excited to
- 2:20:30learn how to do this. What you're going
- 2:20:32to need is you're going to need the
- 2:20:33Codex desktop app and you're going to
- 2:20:34need to install Hyperframes. But I'm
- 2:20:36going to show you guys all that. Before
- 2:20:37we go through the setup, I just wanted
- 2:20:38to show you a few examples of how I've
- 2:20:40been using it and the skills that I'm
- 2:20:41going to give you guys for completely
- 2:20:42free. So this first example is just
- 2:20:43having it help me edit YouTube videos
- 2:20:45because what it's able to do is it
- 2:20:47transcribes the audio and then it can
- 2:20:48cut out mistakes and stutters and things
- 2:20:50like that and you know dead space. So
- 2:20:52for example in this one I gave it two
- 2:20:54different recordings. I gave it my face
- 2:20:56cam and I gave it my screen recording.
- 2:20:57And this was a 14-minute video that got
- 2:20:59cut down to 9 minutes and now it has
- 2:21:01like the background and it also has a
- 2:21:03few cuts. So let me just show you what I
- 2:21:04mean by that. So here's the video. It's
- 2:21:06not live yet but this is how it starts
- 2:21:07off. Today I'm going over the 18 core
- 2:21:10Codex concepts that you actually need to
- 2:21:11know in order to start using it right
- 2:21:13away and getting value from it. It
- 2:21:15doesn't matter if you're not technical
- 2:21:16at all or you've never used codeex
- 2:21:17before. By the end of this video you'll
- 2:21:18understand exactly how this thing
- 2:21:19actually works so that you can start
- 2:21:20using it right away. So as you can see
- 2:21:22that first animation with the 18 cards
- 2:21:24coming into the screen that was
- 2:21:26hyperframes as well as you know codecs
- 2:21:29here driving the actual edits, the
- 2:21:31rounded crops, the background, all of
- 2:21:33this. And so part one we're going to
- 2:21:35start off here is foundations. So we're
- 2:21:37going to kick off with concept number
- 2:21:39one which is projects. Now a lot of you
- 2:21:41guys I'm assuming have used chatbt or
- 2:21:43claude in the now all of that logic was
- 2:21:45once again hyperframes. It basically
- 2:21:47created the card that said you know
- 2:21:48number one projects. It then decided to
- 2:21:50also zoom me in here because it realized
- 2:21:53that for the next you know 10 to 15
- 2:21:54seconds I might be talking about
- 2:21:56something but I'm not showing anything
- 2:21:57on screen. So, I prompted it in this way
- 2:21:59and I gave it the skills in this way to
- 2:22:01understand that when there's nothing
- 2:22:03going on on the screen, maybe you do a
- 2:22:04full screen, but then when it's time to
- 2:22:05do more of a tutorial, you go back into
- 2:22:07this view. So, anyways, throughout this
- 2:22:09whole video, that's basically what it
- 2:22:10does. It's able to jump between full
- 2:22:12screen and this basically, you know,
- 2:22:14template or scene and then it also
- 2:22:16creates those cards in between each of
- 2:22:18the concepts. There is like a transition
- 2:22:20screen. All right, so that's one quick
- 2:22:21example. Let's take a look at some short
- 2:22:23form content which I think is getting
- 2:22:24much much better. So, here I have it
- 2:22:26doing two reels. One is about Claude and
- 2:22:28one is about Nvidia. So, let me just
- 2:22:30real quick play these two for you and
- 2:22:32then we'll break them down. Stop
- 2:22:33prompting Claude. Andre Karpathy thinks
- 2:22:35there's a much better way to work with
- 2:22:36AI. And his method has three layers.
- 2:22:38Layer one is the spec. Instead of giving
- 2:22:40Claude a task and hoping it understands
- 2:22:41you, work with it to create a detailed
- 2:22:43spec first. Have Claude interview you
- 2:22:45about what you're actually trying to
- 2:22:46achieve. [music] Then break the work
- 2:22:47into smaller checkpoints before it
- 2:22:49starts. Layer two is the verifier.
- 2:22:51Before Claude does anything, define
- 2:22:52exactly what a good result looks like.
- 2:22:54Then give it ways to actually check its
- 2:22:55own work. Whether that's another AI
- 2:22:57model or real data or tests that it can
- 2:22:59run on its own. And layer three is the
- 2:23:01environment. Build a workspace where
- 2:23:02Claude already has your instructions,
- 2:23:04your knowledge, your skills, and your
- 2:23:05rules every single time you use it. So
- 2:23:07instead of becoming better at writing
- 2:23:09prompts, you're building an entire
- 2:23:10system around Claude that makes it
- 2:23:12better every time you use it. I made a
- 2:23:14full YouTube video breaking down
- 2:23:15Karpathy's method and exactly how to
- 2:23:16build this yourself. So just comment
- 2:23:18method and I'll send it to you. So
- 2:23:19Nvidia just made building with AI
- 2:23:20basically free. They're giving
- 2:23:21developers free API access to over 80 AI
- 2:23:24models. Here's how you use it. Go to
- 2:23:26Nvidia, choose the model you want, and
- 2:23:27generate an API key. You can choose from
- 2:23:29models like Kimmy, GLM, Deepseek, and
- 2:23:31loads more. Then take that key and plug
- 2:23:33it directly into whatever you're
- 2:23:34building with, so you can experiment
- 2:23:35with AI apps, automations, and other
- 2:23:37projects without immediately racking up
- 2:23:38API costs. And because they're OpenAI
- 2:23:40compatible APIs, you can use them with
- 2:23:42loads of existing AI workflows. [music]
- 2:23:44If you wanted to start building with AI,
- 2:23:45but didn't want to burn your own money
- 2:23:46testing different models, this makes it
- 2:23:48way easier. So just comment key and I'll
- 2:23:50send you the full setup. So I thought
- 2:23:51that those outputs were really, really
- 2:23:53solid. You can see that what it does is
- 2:23:54it kind of starts off with like a visual
- 2:23:56hook. I guess we'll go to the clawed one
- 2:23:57first. We have the three missing pieces
- 2:23:59and what it did here is it's kind of
- 2:24:00creating that open loop because it's
- 2:24:02saying, "Hey, yeah, you know, Andre
- 2:24:03Carpathy uses it in this way and it's
- 2:24:05creating this open loop of, oh wow, I
- 2:24:07need to stay to the end of this short to
- 2:24:08see what all three of these are."
- 2:24:10Because as we go through, we can see we
- 2:24:11have number one. We could talk about
- 2:24:13little number one and then we go into
- 2:24:14number two, the verifier. And then we
- 2:24:16get into number three finally at the
- 2:24:18end, which is about the environment. And
- 2:24:20so about every one or two seconds here,
- 2:24:22it's switching. You know, we've got some
- 2:24:23scenes where it's like this with B-roll
- 2:24:25up top and me down here. It's got some
- 2:24:27where it's a full screen animation. It's
- 2:24:28got some where it's full screen me. And
- 2:24:30all these animations are sort of like
- 2:24:31paper style. There's texture. It's 3D.
- 2:24:33They're always moving. So, we're never
- 2:24:35having one element be still. You've also
- 2:24:37noticed that it went out and it got all
- 2:24:39of this B-roll. It either created it or
- 2:24:40it recorded it itself. And the
- 2:24:43subtitles, it was made by Hyperframes
- 2:24:45and Astra. The music, the sound effects,
- 2:24:48all of this is being synced to the
- 2:24:50actual reel for engagement. And pretty
- 2:24:51much same thing with the Nvidia one.
- 2:24:53This one was shorter, but it's a similar
- 2:24:54style. This one has more green accents
- 2:24:56rather than orange because it's about
- 2:24:57Nvidia, but it has the same types of
- 2:24:59animations. It has the same types of,
- 2:25:00you know, scenes, and it has the same
- 2:25:02way that it's fast-paced. It's always
- 2:25:04switching. We have music. We have real
- 2:25:06B-roll that it went out and it collected
- 2:25:07on its own. And here's even more AI
- 2:25:09generated B-roll that it created on its
- 2:25:11own. I think that this is getting
- 2:25:12really, really good. And I've turned
- 2:25:14this into a skill. And like I said, all
- 2:25:16of these are skills that come inside of
- 2:25:17one sort of student kit is what I'm
- 2:25:19calling it. So, if you want to get it
- 2:25:20all for free, just go into my free
- 2:25:22school community. The link for that is
- 2:25:23down in the description. And then you're
- 2:25:24going to go to the classroom and you'll
- 2:25:25go to either AI skills or all YouTube
- 2:25:28resources. I have it linked in a bunch
- 2:25:29of spots, so it's really easy to find,
- 2:25:31but you'll be able to find it right in
- 2:25:32here in the AI skills section. It'll be
- 2:25:33called like hyperframes student kit.
- 2:25:35Here's another quick example. This one
- 2:25:36is an AIS reel, and it made this one for
- 2:25:38me in both um 16x9 as well as 9x6. But
- 2:25:42let me just real quick play the 9x6
- 2:25:44version for you guys because it's kind
- 2:25:45of like a reel, but it's a little bit
- 2:25:46different cuz this one was meant to be
- 2:25:47an ad. What if you could build the team
- 2:25:49before you hired it? So, you want to
- 2:25:50start a real business, but you can't
- 2:25:52afford people for content, marketing,
- 2:25:54sales, and all the other types of work
- 2:25:55yet? Because you can now install AI
- 2:25:57agents for free. One to help create
- 2:25:58content, one to help follow up with
- 2:26:00leads, and other types of agents to help
- 2:26:02handle the jobs that your first
- 2:26:03employees normally would do for you.
- 2:26:04It's like building your first AI powered
- 2:26:06team before you hire your first
- 2:26:08employee. And that doesn't eliminate
- 2:26:09every future hire. It actually gives you
- 2:26:11the capabilities to start before you
- 2:26:12need them. At AIS Live, I'll show you
- 2:26:14how to build your own AI OS, a
- 2:26:16personalized system of agents, tools,
- 2:26:18and workflows that helps you operate a
- 2:26:20real business. You'll discover the most
- 2:26:22important AI agents for starting a
- 2:26:23company, and how to install them for
- 2:26:25free. You'll also see how an AIOS can
- 2:26:27support content, marketing, sales,
- 2:26:29operations, and even recruiting as your
- 2:26:31company needs to keep growing. So, if
- 2:26:32you want to start a highly profitable
- 2:26:34business without building a huge team,
- 2:26:35then click below for more details on AIS
- 2:26:37Live. Now, it has similar elements,
- 2:26:39right? But what I want you to pay
- 2:26:40attention to here is all of the B-roll
- 2:26:41that it went and collected from me. It
- 2:26:43grabbed YouTube videos of mine. It went
- 2:26:45ahead and it grabbed past AIS live, you
- 2:26:48know, segments. So here's one with me
- 2:26:49and Pat. We can go into here's like a
- 2:26:51VIP session. Here's a video of me like
- 2:26:53talking about my iOS and a second brain.
- 2:26:55So it was able to use the context of
- 2:26:57what the ad was about. Here's a session
- 2:26:58with Hyper Agent and Alex. It used the
- 2:27:01context of what we were talking about
- 2:27:03and it actually put it inside so that
- 2:27:04this doesn't just feel so much like
- 2:27:06theory. It feels more real. and even
- 2:27:08animated this logo here at the end,
- 2:27:09which I thought was just beautiful
- 2:27:11because it had to go to my Google Drive
- 2:27:12to find the AIS Live logo. And I think
- 2:27:14that this turned out really, really
- 2:27:16nice. And I'll play one more for you
- 2:27:17guys real quick, which was a call to
- 2:27:19action that I gave Codeex. Real quick,
- 2:27:21guys, just wanted to say if you're
- 2:27:22enjoying this video and you're enjoying
- 2:27:24the way that I teach, please consider
- 2:27:25subscribing and joining my free school
- 2:27:27community. The link for that is down in
- 2:27:28the description. We're building a global
- 2:27:30community of AI builders and
- 2:27:31entrepreneurs and business owners. We're
- 2:27:33almost at 450,000 members. We're almost
- 2:27:35at half a million, which is super super
- 2:27:36exciting. and it would just mean a lot
- 2:27:38to me if you joined. We have a lot of
- 2:27:40resources in here. We've got agent
- 2:27:41skills. We've got every YouTube video
- 2:27:43where I give away free resources. Like I
- 2:27:45said, skills, guides, templates, docs,
- 2:27:48GitHub repos. I give it all away in here
- 2:27:50for completely free. We also have a
- 2:27:517-day challenge. We have a build your
- 2:27:52AIOS course. We've got a lot of fun
- 2:27:54stuff going on in here. Events,
- 2:27:56certification programs, tons of cool
- 2:27:58things that we're building towards. So,
- 2:28:00if you're enjoying it, if you like the
- 2:28:01way I teach, please definitely check it
- 2:28:03out. Hope to see you guys in there.
- 2:28:04Let's get back to the video. Now, what I
- 2:28:05liked about this is this one felt very
- 2:28:07clean, right? It wasn't super
- 2:28:08fast-paced, super energetic. There's low
- 2:28:10music, but once again, it's proving to
- 2:28:11us that it can go grab screenshots, it
- 2:28:13can crop them, and it can use context
- 2:28:15about me and my business in order to
- 2:28:17make these feel more relevant. It
- 2:28:18grabbed all of these different, you
- 2:28:19know, skills and docs, and it put them
- 2:28:21in here in a really, really nice way. It
- 2:28:23even went and screenshotted some of our
- 2:28:24courses from the community. It grabbed,
- 2:28:26you know, event assets and pictures and
- 2:28:27videos, and it just made this call to
- 2:28:29action feel so much more real, so much
- 2:28:31more human. So, if you guys are getting
- 2:28:33excited by that, go ahead, like I said,
- 2:28:34to my free school community and grab all
- 2:28:36of those skills to be able to replicate
- 2:28:38those sorts of outputs. But now, let's
- 2:28:40actually get started with building this.
- 2:28:41Like, how do you actually do it? And I'm
- 2:28:42going to walk through a real example
- 2:28:43with you guys. If you guys remember the
- 2:28:45intro from this video, I'm going to
- 2:28:47build that out live right now with you
- 2:28:48guys. So, the first thing that I want
- 2:28:50you to do is you're going to open up a
- 2:28:51new project. So, I would just open up a
- 2:28:53new folder in your desktop or something.
- 2:28:55And we'll just call this one for now.
- 2:28:57I'm going to call this um Hyperframes
- 2:29:00demo. And that is the project that we're
- 2:29:03going to be working inside of. So I've
- 2:29:05got this folder on my desktop. Now I'm
- 2:29:07going to come into here on the lefth
- 2:29:08hand side and open up a new project. I'm
- 2:29:10going to choose for this to be a local
- 2:29:12project. I'm going to give this a name.
- 2:29:13So Hyperframes demo. And then I'm going
- 2:29:16to choose that actual folder so that
- 2:29:17Codex can work inside of it. So that was
- 2:29:19on my desktop and it was called
- 2:29:21Hyperframes demo. Cool. So there's the
- 2:29:23project. I'm going to select that and
- 2:29:24open this up. So, when you open up a new
- 2:29:26chat in this project, there's going to
- 2:29:28be nothing in here. So, if I go over
- 2:29:29here and I go to my files, there's
- 2:29:31literally nothing here, but we're going
- 2:29:32to start to fill this up with things
- 2:29:33like video projects and different
- 2:29:35renders and different files and things
- 2:29:36like that. So, the first thing that I
- 2:29:38want you to do is I want you to go to
- 2:29:40this GitHub repo, which I will link in
- 2:29:41the description. And this is called
- 2:29:43Hyperframe. So, you're basically just
- 2:29:44going to copy this URL for the
- 2:29:46Hyperframes GitHub repo. You're going to
- 2:29:48open up Codex, paste that in there, and
- 2:29:50then basically just say, "Hey, Codex, I
- 2:29:53need you to um basically pull in this
- 2:29:55hyperframes repo so that I can um use it
- 2:29:59to have you edit videos for me and
- 2:30:00stuff. I need you to make sure that all
- 2:30:03the dependencies are installed and
- 2:30:05everything like that and, you know, grab
- 2:30:06in some skills or things that you need
- 2:30:08inside of this codeex project in order
- 2:30:10to actually make this stuff useful." So,
- 2:30:12that's the first step. You're going to
- 2:30:13go ahead and shoot that off. Now, while
- 2:30:14this is getting set up, let me talk
- 2:30:16about the mindset and the idea of what
- 2:30:18we're actually doing here when we're
- 2:30:20editing videos with AI. So, there's a
- 2:30:21few steps, right? And in this case, I'm
- 2:30:23taking the example of we're giving it
- 2:30:24some sort of footage and having it edit
- 2:30:26it and animate it and things like that,
- 2:30:28but it can also create things from
- 2:30:30scratch. So, if you gave it um a rough
- 2:30:32outline and you wanted just motion
- 2:30:34graphics and like sizzle reels or like
- 2:30:35SAS demos, it could do that 100%. But
- 2:30:38that's not what I'm going to be focusing
- 2:30:39on right now. You would just remove this
- 2:30:41first step. So basically in my mind,
- 2:30:42step one is you have footage and what
- 2:30:44you need to do is you need to transcribe
- 2:30:46it. And basically that means it needs to
- 2:30:48find out all the text that you're saying
- 2:30:50in that clip and exactly when it
- 2:30:53happens, the exact second to the
- 2:30:54millisecond of when you say certain
- 2:30:57words so that it can sync it up with
- 2:30:58different animations and different text
- 2:31:00like subtitles for example. So
- 2:31:01transcribing comes first. You can do
- 2:31:03this with a free local thing that it can
- 2:31:04set up like called whisper, but I
- 2:31:06actually like to use 11 Labs for this.
- 2:31:08So I'm going to show you how to set that
- 2:31:09up as well. After you transcribe, then
- 2:31:11the next thing that we want to do is we
- 2:31:13probably want to cut. So, basically
- 2:31:15having it cut out um you know, mistakes
- 2:31:17or five seconds of silences and things
- 2:31:20like that. So, that would be step two.
- 2:31:23Step three then in my mind is to plan
- 2:31:25the beats. That's kind of what it calls
- 2:31:26our beats. So, beats are basically
- 2:31:28different scenes. So, for example, with
- 2:31:29this reel, like this install capability
- 2:31:32slide, this is one beat and then like
- 2:31:33this is one beat and this is one beat.
- 2:31:35Every basically scene is a different
- 2:31:37beat. So now what we like to do is we
- 2:31:39like to have Astra, the model, read
- 2:31:41through the text, understand the intent,
- 2:31:43and start to plan out the beats of
- 2:31:45what's actually going to go inside of
- 2:31:46this reel or inside of this video. And
- 2:31:48then what it's going to do is it's going
- 2:31:49to use its, you know, skills, and it's
- 2:31:51going to use hyperframes, and it's going
- 2:31:53to use other tools like this in order to
- 2:31:55actually do the generation of the HTML.
- 2:31:58It's basically just creating HTML and
- 2:32:00it's animating it. And that's what
- 2:32:01really gives us all these 3D elements
- 2:32:02and it gives us this text and it gives
- 2:32:04us what we are basically looking at and
- 2:32:06considering the motion graphics or the
- 2:32:07motion design. And by the way, part of
- 2:32:09planning the beats is like understanding
- 2:32:11that there's going to be like, you know,
- 2:32:13music that's going to sync to it and
- 2:32:15sound effects and all of that is
- 2:32:16basically coming into um this element of
- 2:32:19planning. And the thing about it is we
- 2:32:21have to have transcribed and cut all of
- 2:32:23that first in order to be able to do
- 2:32:24this properly. And then the last step
- 2:32:27here is basically honestly it's kind of
- 2:32:28a loop. So the last step is basically to
- 2:32:30verify, meaning have the agent watch
- 2:32:33through all of this again. It will watch
- 2:32:35through it. It will take screenshots. It
- 2:32:37will look at the transcript again. And
- 2:32:39basically what we want to do is we're
- 2:32:41going to do this loop of verifying and
- 2:32:43then going back into here and using the
- 2:32:45skills again and then verifying and then
- 2:32:47using the skills again. So we get in
- 2:32:49this loop of just going like this until
- 2:32:51the agent is confident that everything
- 2:32:52is in bounds and everything syncs
- 2:32:54properly and everything feels how it
- 2:32:56should. So that's kind of the flow.
- 2:32:58Let's actually go get set up. So, we are
- 2:33:00now inside of this project and it's
- 2:33:03still getting everything set up. So,
- 2:33:04what I want to show you what to do next
- 2:33:05is we want to do um 11 Labs to
- 2:33:08transcribe. So, you're going to go to a
- 2:33:10new page. You're going to open up 11
- 2:33:12Labs and create an account if you
- 2:33:14haven't. You can get started on this for
- 2:33:16like a five bucks a month plan, which it
- 2:33:18it'll last you quite a while. So, you're
- 2:33:20going to come in here and then what
- 2:33:21you're going to do is you're going to go
- 2:33:22on the bottom left to developers and
- 2:33:23you're going to click on API keys right
- 2:33:25here and you're going to go ahead and
- 2:33:26create a new key. I'm just going to call
- 2:33:28this um Astra demo. You can have this
- 2:33:31expire. You could just leave this at
- 2:33:32never. And then as far as restricting
- 2:33:34the key, if you want to say, okay, this
- 2:33:36key is literally only going to
- 2:33:38transcribe stuff. That would be speech
- 2:33:40to text. You could give it access to
- 2:33:42speech to text. If you also want to give
- 2:33:44it access to like sound effects and
- 2:33:45music generation and image and video
- 2:33:47generation, you can give it access to
- 2:33:48other things as well. But right now, I'm
- 2:33:50just going to start with just speech to
- 2:33:51text because all we really need for this
- 2:33:52endpoint is the um transcription. So,
- 2:33:55I'm going to create this key. This gives
- 2:33:56us basically a password. So, don't share
- 2:33:58this with anybody. And I'm going to
- 2:33:59delete this as soon as I'm done filming.
- 2:34:00So, don't even think about it. But,
- 2:34:01you're going to copy this to your
- 2:34:02clipboard. You're going to go into
- 2:34:04codeex and then we're going to come into
- 2:34:06your files right over here. And you can
- 2:34:07see now it's starting to build out some
- 2:34:08folders and files and things like that.
- 2:34:10And what you need is you need a file
- 2:34:11called aenv which lets you store secrets
- 2:34:14and passwords like your 11labs API key.
- 2:34:17So all you have to do is just you can
- 2:34:18chuck in a message even while it's
- 2:34:19working here. And look it actually
- 2:34:21opened up a local host which is kind of
- 2:34:22how we can see our video projects. This
- 2:34:24is just you know it just lets us test
- 2:34:26them and look at them and even like move
- 2:34:27elements around and stuff which is
- 2:34:29really really cool. But I'm just going
- 2:34:30to close out of that because that was
- 2:34:31just an example. But what I can do is
- 2:34:33say when you do the transcriptions I
- 2:34:35want you to use 11 labs. So, can you
- 2:34:38please create me availabs
- 2:34:41API key in there? And also make sure
- 2:34:43that you include this in the agents.md
- 2:34:46that says whenever you know Nate wants
- 2:34:48you to transcribe a video, use 11 Labs
- 2:34:50for that. So, the agents.mmd is this
- 2:34:52file. This basically just explains how
- 2:34:54the project works. So, right now,
- 2:34:56because there's not much information in
- 2:34:57there, it just has some very basic stuff
- 2:34:59from hyperframes. But as you start to
- 2:35:01learn more about the way you like to
- 2:35:02edit videos and the way you like to
- 2:35:04store your projects, you'll basically
- 2:35:05want to come in here and update it with
- 2:35:06certain information. So right here you
- 2:35:08can see it says, "Okay, I will add that
- 2:35:09rule into the agent.MD so that in the
- 2:35:12future I'll stop the whisper setup which
- 2:35:14is that local thing that I talked
- 2:35:15about." Now nothing's wrong with
- 2:35:16whisper, it's just slower. And so 11
- 2:35:18Labs, yeah, you have to pay for it, but
- 2:35:19it's much quicker and I like that speed
- 2:35:21and it's not that expensive. So now ourv
- 2:35:24has a placeholder for the 11 Labs API
- 2:35:26key. So, what you're going to want to do
- 2:35:27is just paste in your 11 Labs API key
- 2:35:29right inside of here. And by the way, if
- 2:35:31it's not letting you paste or type
- 2:35:32anything in here, um, which it should,
- 2:35:34but if it doesn't, then you can just go
- 2:35:36into here and click on open in default
- 2:35:38app. And for me, that opens up in my
- 2:35:39notes. I can paste in the API key. Hit
- 2:35:42save. And then, as you can see, it's
- 2:35:43going to refresh live right inside of
- 2:35:45here. All right. So, at this point, we
- 2:35:46have enough information in here to
- 2:35:48actually be able to use hyperframes. So,
- 2:35:50I'm not going to use my skills that I
- 2:35:52mentioned that I showed all of these
- 2:35:54examples where I use my skills. I'm not
- 2:35:56going to use these right now because I
- 2:35:57just want to show you what this kind of
- 2:35:58looks like at its core. But don't forget
- 2:36:00that you can definitely grab those
- 2:36:01skills and come in here and make these
- 2:36:02things better. And so one thing that you
- 2:36:04can do is you can give it a video file
- 2:36:06and just say, "Hey, edit this for me."
- 2:36:07You know, add motion graphics, make it
- 2:36:09tell a story, blah blah blah. Or you can
- 2:36:11be very specific, which is what I'm
- 2:36:13going to do for this specific example.
- 2:36:14So let me pull up the video that we're
- 2:36:16going to be working with. It's right
- 2:36:18here. It's called edit intro. So let me
- 2:36:19actually open this up and we'll give it
- 2:36:20a quick watch. This is just my natural
- 2:36:22language. I can get all these crazy
- 2:36:23motion graphics right here. And I get
- 2:36:24all these insane animations over here as
- 2:36:25well. I can have subtitles up here at
- 2:36:26the bottom like you see. I can take my
- 2:36:28face. I can put in the bottom right
- 2:36:28corner. Okay. Anyway, so now you see
- 2:36:30that I'm basically planning out what I
- 2:36:32want to happen as it's happening. So
- 2:36:34here's what I'm going to do. I'm going
- 2:36:36to grab the file path of my um video.
- 2:36:40I'm going to come back into here. I'm
- 2:36:42going to do a /goal prompt for this,
- 2:36:44which basically just means that I'm
- 2:36:45setting a goal and it's going to keep
- 2:36:46pursuing until it hits that goal. I'm
- 2:36:48going to paste in the file path of that
- 2:36:50video and I'm just going to start
- 2:36:51instructing what I want to happen when I
- 2:36:54want it to happen. But first, I have to
- 2:36:55set up the loop of transcribing,
- 2:36:58cutting, planning the beats, and then
- 2:37:00it's going to go ahead and get to work.
- 2:37:01But instead of planning the beats, I'm
- 2:37:03kind of directing this a little bit
- 2:37:04more. So, this is kind of like
- 2:37:05plan/accept beats. But anyways, let me
- 2:37:08show you what this is going to look like
- 2:37:09as I instruct this thing.
- 2:37:12So, I've just dropped in a video for
- 2:37:14you. This is going to be an intro. So, I
- 2:37:16need you to help me turn this into a
- 2:37:19super premium and polished edited video.
- 2:37:21So, the first thing you need to do is I
- 2:37:22want you to transcribe the video because
- 2:37:24we need all of the beats and all the
- 2:37:26motion graphics to sync exactly when I
- 2:37:28say them. And then I need you to cut out
- 2:37:30anything if there's mistakes or if
- 2:37:32there's um you know, silence. There
- 2:37:34shouldn't really be like a second or a
- 2:37:36half second of silence. I want this to
- 2:37:37feel fast-paced. After you've done those
- 2:37:39two things, it's time to start actually
- 2:37:41animating this video. So, let me tell
- 2:37:43you what I'm looking for when it comes
- 2:37:44to the animations. So, I start off this
- 2:37:46intro by saying something like, "Isn't
- 2:37:47it crazy that I can use just my natural
- 2:37:49language to get all these crazy motion
- 2:37:51graphics over here and all these crazy
- 2:37:52animations over here?" What I want to
- 2:37:54happen then is I want this whole vibe of
- 2:37:57this video to be very, very like Apple
- 2:37:59style motion design. So, like liquid
- 2:38:01glass, clean animations, um, and they
- 2:38:04feel very professional. So, when I say
- 2:38:08motion graphics over here, I believe
- 2:38:09that my hand was on it was pointing to
- 2:38:12the left side of the screen. So on the
- 2:38:14left third of the screen, I want them
- 2:38:16from top down, sort of like my hands are
- 2:38:20creating these motion graphics. I want
- 2:38:21them to come into screen and just make
- 2:38:23them, you know, against like some sort
- 2:38:25of liquid glass card where we see three
- 2:38:27different types of motion graphics.
- 2:38:28Maybe like cards and documents moving
- 2:38:30around or um laptop screens and charts
- 2:38:34and elements that are being like
- 2:38:35animated and things like that. Just make
- 2:38:37things over here that are impressive and
- 2:38:39that tell a visual story. And then
- 2:38:40similarly when I go into the insane
- 2:38:42animations over here as well that's on
- 2:38:45the right side of the screen. So same
- 2:38:46thing I want them to come in from top
- 2:38:47down like I'm drawing them in. I want
- 2:38:49these to be 3D. So I want these to be
- 2:38:51against some sort of um maybe little
- 2:38:54overlay in the back so that we can see
- 2:38:56them against the background just like a
- 2:38:58subtle dark overlay. And these should be
- 2:39:003D. So um they should maybe look like
- 2:39:03they're spinning. They should have
- 2:39:04shadows. They should have depth. And
- 2:39:06these can just be 3D like letters or 3D
- 2:39:09um animations and I don't know like
- 2:39:11symbols that are moving. Just make
- 2:39:12these, you know, subtle and
- 2:39:14professional, but they should certainly
- 2:39:15give some sort of like wow factor. And
- 2:39:17for those two beats, I want them to stay
- 2:39:20on the screen while I'm talking about
- 2:39:22them. And then I want them to leave the
- 2:39:23screen or sort of like fade away at the
- 2:39:26same moment. So the motion graphics on
- 2:39:27the left come in first, animations on
- 2:39:29the right come in second, and then they
- 2:39:31leave the screen at the same time. The
- 2:39:33next section is pretty simple. When I
- 2:39:35say I can have subtitles appearing at
- 2:39:37the bottom as you see, I just want a
- 2:39:38super super clean um overlay at the
- 2:39:41bottom of the screen with subtitles,
- 2:39:43they shouldn't cover um my face. So,
- 2:39:45they should be sort of near the lower
- 2:39:47third. And if you could just have the
- 2:39:49words like subtly highlighted with maybe
- 2:39:51like the words are white, but they're
- 2:39:53highlighted in blue when I'm actually
- 2:39:54saying them. So, make sure they're
- 2:39:55synced exactly to the transcript when
- 2:39:57I'm saying them. Now, in the next
- 2:39:58section, I go on to say, you know, I can
- 2:40:00take my face and I can move it to the
- 2:40:02bottom right corner or um the top left.
- 2:40:05And what I want you to do for that is I
- 2:40:07want you to take my face cam and sort of
- 2:40:10shrink it into like a um a vertical
- 2:40:13rounded crop. So, you kind of just bring
- 2:40:15in the left and right sides to keep me
- 2:40:17centered. And you make it rounded crop
- 2:40:18with a drop shadow. And you put it
- 2:40:20against this background, which is the
- 2:40:22full screen background. And so here what
- 2:40:24I'm doing is I just pasted in this
- 2:40:25picture which I like to use as uh my
- 2:40:28background in some YouTube videos. And I
- 2:40:29want you to make that background 50%
- 2:40:31opacity so it's a little bit darker. And
- 2:40:33then for the next part when I say I can
- 2:40:35play you guys these um clips from my
- 2:40:37YouTube videos, I want you to go to my
- 2:40:39YouTube video directory locally which
- 2:40:42I'll give you the file path too. So I
- 2:40:43will just copy in this file path real
- 2:40:45quick right there. What I want you to do
- 2:40:46is just grab a few of my most recent
- 2:40:48videos, maybe like four or six, and I
- 2:40:51want you to sort of fan them out, like
- 2:40:52animate the process of you fanning them
- 2:40:54out, and have these be 3D with like a
- 2:40:57little bit of a drop shadow, and they're
- 2:40:58sort of like rotating and swiveling back
- 2:41:00and forth just to create a really cool
- 2:41:02effect. But I want them to actually be
- 2:41:03playing throughout. I don't want them
- 2:41:05just be static images of the videos. I
- 2:41:07want you to just play, you know, all of
- 2:41:08them are sort of actually playing at the
- 2:41:10same time. And then to end off the
- 2:41:12video, what I want you to do is
- 2:41:13basically against that background I gave
- 2:41:15you, if you could bring my main face
- 2:41:17cam, sort of align it on the left side
- 2:41:19of the screen, but make it bigger so it
- 2:41:21kind of touches the top and bottom of
- 2:41:22the screen, but we should still see the
- 2:41:23rounded crop and the drop shadow. And
- 2:41:25then on the right side of the screen, we
- 2:41:27can just have some other text and other
- 2:41:28motion graphics that tell a story based
- 2:41:30on what I'm saying. So like um you know,
- 2:41:32you don't have to be technical, you
- 2:41:33don't have to have used codecs before,
- 2:41:35you don't have to have a video editing
- 2:41:37background. if you could just make
- 2:41:38little icons and animate them in a fun
- 2:41:40way to tell that story. And then at the
- 2:41:43end, you know, it says like, "Hey, I
- 2:41:44don't want to waste any time. Let's just
- 2:41:45get straight straight into the video."
- 2:41:46Do the exact same thing there. And
- 2:41:48that's going to pretty much be the end
- 2:41:50of this intro. After you have then done
- 2:41:52all of that, I just want you to make
- 2:41:53sure it looks good. I want you to sort
- 2:41:55of like verify it by taking some
- 2:41:57screenshots and looking through, making
- 2:41:59sure that the beats actually sync to the
- 2:42:01text that I'm reading out and make sure
- 2:42:03that um everything's in bounds and
- 2:42:05everything looks professional and it's
- 2:42:06giving off the quality that you want it
- 2:42:08to. I'm not looking for a PC or a
- 2:42:10version one. I'm looking for a finished
- 2:42:12product ready to go here. Out of breath
- 2:42:14a little bit. So, that was a massive
- 2:42:15prompt, right? That was me being super
- 2:42:17super specific about what I'm actually
- 2:42:19looking for here. And that's kind of
- 2:42:20what this looks like to start. The way
- 2:42:22that I got to all of these skills that I
- 2:42:24now use for YouTube videos, lessons,
- 2:42:26reals, ads, is because I did that so
- 2:42:28many times. And now if you get an output
- 2:42:31here that you really, really like, you
- 2:42:32say, "Cool, that was awesome. Turn that
- 2:42:34into a skill." And then next time, you
- 2:42:36don't have to give such a big prompt.
- 2:42:37You basically just say, "Hey, edit this
- 2:42:38video. Use this skill." And then when it
- 2:42:40comes back, if you have feedback, say,
- 2:42:41"Hey, I didn't like this. Make sure you
- 2:42:43do this next time." And then update the
- 2:42:45skill, too. So, what I just did is
- 2:42:47really the hardest it'll ever be. And
- 2:42:49then it's always just about iterating on
- 2:42:50those skills. So, I'm just going to go
- 2:42:52ahead and let this cook a little bit. I
- 2:42:53will check in with you guys when this
- 2:42:54one comes back with the first version.
- 2:42:56And hopefully it's what you saw in the
- 2:42:58beginning of this video. Okay. So, I
- 2:42:59have not yet watched this, but you can
- 2:43:01see that this went from the original
- 2:43:03being what was this? A little over a
- 2:43:05minute, a minute and 5 seconds to now
- 2:43:07being cut down to 28 seconds. Let's go
- 2:43:10ahead and give this a watch and see how
- 2:43:12this looks. Isn't it crazy that with
- 2:43:13just my natural language, I can get all
- 2:43:15of these crazy motion graphics right
- 2:43:16here? And I can get all of these insane
- 2:43:18animations over here as well. I can have
- 2:43:20subtitles appearing at the bottom like
- 2:43:21you see. I can take my face and I can
- 2:43:23put it in the bottom right corner or I
- 2:43:24can bring it up to the top left corner.
- 2:43:26I can even play you guys clips of a
- 2:43:28bunch of my different YouTube videos
- 2:43:29right here in this sort of 3D fashion.
- 2:43:32So, by the end of this video, you will
- 2:43:33know exactly how to actually do this
- 2:43:35even if you've never really used codecs
- 2:43:36before or you've never edited videos
- 2:43:38before. So, I don't want to waste any of
- 2:43:39your time. Let's just get straight into
- 2:43:41this video. Okay, not bad. Not bad for a
- 2:43:43first pass. Um, you can see how my very
- 2:43:45specific prompting obviously benefited
- 2:43:48us here. I think that these look good,
- 2:43:49right? Like we get some 3D animations,
- 2:43:51we get some motion graphics. I like it.
- 2:43:52I think that's a good way to start. I do
- 2:43:54think maybe we need something more in
- 2:43:56the first two seconds. Yeah, I think we
- 2:43:58should add something right there in the
- 2:43:59first two seconds just to make it more
- 2:44:00engaging. But the other thing I noticed
- 2:44:02is right here. It's playing videos,
- 2:44:03which is good, but they're like inter
- 2:44:05overlapping with each other, right? So
- 2:44:06they it's not realistic. What we need to
- 2:44:08do is make this feel more legit. And
- 2:44:11also, I want to make this text bigger. I
- 2:44:12don't think that text is big enough. And
- 2:44:14I think this looks fine. This is all
- 2:44:16simple. Nice little animation as well.
- 2:44:17Basically, that's my feedback. So, let
- 2:44:19us shoot off prompt number two. I'm
- 2:44:21going to once again do another slash
- 2:44:22goal prompt. Awesome. So, I mean, that
- 2:44:25was 18 minutes. That was a really good
- 2:44:27first pass, but let's do another version
- 2:44:29here. What I want is a couple things.
- 2:44:32First of all, in the first two seconds,
- 2:44:34I want you to get creative here. Do
- 2:44:36something that feels on brand
- 2:44:38to make the first two seconds more
- 2:44:41engaging. I'm not sure I want like a
- 2:44:43full screen cutaway and I don't want
- 2:44:45anything that's going to lay on top of
- 2:44:48my face, but you know, I start off the
- 2:44:50video by like, isn't it crazy that with
- 2:44:52just my natural language, some sort of
- 2:44:54animation, some sort of motion graphic,
- 2:44:56make that a little bit more emotional?
- 2:44:57Just do something right away to sort of
- 2:44:59like hook the viewer in. Now, besides
- 2:45:01that, the only other changes I need are
- 2:45:04in the part where you show my YouTube
- 2:45:05videos, you show the four. I like how
- 2:45:08these look like they're 3D and they have
- 2:45:10depth, but if you could make this more
- 2:45:12realistic because they clearly are like
- 2:45:16cutting through each other almost like
- 2:45:17it's not realistic. We need this to be
- 2:45:19more like the physics of it need to feel
- 2:45:22better. So, if you could just maybe make
- 2:45:24them thicker and make sure that they're
- 2:45:26they're fanned out behind each other
- 2:45:27rather than like cutting through each
- 2:45:29other, that would be great. And then the
- 2:45:32text on that screen, it says like your
- 2:45:33videos or something. um if you could
- 2:45:35just make that larger because that text
- 2:45:36was pretty small. I want it to be bigger
- 2:45:37and easier to read. But besides that,
- 2:45:39everything else I thought looked really
- 2:45:40good and it matched the transcript
- 2:45:43really well also. So make those changes
- 2:45:45please and then let me know when you're
- 2:45:47done. So that is prompt number two going
- 2:45:48off as another goal. And you can see
- 2:45:50that this is the actual hyperframe
- 2:45:52studio that it created for us. Um this
- 2:45:54is at 107. Let's just reset that. And
- 2:45:56what's cool is I can literally move
- 2:45:57things in here. So I can actually take
- 2:45:59this element and I could increase the
- 2:46:00text size over here. I think font this
- 2:46:02is where I would do it. I could change
- 2:46:04this to what, like 80, and see how that
- 2:46:06looks. So, it would reload and now I can
- 2:46:08just sort of manipulate this stuff right
- 2:46:10from here, which is pretty cool. You
- 2:46:11know, you can move stuff around and um
- 2:46:13you can see there's even different
- 2:46:14elements for the actual like 3D scene
- 2:46:16and the actual video inside. So, I'm not
- 2:46:18going to mess with that right now. I
- 2:46:19honestly don't use that interface too
- 2:46:21much. But if you're editing a really
- 2:46:22long video, let's say it's like a
- 2:46:2310-minute video, and you just want to be
- 2:46:24able to like tweak one thing really
- 2:46:26subtly, it's probably easier to come in
- 2:46:28here, make a little tweak, like just
- 2:46:30move this over and then export it rather
- 2:46:33than having to do the prompt layer
- 2:46:34again. So, it's really nice that
- 2:46:35Hyperframes gives you this local host
- 2:46:37environment as well. But for something
- 2:46:39as simple as what we're doing, I just
- 2:46:41wanted to shoot off another natural
- 2:46:42language prompt. So, I will check in
- 2:46:43with you guys once this one has finished
- 2:46:45up. All right, so that took 10 minutes
- 2:46:46now and V2 has come back. Let me go
- 2:46:48ahead and open this up full screen and
- 2:46:49we'll give it a watch. Isn't it crazy
- 2:46:51that with just my natural language, I
- 2:46:52can get all of these crazy motion
- 2:46:54graphics right here and I can get all of
- 2:46:55these insane animations over here as
- 2:46:57well. I can have subtitles appearing at
- 2:46:58the bottom like you see. I can take my
- 2:47:00face and I can put it in the bottom
- 2:47:01right corner or I can bring it up to the
- 2:47:03top left corner and I can even play you
- 2:47:05guys clips of a bunch of my different
- 2:47:06YouTube videos right here in this sort
- 2:47:08of 3D fashion. Okay, so this was much
- 2:47:10better. The text got larger and this
- 2:47:12feels more 3D and it also just feels
- 2:47:14like, you know, it just looks better.
- 2:47:16And so right here, this is pretty much
- 2:47:17the only section where I didn't tell it
- 2:47:19exactly what I wanted. And this is what
- 2:47:21it came up with. I don't honestly love
- 2:47:22it. But for now, I'm just going to keep
- 2:47:24it in because I told it I didn't want a
- 2:47:25full screen takeaway, which would have
- 2:47:27meant it created its own animation to be
- 2:47:29the full screen. But actually, let's
- 2:47:30just see what that would look like if I
- 2:47:32just shoot off that quick prompts real
- 2:47:33quick and say, "Awesome. Now, can you
- 2:47:35create me one more version where in the
- 2:47:37first two seconds before you cut into
- 2:47:39like the motion graphics and everything,
- 2:47:41let's see what this would look like if
- 2:47:42you did a full screen takeaway." So, it
- 2:47:44would be the full screen is animated um
- 2:47:46against some sort of, you know,
- 2:47:47background that feels on brand with the
- 2:47:49rest of the intro and maybe the text is
- 2:47:52sort of comes in in an animated way that
- 2:47:54says, "Isn't it crazy?" So, anyways,
- 2:47:56just going to shoot that off. This
- 2:47:57really shouldn't take long at all, but
- 2:47:58just to show you guys that it can also
- 2:48:00do things like, you know, a full screen
- 2:48:01cutaway and animate it in that way, too.
- 2:48:04So, I'll show you guys that in just a
- 2:48:05sec. All right, so that one came back
- 2:48:06and it's done. And I have a feeling that
- 2:48:08it's basically just going to be super
- 2:48:09simple right here. I'll play it from
- 2:48:10this interface. Isn't it crazy that with
- 2:48:12just my natural language, I can get all
- 2:48:14of these crazy motion graphics right
- 2:48:15here and I can get all of these Cool.
- 2:48:17So, that is looking good. We can see
- 2:48:18basically all this did was just add some
- 2:48:20very simple animated text that came in
- 2:48:22here. It had a little bit of layering as
- 2:48:23well with these little liquid glass
- 2:48:25looking cards and it kept the background
- 2:48:27consistent to what we actually used
- 2:48:28later over here. So, I thought that's
- 2:48:30pretty good. Anyways, guys, at this
- 2:48:31point, we've set up the project. You've
- 2:48:33set up something to transcribe if you
- 2:48:34decided to use 11 Labs and you
- 2:48:36understand now the loop that we actually
- 2:48:37go through in order to do all this. I
- 2:48:39didn't dive too much into this actual
- 2:48:41interface and how you control it, but
- 2:48:42like I said, it's very, very intuitive
- 2:48:43once you get in here and it's a nice
- 2:48:45touch by Hyperframes. So, don't forget
- 2:48:46if you want to get this student kit that
- 2:48:47has all of the skills that you need to
- 2:48:49build these sorts of reels and videos,
- 2:48:51you can download this for completely
- 2:48:52free by going to my free school
- 2:48:53community. The link is in the
- 2:48:54description. You'll go to classroom and
- 2:48:56click on AI skills and it will be in
- 2:48:58there. Okay. So, now that you have a
- 2:49:00really good understanding, at least I
- 2:49:01assume, of how Codex works when you're
- 2:49:03in it and you're sitting at your desk
- 2:49:04and you want to drive automations and
- 2:49:05drive, you know, skills and different
- 2:49:07sessions like that. But what if you want
- 2:49:09to build automations that run without
- 2:49:10you being there at all? And I know we've
- 2:49:12talked a little bit about scheduled
- 2:49:13automations, but what if you want to
- 2:49:15build things that live somewhere else? I
- 2:49:16don't know if you guys have ever used
- 2:49:17Nitn before or make, but what was cool
- 2:49:19about that is you could just deploy
- 2:49:20those automations in, you know, make
- 2:49:22server or NEN server or something like
- 2:49:24that. But what Codex lets you do is it
- 2:49:26lets you build automations in code. I
- 2:49:28know we don't want to look at code, I
- 2:49:29understand, but it lets you build
- 2:49:31automations in code and then you can
- 2:49:32just deploy those automations somewhere
- 2:49:35else. And now you don't have to have
- 2:49:36your Codex desktop app open or you don't
- 2:49:38have to have your laptop turned on in
- 2:49:39order for those automations to keep
- 2:49:41running all the time. So we're going to
- 2:49:42talk about how you can use Codex to
- 2:49:43build automations and then host those
- 2:49:45automations in something called
- 2:49:46trigger.dev. It is a lot simpler than it
- 2:49:48sounds. So let's just hop into this next
- 2:49:50video. Codex lets you build automations
- 2:49:51right inside of it using scheduled
- 2:49:53tasks. However, because these scheduled
- 2:49:55tasks basically just send messages right
- 2:49:57into an actual chat thread, what that
- 2:49:59means is this eats at your actual weekly
- 2:50:01usage limit. So, it doesn't really make
- 2:50:03sense to build a bunch of your personal
- 2:50:05and company automations inside of
- 2:50:06Codeex. A lot of the most useful
- 2:50:08automations can run somewhere completely
- 2:50:09independently of codecs, but we can use
- 2:50:11codecs to help us build them. So, that's
- 2:50:13exactly what I'm going to show you guys
- 2:50:14today. Even if you have no technical
- 2:50:16background at all, by the end of this
- 2:50:17video, you'll see how to build scheduled
- 2:50:18automations or ones that are triggered
- 2:50:20on some sort of event that live outside
- 2:50:22of codec so they don't eat away at your
- 2:50:24subscription. So, I don't want to waste
- 2:50:25any time today. Let's just get straight
- 2:50:26into this one. So, like I said, when you
- 2:50:28have an automation running inside of
- 2:50:30Codeex all the time, it's eating away at
- 2:50:32your subscription limit. So, what I'm
- 2:50:33going to show you guys today is how we
- 2:50:34actually build our automations with
- 2:50:36Codeex and then instead of actually
- 2:50:37hosting them there, we're going to host
- 2:50:39them in something called trigger.dev.
- 2:50:41Because when we actually build the
- 2:50:42automation, what happens is we can
- 2:50:44actually just turn that into code. So
- 2:50:46instead of having, you know, our
- 2:50:47schedule task run a skill on a certain
- 2:50:50cadence, it basically just turns that
- 2:50:51skill into code that can actually be
- 2:50:53executed and then that code is just
- 2:50:55given from codeex to GitHub and then
- 2:50:57GitHub basically says, "Okay,
- 2:50:58trigger.dev, you're supposed to run this
- 2:51:00code, you know, every hour or you're
- 2:51:02supposed to run this code whenever
- 2:51:03someone submits that form submission."
- 2:51:05So if that sounds overwhelming or scary
- 2:51:06at all, don't worry about it. It's going
- 2:51:08to be super simple. Just think of it as
- 2:51:09the idea that you're giving your skills
- 2:51:11to something else to run it so that it
- 2:51:13doesn't eat your codeex subscription. So
- 2:51:14that way instead of your usage limit
- 2:51:16looking like this, it will look more
- 2:51:18like this one down here, which is green
- 2:51:19and a lot more full. So what you'll need
- 2:51:21for this video is a CEX subscription.
- 2:51:23You'll need a trigger.dev account and a
- 2:51:24GitHub account. That's what we're going
- 2:51:26to be using. Okay. So there are three
- 2:51:27different types of automations that I'm
- 2:51:28going to show you guys today. And I'm
- 2:51:29also going to explain the difference. So
- 2:51:31it's going to be a scheduled automation.
- 2:51:32We're going to have a web hook
- 2:51:34automation. And then we're going to have
- 2:51:35a codeex SDK automation. So let's start
- 2:51:38with just number one. This is a
- 2:51:39scheduled automation, which basically
- 2:51:41just means we're going to have
- 2:51:42trigger.dev dev execute our code that we
- 2:51:45write on a schedule every 30 minutes
- 2:51:476:00 a.m. on Mondays however we want to
- 2:51:49set up that schedule very similar to the
- 2:51:51way that the schedule tasks inside of
- 2:51:53codeex actually work. So the first step
- 2:51:55of planning out the automation is the
- 2:51:56planning. It means we just want to kind
- 2:51:58of brainstorm with um Astra in this case
- 2:52:00to help us make sure that it fully
- 2:52:02understands what we want built and to
- 2:52:04make sure that we're not, you know,
- 2:52:06missing anything. We're not we're
- 2:52:08thinking through all the different
- 2:52:09scenarios. And yes, this example is
- 2:52:10going to be very simple, but it's just
- 2:52:12to show you how this actually works. So,
- 2:52:14I want to build an automation, and I'm
- 2:52:16going to put this on trigger.dev to
- 2:52:18actually be hosted there. But you need
- 2:52:19to help me plan this out and then
- 2:52:20actually build it. So, the first step is
- 2:52:22just the planning. I want this to be a
- 2:52:24morning routine automation. So, this is
- 2:52:25going to go off, let's say, 6 a.m., and
- 2:52:28it's going to look at my calendar for
- 2:52:29the day, and it's going to just give me
- 2:52:30a brief. So, it will have to check my
- 2:52:32Google calendar and it will just look at
- 2:52:34the events I have and then if there's
- 2:52:35any additional research or anything like
- 2:52:37that that it could do to be more
- 2:52:39helpful. It can just sort of help me
- 2:52:40prep for the day. Is there anything else
- 2:52:42important that I'm not thinking of and
- 2:52:43do you understand how this will actually
- 2:52:46be built out? And what's cool is it will
- 2:52:48actually be able to use some of the
- 2:52:50context that's already in your projects
- 2:52:52if you are working inside of a project
- 2:52:53because you can see the first thing it
- 2:52:54said is I'm going to check your existing
- 2:52:56morning routine workflow so that I can
- 2:52:58actually understand a little bit better
- 2:52:59how you want this built. So, trigger.dev
- 2:53:01hosts something called TypeScript, which
- 2:53:03is just basically a coding language. If
- 2:53:05you used something like modal, then you
- 2:53:07would be writing it in Python. So, don't
- 2:53:08worry about this word. It just is
- 2:53:10basically the language of the code that
- 2:53:11we're writing, but don't worry about it.
- 2:53:13So, it's explaining now how is this
- 2:53:14going to work. So, the morning trigger
- 2:53:16would go off at 6:00 a.m. America
- 2:53:17Chicago time. And then we would look at
- 2:53:19Google calendar and we would read
- 2:53:20today's events. We would read their
- 2:53:21descriptions and their times and their
- 2:53:23attendees and locations. And then it
- 2:53:24would decide whether the events need
- 2:53:26prep. So, if it's with someone new, we
- 2:53:28have this stuff that could be helpful.
- 2:53:29if it's a podcast, we could have this
- 2:53:30stuff. So, it's going to do a little bit
- 2:53:31of AI reasoning here, and then it's
- 2:53:33going to use AI once again to write the
- 2:53:35short brief. And then after that, it
- 2:53:37will send it to me, but I didn't
- 2:53:39actually tell it where. So, it said,
- 2:53:40"Where do you want this to be sent? Um,
- 2:53:42what's the timing? What's the calendar
- 2:53:43scope?" It's asking me some questions
- 2:53:45here. So, all of these are good
- 2:53:46thoughts. I want this to actually be
- 2:53:48sent to me in ClickUp. I want this to be
- 2:53:50sent to me as a DM to Nate Herk, so to
- 2:53:53myself and ClickUp. And you will use the
- 2:53:55Upit AI account to actually send this.
- 2:53:58As far as timing, we want this to be
- 2:53:596:00 a.m. And let's just have it be on
- 2:54:01weekdays. It doesn't really matter to me
- 2:54:03how long it takes. If it if I get it at
- 2:54:056:05 or 6:10, that's fine. The calendar
- 2:54:07scope, we're going to use my main work
- 2:54:09calendar. And research boundaries, let's
- 2:54:11just do public web research for now.
- 2:54:14Later, we could always work in other
- 2:54:16things like meeting notes and clickup
- 2:54:17and Google Drive and things like that.
- 2:54:19As far as the research, there's really
- 2:54:20no reason why you should be spending
- 2:54:21more than, you know, 25 cents for
- 2:54:23research at this point, but we can
- 2:54:25increase that later. Right now, let's
- 2:54:27just assume that the calendar is locked
- 2:54:28at 6 a.m. So, don't worry about checking
- 2:54:30in later. So, now that I've given it
- 2:54:31some more information, I will go ahead
- 2:54:32and send that off as well. Now, what I
- 2:54:35wanted you guys to notice here is
- 2:54:37figuring out if the automation is one of
- 2:54:39these two things, which are it's either
- 2:54:42a deterministic automation or it's a
- 2:54:45non-deterministic
- 2:54:47automation. And so, what this means is
- 2:54:49basically is it a predictable process or
- 2:54:51is it unpredictable? And usually the
- 2:54:54unpredictable automations when you get
- 2:54:55into this territory over here, this is
- 2:54:57when we really need to go for, you know,
- 2:54:59a full agentic harness, a full codeex
- 2:55:01routine because these are very
- 2:55:03noneterministic. They're unpredictable.
- 2:55:04And in this specific example, this was
- 2:55:06pretty simple. We basically have, you
- 2:55:08know, a 6 a.m. trigger. So we have 6
- 2:55:11a.m. it goes off. Then what happens next
- 2:55:13is we will read calendar. And so that is
- 2:55:17the next step. And so far there's not
- 2:55:19even any AI involved. where we start to
- 2:55:21get AI is in this next step when we sort
- 2:55:23of do like the decision-m on do we need
- 2:55:25research and you know what would we do
- 2:55:28so this is research and this would be an
- 2:55:30AI step so I would fill in this with
- 2:55:32green just to indicate that this is in
- 2:55:33fact AI the other thing that we would
- 2:55:35need AI for is drafting the message
- 2:55:37because you can't really draft a message
- 2:55:39um with code unless you were just using
- 2:55:42explicit placeholders and it's not as
- 2:55:43flexible and then the final step is to
- 2:55:45send that message to ClickUp so what I
- 2:55:48wanted you guys to see here is this is a
- 2:55:49very deterministic automation because it
- 2:55:51goes 1 2 3 4 5 oops five and it happens
- 2:55:56in that order every single time. So that
- 2:55:58is a deterministic flow. What's not
- 2:56:00deterministic in here are these specific
- 2:56:02AI steps. So step number three and step
- 2:56:05number four are nondeterministic because
- 2:56:07the research is going to be different
- 2:56:08every time based on the input that goes
- 2:56:10in. And then the draft message is going
- 2:56:12to be nondeterministic because it's
- 2:56:13different every time based on the
- 2:56:14research that goes in. So these are
- 2:56:16important things to think about because
- 2:56:17the more deterministic your automation
- 2:56:19gets, the more of a waste it would be to
- 2:56:21use codeex to host that. There's no
- 2:56:23reason to use the full Astra agentic
- 2:56:26loop inside of codeex if your automation
- 2:56:28looks like this. 90% of business use
- 2:56:30cases or personal productivity use cases
- 2:56:32are more like this. They're more of the
- 2:56:33deterministic automations that help you
- 2:56:35out and really save time. So anyways,
- 2:56:37just wanted to call that out. Hopefully
- 2:56:39you're learning some new stuff in this
- 2:56:40video. All right guys, real quick break
- 2:56:42for a message from today's sponsor,
- 2:56:44Hyper Agent. Now, I know that a lot of
- 2:56:45you guys are running AI agencies and a
- 2:56:48very tedious part is teaching the same
- 2:56:49thing over and over again to every new
- 2:56:52client's agents because every client
- 2:56:54wants it done a certain way. So, Hyper
- 2:56:56Agent handles that with skills and
- 2:56:58memories because you can teach an agent
- 2:57:00a skill once like your due diligence
- 2:57:02framework or your press release
- 2:57:03framework or format or you know some API
- 2:57:06that you always end up wiring in and it
- 2:57:08will just keep it in its memory and then
- 2:57:11it will build up memories as it works on
- 2:57:12things like the client's preferences,
- 2:57:14their data sources and how they want
- 2:57:16things formatted. And then you can watch
- 2:57:18the whole fleet from a command center
- 2:57:19that runs eval rubrics and it can do AB
- 2:57:22tests so that you can see which agents
- 2:57:23are actually getting better and what
- 2:57:25each client costs you to actually run.
- 2:57:27They can also sit in your Slack channels
- 2:57:29and they can jump in when something
- 2:57:30needs doing instead of just waiting
- 2:57:32around for you to ask. So if that sounds
- 2:57:33like something that could be useful for
- 2:57:35your agency, then check out Hyper Agent
- 2:57:36through the link in the description and
- 2:57:38you'll get some free credits to start.
- 2:57:40So thanks again to Hyper Agent for
- 2:57:41sponsoring this part of the video. But
- 2:57:42now let's get back to it. So anyways,
- 2:57:44Codex comes back and says, "Okay, cool.
- 2:57:45That gives us a clear plan. Here's
- 2:57:47exactly what we will do. Here's the
- 2:57:48budget. Here's the research. Here's all
- 2:57:50the stuff that's how it's going to work.
- 2:57:51And now, let's actually go ahead and
- 2:57:52start building it. It also calls out
- 2:57:54this hosting detail, which is that
- 2:57:55Trigger Dev's current free plan
- 2:57:57scheduling window can delay a 6 a.m.
- 2:57:59start by up to an hour. So, if we really
- 2:58:01need this right at 6 a.m., you might
- 2:58:02have to upgrade your Trigger.dev plan.
- 2:58:04But, as far as getting started, you can
- 2:58:06start on the free plan. All right. So,
- 2:58:07sounds like you have a good idea of how
- 2:58:08this works. What do you need from me?
- 2:58:10Are there any different API keys that I
- 2:58:12need to get, or do you already have
- 2:58:13access to all of these keys? So that's
- 2:58:15really the next step to be thinking
- 2:58:16about because if you have been building
- 2:58:17inside of codeex and you're using a
- 2:58:19bunch of these plugins, so you've
- 2:58:20connected to all of your accounts inside
- 2:58:22of these codeex plugins, those will not
- 2:58:24transfer over to trigger.dev, which is
- 2:58:27why I really like to use plugins not
- 2:58:29very much. I like to use thev file and I
- 2:58:32like to put all my API keys in there so
- 2:58:34that if I ever need to move them over to
- 2:58:35cloud code or if I need to move them
- 2:58:37over to trigger.dev, I already have them
- 2:58:39in one spot and I can just copy and
- 2:58:40paste them. And I will show you guys how
- 2:58:41that works later. But yeah, now that
- 2:58:43we've done sort of the hard part, the
- 2:58:44heavy lifting of actually planning out
- 2:58:45what we want, the implementation is the
- 2:58:47easy part because now we let Codeex
- 2:58:49actually go ahead and build this out for
- 2:58:50us because as you can see, it says,
- 2:58:52okay, it has access to Google Calendar,
- 2:58:54ClickUp, OpenAI, Perplexity, and we will
- 2:58:57obviously set up trigger.dev, but
- 2:58:58basically everything we need to move
- 2:59:00this over once Codex has built it is
- 2:59:02already here. If you don't yet have your
- 2:59:04API keys for Google Calendar or ClickUp,
- 2:59:06you would just say, "Hey, can you help
- 2:59:07me set that up with Google Calendar,
- 2:59:08with ClickUp?" and it will tell you
- 2:59:10exactly where to go and what to click on
- 2:59:12and what to get. So getting API keys and
- 2:59:14getting that set up is not really
- 2:59:15technical. It's not really a technical
- 2:59:16lift anymore. It used to be a little
- 2:59:17more intimidating in the old days, but
- 2:59:20now it's so easy. So if you don't have
- 2:59:21API keys, just ask Codex to help you
- 2:59:23find where those are and just go grab
- 2:59:25them. But now we're ready to start
- 2:59:26building. So I'm just going to tell it,
- 2:59:27yep, go ahead and start building. And
- 2:59:28I'll check in with you guys when we're
- 2:59:29ready to move this over to trigger.dev.
- 2:59:31But the one thing I'll say is before we
- 2:59:32move it into trigger.dev, we want to
- 2:59:34have Codex do as much verification on it
- 2:59:36as possible. So, if it's able to test it
- 2:59:38out and basically fix it before we move
- 2:59:40it over, then that's good. You know,
- 2:59:41we've talked a lot about giving agents a
- 2:59:43way to verify their own work and telling
- 2:59:44them to verify their own work so that
- 2:59:46you're not getting their first attempt.
- 2:59:47So, go ahead and build this out for me.
- 2:59:49Make sure you feel confident in how it
- 2:59:50works and you've validated that it will
- 2:59:52work before you actually tell me we're
- 2:59:54going to move it into trigger.dev and
- 2:59:55then once you're done, we will move
- 2:59:57everything over and host it. Now, you
- 2:59:58can see what it's doing is it's building
- 3:00:00it out and it's testing it. You can see
- 3:00:01it's using test-driven development. It
- 3:00:03is sending DMs. It is looking at how
- 3:00:06much all this costs and it's basically
- 3:00:07optimizing for what I actually said. It
- 3:00:09even made sure that the DM was the
- 3:00:11correct conversation. And the reason why
- 3:00:12this whole deterministic or
- 3:00:13nondeterministic thing is important to
- 3:00:15understand is because yes, you could
- 3:00:16easily do this as a routine inside of
- 3:00:19codeex, but I actually did this before
- 3:00:21where I had a routine that was very
- 3:00:22simple like this and after about a
- 3:00:25month, it started to just go rogue. It
- 3:00:26just randomly started to send to
- 3:00:27different DM channels and it started to
- 3:00:29send to like the team channel and stuff
- 3:00:31like that. And the reason for that is
- 3:00:32because it was an agent on the back end.
- 3:00:33the agent was just interpreting the
- 3:00:34message different every time and it just
- 3:00:36acted differently. But what we're
- 3:00:38actually building here is we're building
- 3:00:39a script so it doesn't have the option
- 3:00:41to act differently. We're literally
- 3:00:42saying, "Hey, this is what runs and it's
- 3:00:44goes the same way every time." So in
- 3:00:46this code, we're basically like hard-
- 3:00:48coding the direct channel with the
- 3:00:50channel ID and the message ID and
- 3:00:52everything that gets sent rather than
- 3:00:54giving the agent the full connection to
- 3:00:56ClickUp and saying, "Hey, just send it
- 3:00:57to Nate." Anyways, this has just
- 3:00:59finished up. So you can see, whoops,
- 3:01:00that it was built and validated locally.
- 3:01:02And we should have a test brief in
- 3:01:04ClickUp. So let me check that real
- 3:01:06quick. Awesome. So you can see right
- 3:01:07here that I do have this DM where I see
- 3:01:09my Tuesday morning briefing. I have my
- 3:01:11day right here. Every single one of
- 3:01:12these things has a link to my calendar
- 3:01:13as well. If I click on this, for
- 3:01:15example, it pulls up the actual event in
- 3:01:17my calendar. And then it also comes back
- 3:01:19with, hey, here's some useful prep. And
- 3:01:21here are some things to notice. So we
- 3:01:23know that the ClickUp connection works.
- 3:01:24We know that the AI is working on the
- 3:01:26background. And we know that the Google
- 3:01:28calendar connection is working as well.
- 3:01:30And it says that everything has been
- 3:01:31built and validated. And this costed
- 3:01:33about 1.33 cents. So us giving it the
- 3:01:36gate of, you know, maxing each run at 25
- 3:01:38cents seems pretty realistic. Anyways,
- 3:01:40now we have to deploy to trigger.dev and
- 3:01:42validate one run there before we turn on
- 3:01:44the schedule. And the cool thing about
- 3:01:46this is yes, you're going to be able to
- 3:01:47follow this tutorial so you know exactly
- 3:01:48how to connect to trigger.dev, but if
- 3:01:50you didn't, so how exactly do I get this
- 3:01:52automation from the code that you built
- 3:01:55into trigger.dev? And while it's doing
- 3:01:56this, let's go ahead and make sure we
- 3:01:57have a trigger. So, head over to
- 3:01:59trigger.dev. Go ahead and create an
- 3:02:01account. Like I said, you can get
- 3:02:02started for free. Now that I'm in here,
- 3:02:03you can see I'm on a free plan in this
- 3:02:05account and I just created a new project
- 3:02:06up here called YouTube. And this is what
- 3:02:08it will look like in order for you to
- 3:02:10sort of like get everything set up.
- 3:02:11There's nothing yet in here. So, this is
- 3:02:13how it gets really cool because we're
- 3:02:14able to basically just connect codeex to
- 3:02:16trigger.dev and it can basically talk to
- 3:02:19trigger.dev in order to do all of this.
- 3:02:21Now, you can also have it connect to
- 3:02:22GitHub, which is something that I do
- 3:02:24recommend because that way GitHub
- 3:02:26connects to trigger.dev and the three of
- 3:02:27them talk together. And the reason why
- 3:02:29we like to use GitHub is because then
- 3:02:30you can have like different version
- 3:02:31control and you can have other people
- 3:02:33more easily contribute to the automation
- 3:02:35if you need someone else to change it or
- 3:02:37something like that. And Codex can
- 3:02:38connect to trigger.dev and GitHub via
- 3:02:40the command line tool. So, that's what I
- 3:02:42would probably say next is let's get
- 3:02:44connected to GitHub and trigger.dev dev
- 3:02:46via the command line and then we can
- 3:02:49just push our code to GitHub and then
- 3:02:51trigger.dev will sync with that code.
- 3:02:53And what this will do is it will use the
- 3:02:54CLI and it will prompt you to just
- 3:02:56authenticate in. So it will do one of
- 3:02:57those things where it opens up the
- 3:02:59browser, you sign in with GitHub, you
- 3:03:01sign in with trigger.dev, and then you
- 3:03:02come back into Codex and it says boom,
- 3:03:04got it, I'm connected. So here is the
- 3:03:06trigger.dev authorization. I just have
- 3:03:09to go ahead and authorize right here.
- 3:03:10And then it says return to your terminal
- 3:03:12to continue. So Codex should have gotten
- 3:03:13that now. And you can see that because
- 3:03:15my codeex is already authenticated to
- 3:03:17GitHub, I didn't have to do that. But it
- 3:03:18would be the same exact flow if you
- 3:03:20haven't done that yet. So now I want you
- 3:03:22to create a new private repo for this
- 3:03:24automation in my GitHub account. And I'm
- 3:03:27just going to shoot off this message to
- 3:03:28steer codeex in the right direction now
- 3:03:30that we know we're connected to these
- 3:03:31two tools. And you can see that it says
- 3:03:32the API keys, the calendar snapshots,
- 3:03:35and anything sensitive are excluded from
- 3:03:37Git. Even though we're creating it as a
- 3:03:38private repo, it gets completely
- 3:03:40excluded. And that's the whole point of
- 3:03:41the env. So, what we'll have to do in
- 3:03:44trigger.dev is we will have to manually
- 3:03:46move those over. And look at this.
- 3:03:47Because Codeex's browser use is so good,
- 3:03:49it also pulled up my trigger. account in
- 3:03:51this browser. So, if there's anything
- 3:03:53that you're confused about navigating
- 3:03:54the trigger interface, you can have it
- 3:03:56help you out by just clicking around
- 3:03:57right in there. But, it says that we did
- 3:03:59create this GitHub repo for the morning
- 3:04:01brief. So, basically, this repo, just
- 3:04:03think of it like a Google Drive. It
- 3:04:05holds the actual code and the rules for
- 3:04:06how this automation runs. And
- 3:04:08trigger.dev will grab this and actually
- 3:04:09host it. So, it's super simple. But also
- 3:04:12if you make different versions you will
- 3:04:13update it here. So you could roll back
- 3:04:15to previous versions if you want and
- 3:04:17that way if a team member later wants to
- 3:04:18take over this automation all of the
- 3:04:20details are here and they can contribute
- 3:04:21to this project. Okay. So take a look at
- 3:04:23this. It actually created its own
- 3:04:25project in my trigger. So it said in
- 3:04:26morning brief and settings and git is
- 3:04:29the GitHub repo connected. So basically
- 3:04:31what that's saying is okay it got into
- 3:04:33my trigger.dev. You guys saw me create a
- 3:04:35new project called YouTube but it
- 3:04:36decided okay I'm going to create my own
- 3:04:38project. So it made one called morning
- 3:04:39brief. And now what we need to do is we
- 3:04:41need to see if this trigger.dev project
- 3:04:43is connected to our actual GitHub
- 3:04:46codebase called morning brief so that
- 3:04:47they can sync and talk to each other. So
- 3:04:49that's what it's asking me to verify. It
- 3:04:51said to go to the settings in here, but
- 3:04:52you can see that it says GitHub isn't
- 3:04:54connected. So it's actually not syncing
- 3:04:55to anything at all. So what I need to do
- 3:04:57maybe is install the GitHub app. But
- 3:04:59let's see. I'm just going to say I don't
- 3:05:01see any deployments. I don't see any
- 3:05:03integrations. When I go into
- 3:05:05trigger.dev, it tells me that GitHub
- 3:05:07isn't yet connected. So that's what I
- 3:05:09see. But this is really cool, guys. You
- 3:05:11can see how proactive this thing is. It
- 3:05:13actually just went ahead and started
- 3:05:14doing all of the connections rather than
- 3:05:16telling me what to do, which is pretty
- 3:05:17cool. But every once in a while, it hits
- 3:05:19a snag and it will just tell you, hey,
- 3:05:20can you check this? Can you check this?
- 3:05:21But hopefully with this context, it's
- 3:05:23going to be able to help out with that
- 3:05:25now. And now you can see it's going to
- 3:05:26open it up to actually do it itself. So
- 3:05:28you can see the mouse right here moving.
- 3:05:30This isn't me. This is Codeex using its
- 3:05:32browser use to check on all this for us.
- 3:05:34Okay. So you can see that this says
- 3:05:35connected and verified and morning
- 3:05:37delivery remains disabled until the full
- 3:05:39hosted delivery test passes. So that's
- 3:05:40good. And it actually moved it back
- 3:05:42under the personal account where the
- 3:05:44GitHub connection already exists and
- 3:05:46ClickUp still uses UPAI. Now let's go in
- 3:05:49there real quick so I can show you guys
- 3:05:50something. If I go back into the other
- 3:05:52project and it actually did it in a
- 3:05:54different org. So if I go back to my
- 3:05:55personal org, this is where we should
- 3:05:57see the morning brief. There we go. So I
- 3:05:59was just looking at the completely wrong
- 3:06:01one. And if I go to tasks, this is where
- 3:06:02we can actually see the interface of our
- 3:06:05weekday runs. And what's cool is it did
- 3:06:07two different tasks. It did the
- 3:06:09scheduled one. So this one you can see
- 3:06:10is the one that goes off at 6 a.m.
- 3:06:12Monday through Friday. And then what
- 3:06:14actually happens when this runs is it
- 3:06:16calls this other I guess runner. So it
- 3:06:19calls this and this is the one that has
- 3:06:21the actual process in it. So it does it
- 3:06:24in two parts. And that's pretty cool
- 3:06:25because as you create these longer, more
- 3:06:27complex workflows, they call individual
- 3:06:29tasks and pieces that codeex will build
- 3:06:31out for you and it just helps with some
- 3:06:33visibility. So anyways, what we can do
- 3:06:34here is if I go into the schedule and I
- 3:06:36hit test schedule, this basically starts
- 3:06:38running this. So I'll hit run test and
- 3:06:40now this runs in production. And then
- 3:06:42what you can see is that it actually
- 3:06:44calls this other little task. And now
- 3:06:46this thing is running. It's showing us
- 3:06:47how long this is taking. It's going to
- 3:06:49show us all these little different
- 3:06:50details so that if this errors, we could
- 3:06:52copy all this data, give it to Codex and
- 3:06:54say, "Hey, this is what happened. Help
- 3:06:56me fix this." So, you can see that this
- 3:06:57already finished and we can look through
- 3:06:59the payload. We can look at the status
- 3:07:02which came through as disabled. And
- 3:07:03you'll notice that I didn't get a new
- 3:07:05DM. So, we have to figure out why this
- 3:07:06didn't seem to work. My first thought
- 3:07:08would be, okay, well, do we have all of
- 3:07:09our API keys and all of our secrets in
- 3:07:11here? And if I go down to deployments,
- 3:07:13this is where I can click on environment
- 3:07:14variables. And you can see that we have
- 3:07:16a bunch of stuff in here that got added
- 3:07:18today. Today's September 15th. So Codeex
- 3:07:20actually took our ENV and via the CLI
- 3:07:23put everything in here, which is
- 3:07:24awesome. So we should have our ClickUp
- 3:07:26ids. We have our workspace ID, client
- 3:07:29ID, all of this stuff should be working
- 3:07:31in here. So what I would do is I would
- 3:07:32say, okay, so I just ran a test run
- 3:07:35inside of trigger.dev, but I didn't get
- 3:07:37a ClickUp message. I didn't get
- 3:07:40anything. Can you just check and see
- 3:07:42what happened there? So you guys
- 3:07:43remember how it came through and said
- 3:07:44status equals disabled. It's because it
- 3:07:46had morning brief enabled as false. So
- 3:07:49if I go back into our trigger.dev and I
- 3:07:51go to our environment variables, you can
- 3:07:53see right here, morning brief, it says
- 3:07:54false. So we would basically have to
- 3:07:56come in here, edit this, and we would
- 3:07:58have to change this to true. Hit save.
- 3:08:00And now if we go back, let's see if this
- 3:08:02fixes the issue. So I'm going to go into
- 3:08:04the morning brief weekdays and hit test
- 3:08:06and go ahead and run that. And there we
- 3:08:08go. This looks a little bit different
- 3:08:09now because you can see it called on our
- 3:08:10different runner. We can see what's
- 3:08:12actually going on. And we can see this
- 3:08:13thing actually in real time sort of like
- 3:08:15playing out. We got a little bit of a
- 3:08:17green message here. We get output equals
- 3:08:19delivered. Let's open up my ClickUp.
- 3:08:21See, was this a new run? Okay. Well,
- 3:08:24unfortunately, this still doesn't look
- 3:08:25like it was a new run. So, once again,
- 3:08:26we have to go figure out why this didn't
- 3:08:28work yet. Because if you guys remember,
- 3:08:29it worked when Codex was testing it
- 3:08:31locally, but it didn't work yet in
- 3:08:33production. So, okay, that run seemed to
- 3:08:36work. It showed that we actually got the
- 3:08:38output, but I didn't get any DM in
- 3:08:41ClickUp. So, take a look at that run and
- 3:08:44help me figure out how do we fix this.
- 3:08:45Okay, that's really interesting. What
- 3:08:47happened was the run found today's
- 3:08:49earlier test message. So, it skipped
- 3:08:51sending a duplicate because it would
- 3:08:52have been essentially the exact same
- 3:08:53message. So, it seems like that run
- 3:08:55didn't actually fail. It was more so
- 3:08:57done by design. You can see the existing
- 3:08:59message was sent at 12:01 and we just
- 3:09:01tried to run a new one at 12:33, so it
- 3:09:03didn't get sent. That's why it's always
- 3:09:04important that when you're having AI
- 3:09:06build you automations and build you
- 3:09:07code, you don't have to know what every
- 3:09:09single line is doing. But you do have to
- 3:09:10test the stuff out to figure out how it
- 3:09:12works and why it does what it does. I
- 3:09:14mean, look how powerful this is. It's
- 3:09:15basically able to deploy everything,
- 3:09:17debug everything, move around in the
- 3:09:19interface, and then just test it all for
- 3:09:20me. And we're just using our natural
- 3:09:23language, and we're using our intent to
- 3:09:26drive all of this. This is so much
- 3:09:27quicker than the way I used to have to
- 3:09:28build automations. it worked in this
- 3:09:30variable that shows that if it was
- 3:09:32already delivered and it got triggered
- 3:09:33twice for some reason to not send a
- 3:09:35duplicate message. So duplicate
- 3:09:37prevention remains intact. I think it
- 3:09:39did a really good job there. But I do
- 3:09:40want to prove to you guys this works. So
- 3:09:41I'm telling it to override that just so
- 3:09:43we can see that it works as expected.
- 3:09:45And then we'll move on to the next
- 3:09:46automation. There we go. We can now see
- 3:09:48that we got the hosted proof test for
- 3:09:50the morning brief. And now we can feel
- 3:09:52better about turning this thing on
- 3:09:54inside of trigger.dev and knowing that
- 3:09:55it will actually have all the
- 3:09:56connections set up because we've seen it
- 3:09:58in here. We've tested it. We know that
- 3:10:00all of the environment variables are
- 3:10:01also moved over. So, we are all good to
- 3:10:03go. All right. So, we took care of the
- 3:10:05scheduled type of automation. Let's look
- 3:10:07at one that is triggered by something.
- 3:10:09So, typically this is triggered by a web
- 3:10:11hook. So, we're going to go back into
- 3:10:13Codex and we're just going to start
- 3:10:14having it do two things for us. It needs
- 3:10:16to build sort of like a very simple
- 3:10:17front end that will actually be our
- 3:10:19trigger for this specific example. And
- 3:10:21then we'll build the back end, too. So,
- 3:10:22I'm actually going to try to do this in
- 3:10:23one fell swoop. I'm going to do a slash
- 3:10:25goal. And here we go. So now I want to
- 3:10:29build a web hook triggered automation
- 3:10:32inside of trigger.dev. So this is kind
- 3:10:34of a two-parter. The first part is we
- 3:10:36need to use or sorry I need you to help
- 3:10:38me build just a very simple local host
- 3:10:40that is sort of like a form submission
- 3:10:42that you might see on a website.
- 3:10:43Collects information like the name, the
- 3:10:45email and um team size and what they're
- 3:10:48looking for. So, a simple landing page,
- 3:10:51a simple form that we can fill out, and
- 3:10:52then when the user hits submit on that,
- 3:10:54that's what I want to trigger the actual
- 3:10:57automation on the back end that we're
- 3:10:58going to build and put in trigger.dev.
- 3:11:00So, basically what I want is for um that
- 3:11:02automation to just send me a ClickUp DM.
- 3:11:06So, same DM as before, and I want this
- 3:11:08one to just say, "Hey, you got a new
- 3:11:09form submission. It's for this person.
- 3:11:10Here's what they asked for, and here is
- 3:11:12how I recommend you reach out to them."
- 3:11:14So, sort of like just creating a draft.
- 3:11:16So, I'm going to shoot that off and it
- 3:11:17might have some questions for us. But
- 3:11:18this is a very simple use case just to
- 3:11:20show you how it works with a web hook.
- 3:11:22But this is still a very much a
- 3:11:24deterministic automation. Even though it
- 3:11:26is AI, it is still deterministic. So,
- 3:11:28let me show you real quick what this one
- 3:11:29looks like. This one is a web hook
- 3:11:31trigger. So, this is just going to be a
- 3:11:32form submission. That is basically what
- 3:11:34kicks this thing off over here. From the
- 3:11:36form submission, what happens is it's
- 3:11:37going to actually read the form and it's
- 3:11:40basically just going to have to do one
- 3:11:41thing, which is create the actual um
- 3:11:44message that will go to us in ClickUp.
- 3:11:45and then it will basically send it to us
- 3:11:48in ClickUp. So, it takes a very similar
- 3:11:49shape as to what we've already seen,
- 3:11:51except for instead of going off on a
- 3:11:53schedule at 6:00 a.m., it goes off based
- 3:11:55on the actual action which is triggered
- 3:11:58by us. And then once again, the only
- 3:11:59time we actually see AI inside of this
- 3:12:01process is just here. Now, you could
- 3:12:02actually do this with no AI. If you
- 3:12:04wanted the form submission to just shoot
- 3:12:06you a message in ClickUp, you could do
- 3:12:08it with placeholders and you could do
- 3:12:09that. you'd save yourself some money and
- 3:12:11some time because there'd be no AI
- 3:12:12message or sorry, there'd be no AI step
- 3:12:14and that would be 100% deterministic
- 3:12:16every single time. Now, obviously, there
- 3:12:17are some edge cases you might want to
- 3:12:19think of. You know, how do you build the
- 3:12:20form so that they can't, you know, spam
- 3:12:22it a thousand times in a second? How do
- 3:12:24you build the form so that if you're
- 3:12:25expecting an email field, you're
- 3:12:26actually going to get only email fields?
- 3:12:28There's a lot of other edge cases to
- 3:12:29think through. But what's cool about
- 3:12:30that is after you build this, you could
- 3:12:32say, "Okay, cool. Now, spin up 50
- 3:12:34different sub aents and have all of them
- 3:12:35test this thing, try to break it, and
- 3:12:37then let me know what you find." and you
- 3:12:39can really stress test and QA your own
- 3:12:41automations before you actually have any
- 3:12:43sort of human or customer find these
- 3:12:45bugs or anything like that. So, that's
- 3:12:46pretty cool. It's going to spin up a
- 3:12:49local page. It's going to use the
- 3:12:50existing trigger.dev deployment on the
- 3:12:52back end. And now, because we've already
- 3:12:53set up the connection with like GitHub
- 3:12:55and trigger.dev, all of the automations
- 3:12:58we want to build in the future are going
- 3:12:59to be much much easier. So, I'm
- 3:13:01basically just going to check in with
- 3:13:01you guys when this one is done and show
- 3:13:02you how that works. Okay, so it says
- 3:13:04that this is done. You can see that we
- 3:13:06have this form here to fill out. It says
- 3:13:08that it has built and verified this end
- 3:13:10to end. So submitting it now will start
- 3:13:12the hosted trigger. Task. It will
- 3:13:14generate the outreach recommendation and
- 3:13:15it will send everything to ClickUp. So
- 3:13:17let's go ahead and real quick check in
- 3:13:19trigger.dev. We do have this new one
- 3:13:21right here. I'm going to refresh just in
- 3:13:22case. And you can see that we have this
- 3:13:23new one called form submission. It has
- 3:13:26already been triggered a few times. It
- 3:13:27looks like actually let me just zoom out
- 3:13:29so that we can see this better. It has
- 3:13:31been triggered five times. So this was
- 3:13:32Codeex testing it. And now let us go
- 3:13:34ahead and test it ourselves. So, I'm
- 3:13:36going to go back into Codex. I'm going
- 3:13:37to fill out this form. So, we're just
- 3:13:38going to say Alex Morgan, Alex
- 3:13:42company.com.
- 3:13:44Team size, we will just put 20 plus. We
- 3:13:47are spending so much time on, you know,
- 3:13:51making hamburgers and it's just becoming
- 3:13:53a real bottleneck. So, I'm looking to
- 3:13:54see if we can automate that. And let's
- 3:13:55go ahead and shoot that off. What do we
- 3:13:58get on this front end? Your request is
- 3:13:59in. Let's go back to trigger.dev. Let's
- 3:14:01see if we get some sort of run. I'll go
- 3:14:03back to the tasks. We should see it
- 3:14:05looks like there's a new one executing
- 3:14:06right now. If I click into this run, we
- 3:14:09should see that all of this is actually
- 3:14:10going on. So that proves that this web
- 3:14:12hook has been set up correctly. Now, in
- 3:14:14the old days, we would have had to set
- 3:14:15up the URL and kind of sync them
- 3:14:16together and like build out the payload.
- 3:14:18But we don't have to do that at all and
- 3:14:20it this is the payload basically like
- 3:14:21the information that comes through and
- 3:14:22now we have the output. Let me check and
- 3:14:24see if we got this in ClickUp. Sweet.
- 3:14:26Okay, so we actually got a lot of these.
- 3:14:27These are all of the demo examples that
- 3:14:29it tried, you know, the new form
- 3:14:31submissions. But this is the one that we
- 3:14:32just submitted. You can see right here,
- 3:14:33Alex Morgan. Here's the email. Here's
- 3:14:35the team size. Now, this took no AI.
- 3:14:38This is what I meant when I said that
- 3:14:40this could be just placeholders. So, if
- 3:14:41you wanted just to just get this
- 3:14:43notification, no AI needed. But here's
- 3:14:45where we needed AI, the recommended
- 3:14:46outreach and the email draft. And this
- 3:14:48is what had to be obviously generated
- 3:14:50with artificial intelligence. So, that's
- 3:14:52kind of the trade-off there. But, super
- 3:14:54simple, super easy. That took me no time
- 3:14:56to build. And I barely had to do
- 3:14:58anything as far as like wiring things up
- 3:15:00by hand. So, that is a web hook
- 3:15:01automation. That doesn't obviously have
- 3:15:02to be just a form submission. That can
- 3:15:04be a new record in the CRM. That can be
- 3:15:06a new email entered the inbox. There's
- 3:15:08so many different event-based triggers
- 3:15:09that you can have for a web hook style
- 3:15:12automation. By the way, guys, I've got
- 3:15:13this completely free SOP for you about
- 3:15:15getting your first aation client. It's
- 3:15:17going to go over the exact steps that
- 3:15:18has been proven for hundreds of our AIS
- 3:15:20Plus members to get their first paid
- 3:15:22gigs. It goes over the one sentence
- 3:15:24service pitch that can get you started
- 3:15:25today. Why your first client should cost
- 3:15:27you money. The five-minute video that
- 3:15:28answers, can this person actually
- 3:15:29deliver before you've actually received
- 3:15:31any money, what to do when you have zero
- 3:15:33case studies. There's so many good
- 3:15:35things in here that are going to help
- 3:15:36you out. Even if you already do have
- 3:15:38clients, I would recommend grabbing this
- 3:15:39because, like I said, it's yours
- 3:15:41completely free. So, if you want to grab
- 3:15:42this, there's a link for it down in the
- 3:15:43description. Let's get back to the
- 3:15:44video. All right. And for this last use
- 3:15:46case, now that we've done the web hook
- 3:15:47as well, this one's a little bit
- 3:15:49special, and this is when you kind of
- 3:15:50want that full agentic loop that you're
- 3:15:53used to getting inside of codeex, but
- 3:15:54now you're getting it programmatically,
- 3:15:56and that is through the codeex SDK. So,
- 3:15:59a really good opportunity to see if you
- 3:16:01want to actually use the Codex SDK is
- 3:16:04when you need to bring something that
- 3:16:06might be an already existing routine
- 3:16:09into something like trigger.dev. Now the
- 3:16:11thing about this is most of the time if
- 3:16:13you're running an automation in a
- 3:16:14scheduled task inside of codecs, I would
- 3:16:16recommend you keep these as a codec
- 3:16:18scheduled task because then it actually
- 3:16:20eats away at your subscription. But if
- 3:16:21you need to do this at scale and you
- 3:16:23can't use your subscription
- 3:16:24programmatically, then you probably want
- 3:16:26to build the automation with the Codex
- 3:16:27SDK and then host it on something like
- 3:16:28trigger.dev. So if that ever does become
- 3:16:31the case, you kind of get the full
- 3:16:33agentic loop. So here is what that might
- 3:16:35look like. We saw these first two
- 3:16:37examples. Now, what I'm going to do here
- 3:16:38is this is going to be a scheduled task.
- 3:16:40So, we're just going to call this
- 3:16:42actually let's just call this every 30
- 3:16:43minutes. And I actually do this for real
- 3:16:46inside of a a scheduled task inside of
- 3:16:47my codeex where 30 minutes it wakes up
- 3:16:50and we basically just have this full
- 3:16:51agent. And what it does is it basically
- 3:16:53it looks at my I'm just going to call
- 3:16:55this an agent because it does a lot of
- 3:16:56things. It looks at my alpaca account
- 3:17:00and actually let me just like go down
- 3:17:01here and make some tools. Just pretend
- 3:17:02that these are tools. So, it has to like
- 3:17:05check alpaca, which is where I have like
- 3:17:07some trading going on and my agent here
- 3:17:09helps me trade. It will also, you know,
- 3:17:12like do research. It can also just like
- 3:17:16look at like past logs and stuff like
- 3:17:19that. So, it's leaving logs for itself
- 3:17:21every time so that the next time the
- 3:17:22agent wakes up, it's not completely
- 3:17:24stateless and just like doesn't know
- 3:17:25what to do. So the reason why this is
- 3:17:28the SDK rather than just being more of
- 3:17:31like one of these deterministic AI
- 3:17:33automations is because we don't know how
- 3:17:35many times it might need to look at the
- 3:17:36logs and read things or look at the
- 3:17:38research or check back in in alpaca if
- 3:17:40it's doing research for 10 minutes and
- 3:17:41it's like wait let me just check the
- 3:17:42current holdings one more time because
- 3:17:44things may have shifted this it it has
- 3:17:46the ability to basically just go back
- 3:17:47and forth you know it can go like this
- 3:17:48and then this and then this and then
- 3:17:50this and then this and then this and
- 3:17:51then this and it's basically an agent
- 3:17:53and it has that luxury and then what
- 3:17:55happens is It essentially just sends an
- 3:17:57email with a recommendation. So, I'll
- 3:17:59just call this email. And it sends that
- 3:18:00to Grockbot. And then Grockbot actually
- 3:18:02places the trade because OpenAI models
- 3:18:04and cloud models don't actually let you
- 3:18:06trade anymore. They just won't do it. I
- 3:18:07think it's cuz all this security stuff
- 3:18:08is going on. But Grock still lets you
- 3:18:10trade. So, my Grockbot is the one
- 3:18:11placing the trades and Astra is doing
- 3:18:13all of this heavy lifting research. Now,
- 3:18:15yes, you could do this exact same sort
- 3:18:17of thing in a deterministic flow, but
- 3:18:20that would just look different. It would
- 3:18:21basically be a predetermined amount of
- 3:18:23research. So, for example, it could hit
- 3:18:26like a research step or maybe it hits,
- 3:18:30you know, two research steps. So, I'm
- 3:18:32just going to like sort of demonstrate
- 3:18:35this like this where maybe it does two
- 3:18:37different like research sources. So,
- 3:18:39maybe this one is firecrawl and this
- 3:18:40one's perplexity or something. And then
- 3:18:42it basically will send it to an agent to
- 3:18:45um look at the research and can create
- 3:18:47the draft, create the email and then
- 3:18:48send it off. But what if it looks
- 3:18:50through this and it's like, "Wait, I
- 3:18:52want to do some more research." Because
- 3:18:53now the AI agent has analyzed it and
- 3:18:55it's like, "Wait, let me do some more."
- 3:18:56We would have to work in somehow the
- 3:18:58ability for the agent to loop back and
- 3:19:00do more research. And that's essentially
- 3:19:01why we would just use something like
- 3:19:02this where we're now using the SDK to
- 3:19:05give us that sort of autonomy. So, let's
- 3:19:07go ahead and build that out real quick.
- 3:19:09Now, this would be a very similar
- 3:19:11process where I would like you to say,
- 3:19:12"Hey, I want you to interview me about
- 3:19:13this process. I want to get very clear
- 3:19:15on what this does." Especially the more
- 3:19:16autonomous your systems get, there's
- 3:19:18more room for error. There's more room
- 3:19:20for, you know, you're increasing the
- 3:19:21risk, you're increasing the cost, you're
- 3:19:23increasing complexity, you're increasing
- 3:19:24the maintenance, you're increasing a lot
- 3:19:25of things. So build the simplest
- 3:19:27solution possible and only move up the
- 3:19:29sort of AI systems pyramid as I call it
- 3:19:31when you truly need that functionality.
- 3:19:32So I would go through this whole process
- 3:19:34of interviewing to get what I want. But
- 3:19:36we have this luxury here of all of this
- 3:19:39is already set up because we have these
- 3:19:40routines going. You can see that I have
- 3:19:42this um thread called challenge thread
- 3:19:44where I'm doing this actual trading and
- 3:19:46you can see that every like 30 minutes
- 3:19:47or so it's basically just starting this
- 3:19:49routine and it's doing the whole agentic
- 3:19:51loop. We see that one is actually
- 3:19:52running like right now and this is the
- 3:19:54this is codeex. This is codeex working
- 3:19:56and we're basically just trying to move
- 3:19:57this over to trigger.dev. Now I'm not
- 3:19:59really wanting to I obviously want to
- 3:20:01keep that using my subscription but for
- 3:20:02the sake of the demo. So what I need you
- 3:20:03to do now is I want you to build me a
- 3:20:06codeex SDK automation and we're going to
- 3:20:09host this inside of trigger.dev. dev.
- 3:20:11Now, what I want to do is I basically
- 3:20:13want to transition the Astra trading
- 3:20:16challenge, that thread, I want to move
- 3:20:18that routine over to trigger.dev. So, we
- 3:20:21basically wake up like every 30 minutes
- 3:20:22during trading hours, one check before,
- 3:20:25one check after, and we do this research
- 3:20:27loop. We check Alpaca, we check the
- 3:20:28logs. So, familiarize yourself with that
- 3:20:30actual process, and then just turn this
- 3:20:32into a Codex SDK automation for me and
- 3:20:34host it in trigger.dev. Let me know if
- 3:20:36there's anything else that you need to
- 3:20:37build this. So once again, the hardest
- 3:20:39part is already done here, and now we're
- 3:20:41just going to get that built out. So
- 3:20:42I'll check in with you guys when this is
- 3:20:43all done. Okay, so this run took about
- 3:20:4515 minutes, a little over 15 minutes.
- 3:20:47And we can see now that this is ready
- 3:20:48for our review. It did 44 passes or 44
- 3:20:52tests and it paused the old desktop
- 3:20:55routine, which I'm going to say, you
- 3:20:56know, I don't want it to pause that.
- 3:20:58This was just a demo. So anyways, um it
- 3:21:00was being pretty proactive there,
- 3:21:01though. But anyways, what it did is it
- 3:21:02made a new project. So it created a full
- 3:21:04new one rather than what we were doing
- 3:21:06earlier. And I'm actually just going to
- 3:21:07open this up full screen in my desktop
- 3:21:10over here. Give this a quick refresh.
- 3:21:12You can see what it did is it basically
- 3:21:14made a few different like tasks or
- 3:21:16runners or whatever you want to call
- 3:21:17them. And they kind of call on each
- 3:21:19other. So you can see here there's the
- 3:21:20check. This ran 18 times. Um then we
- 3:21:23have an install plan. We have a
- 3:21:25pre-flight and we have a proof. So I'm
- 3:21:26not exactly sure the right order to run
- 3:21:28this in, but this is what it did. And if
- 3:21:30I go to the runs, we can see exactly
- 3:21:31what happened. So it basically runs, you
- 3:21:34know, these failed, these completed.
- 3:21:36We've got these checks. So what we're
- 3:21:38going to want to do is we're going to
- 3:21:38want to figure out what does each one of
- 3:21:40these do? So I'm going to go ask um
- 3:21:42Codex that real quick. So I can see that
- 3:21:44you actually created four different
- 3:21:46tasks. We've got a check, an install
- 3:21:48plan, a pre-flight, and a proof. Can you
- 3:21:49just explain to me what these are and
- 3:21:51how they work together and how these use
- 3:21:53the codeex SDK to give us sort of that
- 3:21:55full agentic loop that we're looking for
- 3:21:57here? Okay, so Astro check is the actual
- 3:21:59recurring worker. that is the one that
- 3:22:01should fire on the schedule. The other
- 3:22:03three help it do things. So the
- 3:22:05pre-flight checks the codeex runtime,
- 3:22:07alpaca connection, SIP market data,
- 3:22:10policy files, and the checkpoint
- 3:22:11storage. And it can optionally make a
- 3:22:13small model test call. So this really
- 3:22:15only needs to run during setup or
- 3:22:16troubleshooting. We have proof, which
- 3:22:18runs the full research and reporting
- 3:22:20workflow, including the the test label,
- 3:22:22email, and clickup delivery. It prevents
- 3:22:24actionable trade tickets during the
- 3:22:25test, and this will run before enabling
- 3:22:28the routine. currently disabled after
- 3:22:30verification, the install plan during
- 3:22:32setup, and then the check at each
- 3:22:34scheduled check. So, it's actually
- 3:22:35interesting. It made all of those to
- 3:22:36help it build and to check. But now, I'm
- 3:22:39not sure because each check now follows
- 3:22:41this process where we have it loads up
- 3:22:42the context. It asks codeex what needs
- 3:22:45investigation. It finds the evidence. It
- 3:22:47has Codex evaluate all that. And this is
- 3:22:49with the SDK, of course. It validates
- 3:22:51and delivers, and then it saves the
- 3:22:52handoff. So the agendic part is Astra
- 3:22:54choosing what to research using web
- 3:22:56tools, doing deeper evidence, and
- 3:22:58adjusting its conclusions based on the
- 3:22:59result, which is pretty interesting.
- 3:23:00This is the only one that will actually
- 3:23:02run now that this is all built. So I'm
- 3:23:04going to go back into here and we're
- 3:23:05going to go to Astra check and we're
- 3:23:06going to hit test and just test this
- 3:23:08thing out right here. Now look at this.
- 3:23:09It just called this attempt right here.
- 3:23:12And we open this up, we can see
- 3:23:13everything that's actually going on. Now
- 3:23:15this is pretty cool. The reason why this
- 3:23:16did nothing is it actually sent this
- 3:23:19payload over which was like the time and
- 3:23:21the date and then the output was hey
- 3:23:23this is early because this was set up by
- 3:23:25Astra in a way where it's supposed to
- 3:23:26check in during market hours and right
- 3:23:28now the market has already closed. So
- 3:23:30the next scheduled run isn't until
- 3:23:31tomorrow morning before the market
- 3:23:33opens. So that's why it basically fired
- 3:23:35and it was like oh you know this
- 3:23:36actually isn't right. I'm not supposed
- 3:23:37to run yet so I'm outputting the word
- 3:23:39early. But I do know that this works
- 3:23:41because if I go into my ClickUp thread
- 3:23:42where I'm actually getting these
- 3:23:43notifications every day from the actual
- 3:23:46routine, you can see that these started
- 3:23:47to come in. You can see we have the test
- 3:23:49migration here, 341, 347. We have
- 3:23:51another test migration for Nate's
- 3:23:53review. It's doing the research. It's
- 3:23:54giving us assessments. It's linking
- 3:23:56things. It's looking at our account.
- 3:23:57It's pulling in all the sources. It did
- 3:23:58another one at 352 for the market close.
- 3:24:01So, this is currently working. And what
- 3:24:02it's actually doing is it is messaging
- 3:24:04Grockbot. This is my trading Grockbot
- 3:24:06that actually places these, you know,
- 3:24:08trades based on getting an email. So
- 3:24:10everything transferred over to the point
- 3:24:12where it's getting these test migration
- 3:24:13reports as you can see because what's
- 3:24:15going on is trigger.dev automation with
- 3:24:17the Codex SDK like it said, it's
- 3:24:19checking, it's doing research, it's
- 3:24:21looking at the evidence, and then it's
- 3:24:22making recommendation and it's sending
- 3:24:24it to ClickUp, but also as an email to
- 3:24:27my Grockbot. So you can see we now have
- 3:24:29already set that up in trigger.dev dev
- 3:24:30to basically mimic a codeex routine with
- 3:24:33Astra on the back end with a full codeex
- 3:24:35agentic loop, but we could now trigger
- 3:24:37this programmatically and we could do it
- 3:24:38by web hooks and on different schedules.
- 3:24:40And like I said, you only would really
- 3:24:42do this when you need it to truly be
- 3:24:44programmatic and at scale because
- 3:24:45otherwise you want to throw it on your
- 3:24:46subscription because paying for Astra
- 3:24:48via API credits is obviously more
- 3:24:50expensive than paying for Astra via
- 3:24:51subscription. But that is how these
- 3:24:54three types of automations differ. And
- 3:24:55now I can come back in here and I can
- 3:24:58cross out Codex SDK. So Codex also has
- 3:25:01this really cool feature inside the
- 3:25:02desktop app called voice mode, which
- 3:25:04basically just means that you can
- 3:25:05actually just talk to it and it will
- 3:25:06talk right back to you. So it's super
- 3:25:08super cool. Let's hop into this next
- 3:25:09section about how to use Codex's voice
- 3:25:12mode. Hey there Astra. I've got a few
- 3:25:14tasks today. The first one that I want
- 3:25:15you to help me do, I want you to
- 3:25:16delegate off to a codeex thread where we
- 3:25:18take a video that I made on YouTube a
- 3:25:20few days ago. Or actually no, it was
- 3:25:21just yesterday. So, I want you to turn
- 3:25:22that into an X article for me and put it
- 3:25:24in my X account.
- 3:25:26>> Got it. Let's line that up. So, that X
- 3:25:28article is ready. Let me open this up
- 3:25:29and we'll take a look. We can see that
- 3:25:31we have all of this formatted. We have
- 3:25:33that 5x2 thumbnail that we wanted that
- 3:25:34matches our YouTube one. And now we can
- 3:25:36see that these screenshots are coming
- 3:25:37through. So, the prompt, it analyzed the
- 3:25:39entire video, screenshotted it, put this
- 3:25:41in here. It even cropped it. If you guys
- 3:25:42notice, this isn't the full screen, but
- 3:25:44it's matching these up because it
- 3:25:45understands what am I seeing in this
- 3:25:47image and where do I put it in the text
- 3:25:49based on the actual article that was
- 3:25:50written. Let's also try something else.
- 3:25:52So, inside of my school community, AI
- 3:25:54Automation Society Plus, we just did a
- 3:25:56little bit of a revamp to the actual
- 3:25:58course. So, if you could sort of just
- 3:26:00like look in there, get familiarized
- 3:26:02with that course structure and just help
- 3:26:04me build a really simple, you know, AIS
- 3:26:06branded landing page.
- 3:26:07>> Yep. I'll start on that.
- 3:26:09>> All right. I just got word that the AIS
- 3:26:10Plus curriculum page is done and it's
- 3:26:12looking pretty impressive. Let's open it
- 3:26:13up full screen. Holy, this is actually
- 3:26:16the courses. So, I just built this
- 3:26:18entire personal OS with just my voice
- 3:26:20using GBD6 Astra voice mode. Now, at a
- 3:26:22glance, it might not seem too
- 3:26:23impressive, but it had to figure out how
- 3:26:25to connect my entire Herk Brain. It had
- 3:26:28to go pull all the data from these
- 3:26:29sources. It had to connect my real
- 3:26:30calendar, and this is actually
- 3:26:31functional, meaning here's my actual
- 3:26:32calendar, and if I came here and I
- 3:26:34wanted to add an event right here, we'll
- 3:26:35just call this test, save event. This is
- 3:26:37actually going to reflect on my real
- 3:26:39calendar over here. Now, there are still
- 3:26:40a few things that I'm looking to
- 3:26:42connect, but I can communicate with my
- 3:26:43team through Slack, ClickUp, and email
- 3:26:45right from this interface. So, I can
- 3:26:47open up this conversation and I could
- 3:26:48reply to it right here. And I can also
- 3:26:50kick off new threads via email or
- 3:26:51ClickUp from here as well. And I also
- 3:26:53have community polls from YouTube, from
- 3:26:55my school communities, what they're
- 3:26:56stuck on. And I also have industry polls
- 3:26:58from X, YouTube, things like that. The
- 3:27:00reason voice mode is so powerful is
- 3:27:02because it just uses codecs on the back
- 3:27:03end, which means it can look through
- 3:27:05your AI operating system and it can do
- 3:27:06everything that Codeex can do, but you
- 3:27:08can control it with your voice and
- 3:27:09multitask, kick off different threads.
- 3:27:11You can do it from your phone, so you
- 3:27:12can literally be on a walk and doing
- 3:27:14everything you need to be doing. And the
- 3:27:15browse use capabilities from GBD6 Astra
- 3:27:17are the best I've ever seen. Right here
- 3:27:18on screen, you can just see a silly
- 3:27:20example where I gave it two pictures and
- 3:27:21I had it paint me in Canva. And this did
- 3:27:23take quite a while, but if you think
- 3:27:25about all of the little strokes and
- 3:27:26everything that needs to do, both on the
- 3:27:28vision side and the browser use side,
- 3:27:30it's actually pretty impressive. Now, on
- 3:27:31top of all of the capabilities, you also
- 3:27:32need it to have the right context and
- 3:27:34connections to actually be useful. So,
- 3:27:35when I tell you guys that I've been
- 3:27:36talking to Astra in voice mode, and it's
- 3:27:38just been feeling like it actually knows
- 3:27:40everything about me and my business,
- 3:27:41it's on another level. So, we just had
- 3:27:43an event in July called AIS Live and
- 3:27:44right here there's over 150 GB of
- 3:27:47footage to look through. So, I gave
- 3:27:48Codex this link and I basically just
- 3:27:50said, "Hey, we need like a 60-second
- 3:27:52sizzle reel from this event, like a
- 3:27:54recap." And in only two prompts, this is
- 3:27:56what I get back, which is really
- 3:27:57impressive. Logged off yesterday just
- 3:27:59buzzing. I had goosebumps. I was so
- 3:28:00excited and I had so much fun.
- 3:28:04[music]
- 3:28:13>> [music]
- 3:28:20>> Cost doesn't justify price. Price
- 3:28:23justifies cost.
- 3:28:25[music]
- 3:28:29What is the value of what [music] you
- 3:28:31know and what you know how to do?
- 3:28:36>> [music]
- 3:28:38>> This culture that we have here at AIS is
- 3:28:40is just really special.
- 3:28:43[music]
- 3:28:52[music]
- 3:28:55If you think about the actual story that
- 3:28:56it told and how much data it looked
- 3:28:58through and what it was able to create,
- 3:29:00that was really, really impressive, like
- 3:29:02I said. So anyways, let me stop yapping
- 3:29:03to you guys and start yapping to GBD6
- 3:29:05Astra and show you guys what it can do.
- 3:29:06So all you're going to do is down here
- 3:29:07you can see there's a little button
- 3:29:08called start new voice chat. You're
- 3:29:10going to click on that. And now it's
- 3:29:11time to put my headphones in. Hey there
- 3:29:13Astra. I've got a few tasks today. The
- 3:29:15first one that I want you to help me do,
- 3:29:16I want you to delegate off to a codeex
- 3:29:18thread where we take a video that I made
- 3:29:20on YouTube a few days ago or actually no
- 3:29:21it was just yesterday. It was called
- 3:29:23like Astra made this. It made the whole
- 3:29:25video. So, I want you to turn that into
- 3:29:27an X article for me um and put it in my
- 3:29:30X account.
- 3:29:31>> Got it. Let's line that up.
- 3:29:33>> And while you're working on that, can we
- 3:29:34have a separate thread actually start
- 3:29:36making the thumbnail for that?
- 3:29:39>> Sure. Adding that as well.
- 3:29:41>> Now, I do want to make sure that you're
- 3:29:42using all my right skills and my right
- 3:29:44context. So, make sure you're working
- 3:29:45inside of my Herk 2 project for all of
- 3:29:48these little tasks that we're going to
- 3:29:50kick off today.
- 3:29:52>> Got it. I'll make sure that's the
- 3:29:54context going forward. The original
- 3:29:56article task has been told to stop and
- 3:29:58hand off any verified finds. And the
- 3:30:00thumbnail task has just been told the
- 3:30:02same.
- 3:30:03>> Okay, awesome. Yeah, cuz I see that you
- 3:30:04kicked off two regular chats and these
- 3:30:06wouldn't have had the same context. So,
- 3:30:08it's important that I told you to work
- 3:30:09inside my Herk 2 project and now you
- 3:30:12should be kicking off the right ones up
- 3:30:13there. And it looks like you are.
- 3:30:17>> Yep. Doing that now. And the article
- 3:30:20task now has the thumbnail tasks details
- 3:30:22so they can coordinate the source, the
- 3:30:24angle, and the finished cover. Both
- 3:30:26replacements are running inside Herk 2.
- 3:30:30Awesome. Now, for the actual thumbnail I
- 3:30:33want for this X article, I actually
- 3:30:35already made one that I really like. um
- 3:30:37the one that I put on my actual YouTube
- 3:30:39channel, it was like an image of
- 3:30:42um it was like me with my feet up on the
- 3:30:45desk and the
- 3:30:48OpenAI logo was editing the video. So,
- 3:30:50all I really actually need you to do
- 3:30:51there is if you could just turn that
- 3:30:52into a 5x two version so it fits into
- 3:30:55the X article.
- 3:30:58>> Sure. I'll update the thumbnail work
- 3:31:00with that.
- 3:31:01>> Awesome. So, I know that probably will
- 3:31:03take a little bit of time, so just let
- 3:31:04me know when those are done. Let's also
- 3:31:07try something else. So, inside of my
- 3:31:09school community, a automation Society
- 3:31:10Plus, we just did a little bit of a
- 3:31:12revamp to the actual course. So, if you
- 3:31:14could sort of just like look in there,
- 3:31:16get familiarized with that course
- 3:31:18structure. Um, I'm talking about the AI
- 3:31:20Automation Society Plus group inside of
- 3:31:23that one. if you could just analyze the
- 3:31:24structure real quick and understand sort
- 3:31:25of like the transformation from a new
- 3:31:28joiner to the end state once they've
- 3:31:30taken all the courses and just help me
- 3:31:32build a really simple you know AIS
- 3:31:34branded landing page that I could
- 3:31:36basically send people to if they're like
- 3:31:37hey you know what is AIS plus I could
- 3:31:39send them there and they could get a
- 3:31:41rundown of what that curriculum looks
- 3:31:42like and I want you to also design that
- 3:31:44in a way where you use my scrollcraft
- 3:31:47skill to have layers and to have it be
- 3:31:49you know dynamic and visual but don't
- 3:31:52make it too overwhelming So if you could
- 3:31:53start designing that website as well
- 3:31:54based on that knowledge.
- 3:31:57>> Yep. I'll start on that and I'll keep
- 3:32:00the design dynamic but not overwhelming.
- 3:32:03>> Awesome. So now I see that I'm talking
- 3:32:04to you in one thread but then I can also
- 3:32:06see when I look at the all the threads
- 3:32:08that you're kicking off. I can see how
- 3:32:09you're clearly multitasking. And then by
- 3:32:11the way I see the thumbnail that you
- 3:32:12made. That looks great. That's exactly
- 3:32:14what I was looking for. Will my other
- 3:32:16thread that is making the actual article
- 3:32:19it knows that that's there. Oh yeah. It
- 3:32:21looks like you sent that exact message
- 3:32:22where I said that looks great. That's
- 3:32:23exactly what I was looking for. So, you
- 3:32:24are able to actually coordinate these
- 3:32:26different codec threads to talk to each
- 3:32:28other.
- 3:32:29>> That's right. The article task has the
- 3:32:31approved cover and your go ad.
- 3:32:33>> Awesome. Well, while I wait for me to be
- 3:32:36able to verify that that all worked,
- 3:32:38could you let's make another improvement
- 3:32:40to the Upit OS. Let's see if we could
- 3:32:42add I guess let's add another section
- 3:32:44for meetings. I know we already can see
- 3:32:47all of my meetings in the actual
- 3:32:48Herkbrain visual, but let's add another
- 3:32:51tab on the left that just shows the
- 3:32:53meeting. So, you pull that in from
- 3:32:54Fireflies and it's just kind of like a
- 3:32:56nice visual of the date that the meeting
- 3:32:59was and just like a one sentence summary
- 3:33:01of what was discussed or if there's any
- 3:33:03important action items from there just
- 3:33:04so I can visually see all the meetings
- 3:33:05that I had. Um, and maybe if there are
- 3:33:08any important meetings coming up as
- 3:33:10well.
- 3:33:11>> Okay, I'll take that on.
- 3:33:13>> Cool. So, while you're cooking on that,
- 3:33:15I'm going to go ahead and actually pause
- 3:33:16this voice chat. I'll pick it back up
- 3:33:18with you in a bit when you have some
- 3:33:19updates for me. But, I want to kick off
- 3:33:21some voice threads on my mobile device
- 3:33:24and see how that all works as well. So,
- 3:33:25I'll be back in a bit. Okay. So,
- 3:33:26anyways, just wanted to show you guys
- 3:33:27what was actually going on here. So, it
- 3:33:29made this thumbnail, which was actually
- 3:33:32based on if I go to my thumbnails, this
- 3:33:34one. So, this is the one that I wanted
- 3:33:36to use because this is the one that did
- 3:33:38best on YouTube. And I just needed it to
- 3:33:40make that a 5x two. And what you guys
- 3:33:42will notice is I didn't tell it to find
- 3:33:43this exact thumbnail. It looked through
- 3:33:45my YouTube probably to see which one was
- 3:33:47the live one. It looked through the
- 3:33:48local files to actually build that. Same
- 3:33:50thing over here. What happened was I
- 3:33:52don't know what it delegated around, but
- 3:33:53all I said was I need an ex article for
- 3:33:55this. And so what it did is it had to
- 3:33:57pull in the actual video. It's
- 3:33:58transcribing it. And now what it's doing
- 3:34:00is it's making the expost, but it also
- 3:34:02took 11 screenshots. And so it's going
- 3:34:03to place those screenshots in the X
- 3:34:05article at the right spot. So we'll get
- 3:34:06to view that in a sec. But I just wanted
- 3:34:08to call out what's actually going on on
- 3:34:10the back end there cuz it's very
- 3:34:11impressive. But now I need you guys to
- 3:34:13look at this. So you can also do this on
- 3:34:15your phone. So imagine you're on a walk
- 3:34:17or you're in your car driving and you
- 3:34:19need to kick off some tasks or delegate
- 3:34:20some work to your team or whatever it
- 3:34:22might be. Because it has all the
- 3:34:24connections in here. You're going to go
- 3:34:25to remote. You're going to scroll down
- 3:34:28and this is where you'll Oh, wait. Right
- 3:34:29here. This button right here. This is
- 3:34:30where you'll start a new voice chat. And
- 3:34:33now it's going to be the same thing.
- 3:34:38Hey, so I need you to help me kick off a
- 3:34:40thread inside of um my Herk 2 project.
- 3:34:43What I want you to do is I just need a
- 3:34:45nice rundown of the best use cases that
- 3:34:48you've seen, the best viral demos that
- 3:34:50you've seen of Astra 6 um you know, GPT
- 3:34:53Astra 6 on X.
- 3:34:56>> Sure.
- 3:34:57>> So, the point I'm trying to make here is
- 3:34:58that I can keep scrolling on my phone
- 3:35:01even when it's doing things on the back
- 3:35:02end. As you can see, it's still keeping
- 3:35:03this conversation
- 3:35:06Got it. I'll keep going here. Yep,
- 3:35:08that's exactly it. I kicked off the
- 3:35:09Astro research thread and we can keep
- 3:35:11talking here meanwhile.
- 3:35:12>> Awesome. So, while we're while we're
- 3:35:14waiting on all this, could you just take
- 3:35:15a look at my, you know, the look at the
- 3:35:18YouTube space, look at some of my past
- 3:35:20videos and look at what else is going on
- 3:35:22in the space and give me a quick
- 3:35:24analysis of maybe some trends and maybe
- 3:35:26some things to start preparing for. I'll
- 3:35:28scan the current landscape. Now, big
- 3:35:29trends heading toward 2027 are more AI
- 3:35:33assisted production and personalization,
- 3:35:35deeper community play with memberships
- 3:35:37and owned audiences, short and long form
- 3:35:40blending into one funnel, and creators
- 3:35:43rewarded for building recognizable shows
- 3:35:45and perspectives, not just one-off
- 3:35:47uploads.
- 3:35:48>> So, the point I was trying to make there
- 3:35:49was just that you can do this on your
- 3:35:50phone as well, and it's a super clean,
- 3:35:52crisp syncing between your desktop and
- 3:35:55your phone. So, we've got the Up OS
- 3:35:57working on adding a Fireflies task or a
- 3:35:59Fireflies meeting section. So, I can see
- 3:36:00that stuff. We've got the X article
- 3:36:02being finalized and we'll pull that up.
- 3:36:03So, that X article is ready. Let me open
- 3:36:05this up and we'll take a look. We can
- 3:36:07see that we have all of this formatted.
- 3:36:09We have that 5x2 thumbnail that we
- 3:36:10wanted that matches our YouTube one. We
- 3:36:12have all of this formatted in a good
- 3:36:14way. And now we can see that these
- 3:36:15screenshots are coming through. So, the
- 3:36:17prompt, it analyzed the entire video,
- 3:36:19screenshotted it, put this in here. It
- 3:36:20even cropped it. If you guys notice,
- 3:36:22this isn't the full screen. We keep
- 3:36:23going down. This is the full screen.
- 3:36:25This is a full screen. Full screen. But
- 3:36:27it's matching these up because it
- 3:36:28understands what am I seeing in this
- 3:36:30image and where do I put it in the text
- 3:36:32based on the actual article that was
- 3:36:33written. And there is nothing sensitive
- 3:36:35in this video. But what it will actually
- 3:36:36do too is if I have like an API key in
- 3:36:38there or if I have like a personal
- 3:36:40email, it actually will like blur it
- 3:36:42out. So it does that automatically. And
- 3:36:44now that is just worked into the skill.
- 3:36:46So this is now an article that I would
- 3:36:47just be able to read through. All of
- 3:36:49these screenshots are taken. They're
- 3:36:50cropped if they needed to be. They're
- 3:36:51blurred out if they needed to be. And
- 3:36:53then I can just go ahead and publish it
- 3:36:54because it did all of that with just my
- 3:36:56voice. We can also see that this one
- 3:36:58that I kicked off from my phone is done.
- 3:37:00It searched through X and it found um
- 3:37:02multiple posts. One from Ben Davis, one
- 3:37:04from Ethan, one from Thomas. You can see
- 3:37:05that it gives us the actual link to all
- 3:37:07of these threads. So if I want to click
- 3:37:08on, you know, this 3D graphics build, I
- 3:37:11get taken straight to that actual post.
- 3:37:121.3 million views. I did see this one.
- 3:37:14This one was pretty impressive. We can
- 3:37:16see that now we have this little
- 3:37:17rundown. And if we would have stayed on
- 3:37:19the voice chat, it would have, you know,
- 3:37:20told us about all this. It looks like
- 3:37:22the upp one is also done. Let's see what
- 3:37:24happened over here. Meetings is live.
- 3:37:25So, let me open this up and we'll see.
- 3:37:27Actually, I'm going to open this up full
- 3:37:28screen. We now have a section called
- 3:37:30meetings. You can see on the calendar,
- 3:37:31I've got one tomorrow. I've got one on
- 3:37:33Monday. And we can actually pull up the
- 3:37:35actual calendar details. So, if I open
- 3:37:37this one up, does this take us straight
- 3:37:39to the calendar? Yeah, it takes us
- 3:37:40straight to the actual calendar. And we
- 3:37:41see the event details there. And now we
- 3:37:42can also see past meetings. So this all
- 3:37:44hands we can see the team reviewed
- 3:37:46learning community growth product
- 3:37:47updates operational improvements AI
- 3:37:48tools and a shared mission centered on
- 3:37:50customer happiness. We have an action
- 3:37:52highlight Nate Herk share bookmark links
- 3:37:54to the most used Google Drive folders.
- 3:37:56Yep. Yep. Yep. And we can see other
- 3:37:58action items here and we can also open
- 3:37:59up this recording which takes us
- 3:38:01straight to Fireflies. So that quickly
- 3:38:03we were able to develop a new section
- 3:38:05into our actual um OS here which is a
- 3:38:09combination of Fireflies and Google
- 3:38:10Calendar. And I don't even know if it
- 3:38:12uses any AI in the background because
- 3:38:13Fireflies already processes these AI
- 3:38:15action items. Maybe it hit, you know,
- 3:38:17the GBT endpoint or the Enthropic
- 3:38:19endpoint to actually put this stuff in
- 3:38:21here, but really impressive. All right.
- 3:38:22And I just got word that the AIS Plus
- 3:38:24curriculum page is done. And it's
- 3:38:25looking pretty impressive. Let's open it
- 3:38:26up full screen. Build the skills shape.
- 3:38:28What's next? And we've got Holy, this is
- 3:38:32actually the courses. I wonder if it
- 3:38:35Okay, it takes you down to here. There's
- 3:38:36like a curriculum section. Now, I think
- 3:38:38this looks really good. It pulled a
- 3:38:39picture of me. You also will realize
- 3:38:41that the background, right? The
- 3:38:43background, you can see these layers.
- 3:38:44The mountains are in different spots.
- 3:38:45They move in different, you know, paces.
- 3:38:48But what I love about this that it did
- 3:38:51is if I actually go to our AI Automation
- 3:38:52Society website, it kept a very similar
- 3:38:55vibe, right? Like the blue feeling, the
- 3:38:57backgrounds um of the mountains and
- 3:38:59stuff, even on the certification tab, it
- 3:39:01kept the consistency. Now, honestly, I
- 3:39:03think these mountains look a little bit
- 3:39:04cooler and there's like fog, but it kept
- 3:39:06everything on brand because it was able
- 3:39:08to crawl through my workspace, look up
- 3:39:09what it needed to look up to make sure
- 3:39:11that it actually understood what I was
- 3:39:12asking for. And that's what I think is
- 3:39:14so cool about Astra is when you give it
- 3:39:16a task, it feels like it just will
- 3:39:18contextualize and ask you questions and
- 3:39:19crawl through things first because it
- 3:39:21wants to make it more specific to you.
- 3:39:23And that's what is just really, really
- 3:39:25cool. I mean, this is this is the actual
- 3:39:26curriculum. If I go to AS+ and I go to
- 3:39:28classroom, that's exactly what happens.
- 3:39:30you know these are the actual courses
- 3:39:31and then over here uh wrong one these
- 3:39:34are the actual courses that we have in
- 3:39:35there so it was able to look through all
- 3:39:36of that and build out the actual
- 3:39:39curriculum I think this is really really
- 3:39:41slick all right so voice mode is one of
- 3:39:43the big differentiators between cloud
- 3:39:44code and codec but there are a lot of
- 3:39:46other things and I will be honest in
- 3:39:48this space you are jumping back and
- 3:39:50forth a lot like I use both of them
- 3:39:51every day and sometimes you know codeex
- 3:39:53will come out with a big new release and
- 3:39:55then the next week claude code will come
- 3:39:57out with a new big release or maybe
- 3:39:58there's a new tool like Hermes agent or
- 3:40:00Grockbot or whatever you know is going
- 3:40:02on in the AI world because it moves very
- 3:40:04fast. Now the first thing I want to say
- 3:40:06is that you should not be worried
- 3:40:08because this whole course we've been
- 3:40:10talking a lot about building your own
- 3:40:11AIOS and when you're building your own
- 3:40:13AIOS you guys know that it's just a
- 3:40:15bunch of folders and files and those
- 3:40:16live on your machine. So whether you are
- 3:40:18using codeex or cloud code or Hermes
- 3:40:20agent or whatever the next tool is you
- 3:40:23just say hey Mr. Mrs. AI agent, look at
- 3:40:26my AIOS, get familiarized with it, and
- 3:40:28now you can in a matter of 5 seconds,
- 3:40:30all your skills transfer over, all your
- 3:40:32instruction files transfer over. You
- 3:40:33don't have to be worried about being
- 3:40:35like vendor locked. Everything that
- 3:40:36you're building, not only your local
- 3:40:37files and folders, but also your brain,
- 3:40:39your knowledge on how to work with AI
- 3:40:41and how to talk to it and the mindset
- 3:40:42around it, that is all your IP. And it's
- 3:40:45very tool agnostic. It's very model
- 3:40:47agnostic. So, you can be flexible. Don't
- 3:40:49worry about it. But I do want to show
- 3:40:52you guys this next video which is about
- 3:40:54me running a bunch of different use
- 3:40:55cases on these two coding agents and
- 3:40:58just showing you the difference what
- 3:40:59that might look like sometimes with cost
- 3:41:01and speed and just like quality. So hope
- 3:41:03you guys enjoy this next section which
- 3:41:04is kind of just like an experiment where
- 3:41:06I show off the differences between these
- 3:41:07tools in these different use cases. I
- 3:41:09just put GPT6 Astra headto-head against
- 3:41:11Fable 5.1 across 15 different use cases
- 3:41:13that actually apply to my day-to-day.
- 3:41:15I'm talking things like web design,
- 3:41:17personal organization, taxes, browser
- 3:41:20use, vision, and building actual
- 3:41:22automations and software. For every
- 3:41:24single one of these use cases, I break
- 3:41:25down which model did better, how long it
- 3:41:27took, and what it cost. So, if you've
- 3:41:29been wondering what model to use for
- 3:41:30which use case, then this is the video
- 3:41:31for you. I don't want to waste any time.
- 3:41:33Let's just get straight into it today.
- 3:41:34All right, so these are the 15 different
- 3:41:35use cases that we have. And like I said,
- 3:41:37I chose these because these are things
- 3:41:38that I actually see myself doing with
- 3:41:40these models. So, there are timestamps
- 3:41:41down below if you want to jump around.
- 3:41:42And at the very end, I'm going to show
- 3:41:43the final sort of like takeaways and
- 3:41:45final cost and time breakdown. And by
- 3:41:47the way, guys, I've eaten through like
- 3:41:48three CEX subscriptions, four cloud
- 3:41:50subscriptions, and thousands of dollars
- 3:41:51in usage credits. So, I really hope that
- 3:41:53this one's helpful. All right, so for
- 3:41:54number one, I'm not going to read off
- 3:41:55every single prompt, but here's
- 3:41:56basically what I asked. I wanted a
- 3:41:58McKenzie level presentation on the state
- 3:42:00of SMB report for my Hercules Consulting
- 3:42:03brand. So, I needed this to be branded
- 3:42:04and I needed it to apply to our target
- 3:42:06audience with this fake consulting
- 3:42:08business. So, that was Claude. You can
- 3:42:09see that I gave Codex the exact same
- 3:42:10prompt. Now, immediately what I've
- 3:42:12noticed is that Codex asks more
- 3:42:13questions or GBT6 Astra specifically
- 3:42:16asks more questions. You can see right
- 3:42:17here, it already asked me questions,
- 3:42:19whereas Fable just took this and pretty
- 3:42:21much just ran. And that is something I
- 3:42:22noticed overall. So, I just wanted to
- 3:42:24call that out to start. So, let's take a
- 3:42:25look at these two decks. Here is deck
- 3:42:27number one. I won't tell you which model
- 3:42:28it was yet, but right away, I love the
- 3:42:29way it feels. It is, you know,
- 3:42:31capitalized. It has the branding. This
- 3:42:34feels pretty professional so far. Now, I
- 3:42:36already did check both of these for the
- 3:42:37sources and the verification, and they
- 3:42:39are good from like the fact perspective,
- 3:42:41right? The research they both did was
- 3:42:43good. Now, I'm looking more so at the
- 3:42:45way they actually formatted this
- 3:42:46deliverable. I think we're at a place
- 3:42:47now where GBT6 Astra and Fable 5.1,
- 3:42:49they're both so so intelligent. You
- 3:42:51don't have to worry as much about
- 3:42:52fanning out agents and doing research.
- 3:42:54It's more to me now about how do they
- 3:42:56come back and present this in a way that
- 3:42:58tells a story and in something that I
- 3:42:59can actually like present over, right?
- 3:43:00because that's what we want is we want
- 3:43:02to turn that messy data into something
- 3:43:03that's readable and something that's
- 3:43:05high quality. And right now what I'm
- 3:43:06feeling is that this is very very
- 3:43:08branded. It matches our color scheme. We
- 3:43:10have a footer down here which I love.
- 3:43:11It's consistent. We have like structure.
- 3:43:14There's columns. Like it did a good job
- 3:43:15creating this. The thing that I think
- 3:43:17right now though is it's very very
- 3:43:18wordy. As far as being able to present
- 3:43:20over something like this, it would be
- 3:43:22tough because there's so many words and
- 3:43:24it's tough to control like where
- 3:43:25people's eyes will be looking and what
- 3:43:26they'll be speaking about. So, as far as
- 3:43:28a presentation deck, not the best, but
- 3:43:30as far as like a deliverable you might
- 3:43:32turn in, you might send to someone and
- 3:43:33not present, this is very good. And
- 3:43:35there's a lot of sources down here. This
- 3:43:37came out to 35 slides. Now, let's look
- 3:43:39at the other version, which right away I
- 3:43:40don't like how this isn't capitalized. I
- 3:43:42don't know. I just feel like that feels
- 3:43:43very professional. We have some images
- 3:43:44here, and we have the color scheme as
- 3:43:46well. But let's flick through. We
- 3:43:48immediately notice that this one is not
- 3:43:49as wordy. I immediately feel like this
- 3:43:51one's less professional though because
- 3:43:52of the structure. The other one had like
- 3:43:54I don't know, you could tell in each
- 3:43:55slide there was like columns. there was
- 3:43:57a little bit of like a I don't know
- 3:43:59there was a structure to it whereas this
- 3:44:00one feels a little bit more abstract. It
- 3:44:02feels like one big element. You've got
- 3:44:04one big, you know, title thing. And I
- 3:44:05think that this one would be a much
- 3:44:07better deliverable if I was going to
- 3:44:08present in, you know, in a room full of
- 3:44:10people. But overall, as we flick
- 3:44:11through, this one just feels less
- 3:44:13professional. The vibe I get is less
- 3:44:15McKenzie and it's more Google Slides or
- 3:44:18like Canva. I don't know. There's also
- 3:44:20not like a consistent footer or our logo
- 3:44:22isn't on every slide. So, you know, this
- 3:44:23sort of slide, it just doesn't feel very
- 3:44:25structured and as professional. Even the
- 3:44:27sources section is way smaller, although
- 3:44:29that does link the actual sources. So
- 3:44:30anyways, this one was GBT6 Astra and
- 3:44:33this one was Fable 5.1. And I will say
- 3:44:37in this specific example, I liked Fable
- 3:44:40better. So what I'm going to do here is
- 3:44:42I'm going to give Fable the win. So I'm
- 3:44:43just going to put this down here so we
- 3:44:44know Fable won this experiment as far as
- 3:44:47just the output. But let's look at the
- 3:44:50time and cost. Fable was 37 minutes and
- 3:44:52Astra was 23 minutes and Fable was 26
- 3:44:54bucks and Astra was 12 bucks. So Astra
- 3:44:56was a little less than half the cost as
- 3:44:58well as you know 14 minutes faster. So
- 3:45:02the question would be if you used Astra
- 3:45:04and spent another you know 14 bucks
- 3:45:06using Astra to improve V1 would you get
- 3:45:09somewhere that you feel better about
- 3:45:12than this first version of Fable. So
- 3:45:14anyways right now we'll look at the time
- 3:45:16and cost at the very end but as far as
- 3:45:18the deliverable I liked Fable's output
- 3:45:20better here. All right moving on to
- 3:45:21experiment number two. We did a sales
- 3:45:23letter. So basically I said based on
- 3:45:25what you know about our certification
- 3:45:27program I didn't give it information. I
- 3:45:28told it to you know use my AIOS use my
- 3:45:31second brain use any research to put
- 3:45:33together a sales landing page. So
- 3:45:35writing the copy that's optimized to
- 3:45:36convert. So think about our avatar think
- 3:45:38about their pain. Think about all this
- 3:45:39and I don't want it to sound AI
- 3:45:41generated. So I'm not going to read the
- 3:45:42entire sales letters. That would be
- 3:45:43super boring. Just going to do a quick
- 3:45:45skim through and talk about which one I
- 3:45:47actually thought was better and why. So
- 3:45:50here was the one that Fable 5.1 wrote
- 3:45:52out. You can see it is pretty long. This
- 3:45:55one was longer. It was about 2,800
- 3:45:57words. And if we go over to GBT's
- 3:45:59version, if I come down here, it just
- 3:46:00output it right in the chat to me. This
- 3:46:02was the one that Astra wrote up. And
- 3:46:04this one was more like 1,700 words. So,
- 3:46:07it was definitely shorter. Now, quick
- 3:46:10disclaimer here. I am not an expert
- 3:46:12copywriter. I would want, you know, like
- 3:46:14John or someone on my team who is really
- 3:46:16good at writing copy to tell me which
- 3:46:17one they thought is better. I will say
- 3:46:19so far what I've been hearing from my
- 3:46:21team is that they are liking the GBT
- 3:46:23models lately better for writing whereas
- 3:46:24in the past Claude always dominated when
- 3:46:26it came to writing. But I will say if I
- 3:46:28was a prospective you know certification
- 3:46:31student and I was reading these I
- 3:46:32honestly here would have picked Fable
- 3:46:34and the reason is because this is not a
- 3:46:35cheap program and Fable had things like
- 3:46:38questions people ask. Will this get me
- 3:46:40hired? Does everyone who joins get
- 3:46:41certified? Am I technical enough? I
- 3:46:42already run an agency. What if I miss
- 3:46:44live classes? The way it formatted this
- 3:46:46makes me feel a little bit more
- 3:46:47confident or makes me feel more aware of
- 3:46:48what I'm getting myself into. It also
- 3:46:50had a section about tuition. It also had
- 3:46:51a section about who's teaching, who it's
- 3:46:53for, who it's not for, how this actual
- 3:46:54program works. Whereas the sales letter
- 3:46:56over here with Astra just felt a little
- 3:46:58bit more, I don't know, it just felt a
- 3:47:00little bit more high level. And I think
- 3:47:01with the sales letter, it's supposed to
- 3:47:02be detailed enough to answer all the
- 3:47:04questions and pain that someone's
- 3:47:05thinking about when they're considering
- 3:47:07paying for a program. So, I will say in
- 3:47:09this specific example, I do think that I
- 3:47:11liked Fable 5.1's copy more. So here is
- 3:47:13actually what this looked like. Fable
- 3:47:15was about four minutes. Astra was about
- 3:47:17three minutes. Fable wrote Okay, I was
- 3:47:19wrong about Astra. It wasn't 1,700. It
- 3:47:21was actually more like 1300. So about
- 3:47:22half of the words that Fable wrote. But
- 3:47:24Astra came in at $143 whereas Fable came
- 3:47:27in at almost four bucks. So here I will
- 3:47:30have to say again that I think that
- 3:47:31Fable won this challenge as far as just
- 3:47:33like a oneshot prompt and the initial
- 3:47:36output. All right, so moving on to use
- 3:47:38case number three. Now, unfortunately,
- 3:47:39guys, I'm going to have pretty much
- 3:47:41everything blurred out here because this
- 3:47:42is about my taxes, right? And this is
- 3:47:45one of those things where you want
- 3:47:47something that is very smart to be able
- 3:47:49to do this and in a way that you trust.
- 3:47:51So, I'm just going to go through high
- 3:47:52level the way I feel when I look at
- 3:47:53these outputs. This first one was Fable
- 3:47:555.1. And before I get into it, I do want
- 3:47:57to say in this example, GBT Astra asked
- 3:48:00me like seven questions before it
- 3:48:01started, whereas Fable didn't. So, that
- 3:48:04alone makes me feel more confident in
- 3:48:05Astra's output. But let's look through
- 3:48:07here. We go through basically from
- 3:48:09January of this year to June of this
- 3:48:10year. So Q1 and Q2 of this year. It
- 3:48:13pulls all of the numbers. It looks at
- 3:48:14federal and state. It looks at things
- 3:48:16that I need to read. It goes to the
- 3:48:18input and it shows me all of these
- 3:48:19different things as far as the
- 3:48:20parameters, the notes, the values, the
- 3:48:22sources. And Fable gave me like a
- 3:48:24decision logic. It said, "Hey, if you
- 3:48:26fit into this bucket, do this. If you
- 3:48:27fit into this bucket, do this." Whereas
- 3:48:28Astra just asked me those questions and
- 3:48:30then gave me that recommendation.
- 3:48:31Anyways, we have a um P&L. We have some
- 3:48:34projections. So for the rest of the
- 3:48:35year, what I might be estimating to pay.
- 3:48:37We have tax calculations here for the
- 3:48:39individual return and all of the
- 3:48:41entities that are passed through. We've
- 3:48:42got payments. We've got things that are
- 3:48:44flagged. And then we have the actual
- 3:48:45sources it pulled from. So it's not bad.
- 3:48:47It's a really good deliverable like
- 3:48:49light years ahead of what used to
- 3:48:50happen. But now we have the one from
- 3:48:52Astra. We have the actual information as
- 3:48:55far as start here. This is what it did.
- 3:48:56We have monthly results. We have the tax
- 3:48:58forecast. We have Okay, that looks very
- 3:49:01good. We have all these payments. We
- 3:49:02have inputs and open items. So, these
- 3:49:04are other things that I might need to
- 3:49:05answer, but this is just way more
- 3:49:07tailored. It's easier to read as well.
- 3:49:08We have the source checks. We have the
- 3:49:10transaction ledger. And this one is
- 3:49:12basically every single transaction here.
- 3:49:14So, this Oh my gosh, this is probably
- 3:49:16thousands and thousands of rows. This is
- 3:49:183739 rows of transactions here. And then
- 3:49:21we've got all the sources that it pulled
- 3:49:23from. Okay, cool. So, overall, this one
- 3:49:25I think that Astro won, like no
- 3:49:27question. It was more tailored. It asked
- 3:49:28questions. I think with taxes alone, it
- 3:49:31gave me that that feeling of confidence
- 3:49:32and the structure of the deliverable was
- 3:49:34just better to me. Now, Astra did cost
- 3:49:36more here and it took longer to run. So,
- 3:49:3840 minutes compared to 22 minutes and 22
- 3:49:40bucks compared to 13 bucks. So, even
- 3:49:42though it's more expensive, we're kind
- 3:49:44of putting that aside right now. As far
- 3:49:46as the deliverable, I liked Astra here
- 3:49:48more. So, right now, we're sitting at
- 3:49:50Astra 1 Fable 2. All right. So, this
- 3:49:52next one's interesting. I asked it to go
- 3:49:54through two of my email accounts and
- 3:49:55just look through, you know, potentially
- 3:49:57years of emails to help me find
- 3:49:59information about subscriptions that I'm
- 3:50:00paying for, how much total I've paid
- 3:50:02them, you know, um if prices have
- 3:50:04increased and things that I need to
- 3:50:05cancel, things like that. So, let me
- 3:50:07pull up the two outputs. All right, so
- 3:50:08here is output number one from Fable
- 3:50:105.1. Now, as we get into the next tabs
- 3:50:12on the sheet, I'm sorry, guys, but I am
- 3:50:14going to have to blur some of the stuff
- 3:50:14out. It's it's financial information.
- 3:50:16So, anyways, we have the rundown right
- 3:50:19here of the audit. We have the different
- 3:50:21emails it looked at. We have the
- 3:50:22different types of services,
- 3:50:23memberships, programs, retainers,
- 3:50:25contractors, insurance, all this type of
- 3:50:26stuff. It's looking at the top 10. It's
- 3:50:28looking at our different categories, and
- 3:50:30it's showing us how to read this sheet.
- 3:50:32So, we have all of our active
- 3:50:33subscriptions. It's showing me what
- 3:50:35email they're associated with. It's
- 3:50:36telling me which day of the month I'm
- 3:50:38getting build on. It's showing the
- 3:50:39current price. It's showing the monthly
- 3:50:40equivalent. It's showing me all of the
- 3:50:42stats across all of the different
- 3:50:43payments that we've been making on some
- 3:50:45sort of recurring basis. It's flagging
- 3:50:47things and it's ranking them as far as
- 3:50:48priority. We can see um the different
- 3:50:50actual flag notes. So some of these it
- 3:50:52might be unpaid. This one's getting a
- 3:50:53price jump. This one's having a seat
- 3:50:54creep. This one's having I have two
- 3:50:56subscriptions at once. So maybe we need
- 3:50:57to figure out why do I have two
- 3:50:58different subscriptions on both emails
- 3:51:00there. So the point being it's able to
- 3:51:01look through so much information and
- 3:51:03give you all of these things that are
- 3:51:04high priority and you can flag them. You
- 3:51:06can check them off. You can, you know,
- 3:51:07delegate that out to your team. We have
- 3:51:09all charges which actually this sheet
- 3:51:11came back completely empty. I'm not sure
- 3:51:13why, but this sheet is completely empty.
- 3:51:15And then we have contractors and big
- 3:51:16payments. So, these are things it looks
- 3:51:18like. Anything above
- 3:51:20um either a few thousand or if it was a
- 3:51:23contractor that was like a onetime
- 3:51:24invoice, then it flagged it here as
- 3:51:26well. And now, let's go over to the
- 3:51:27Astra 6 version. You can see that they
- 3:51:29consistently just format Google Sheets a
- 3:51:31little bit different. Like, you can kind
- 3:51:32of tell Astra likes to make the cells
- 3:51:34larger and and wider and taller, whereas
- 3:51:37Fable usually keeps them the same. And,
- 3:51:38you know, they still do a good job
- 3:51:39formatting, but either way, same sort of
- 3:51:41thing. At the audit up front, we have
- 3:51:43the current recurring baseline of
- 3:51:45subscriptions. We have the annualized
- 3:51:46baseline tool payments, tool invoices
- 3:51:48without confirmation, claude usage,
- 3:51:50other bills, um price changes. It's
- 3:51:53flagging everything right here. Priority
- 3:51:54review right here. You know, there's a
- 3:51:56discount ending, there is contract tier
- 3:51:58or a contact tier, there is a price
- 3:52:00increase, there's a term change, there's
- 3:52:01seats and tool overlap. So, it's
- 3:52:03flagging all that right away. We then
- 3:52:04have all the subscriptions. So, this
- 3:52:06shows me a list of every single one. It
- 3:52:08gives me the price. It gives me the um
- 3:52:10last known bill. Now, this isn't showing
- 3:52:13me what date. Oh, there it is. Yeah,
- 3:52:14it's showing me every single day of the
- 3:52:16month that we have the next one. We have
- 3:52:17other bills as well. So, these are kind
- 3:52:19of the bigger like bigger contracts um
- 3:52:21contractors or or bigger bills. We have
- 3:52:24payments. So, this is also like every
- 3:52:25single one. And I think this is what
- 3:52:27Fable was trying to do on this tab with
- 3:52:29all charges, but for some reason just
- 3:52:30didn't come through. So, we have this
- 3:52:32here. And then we have also other notes
- 3:52:34as far as different topics and different
- 3:52:35things to sort of be aware of. So, this
- 3:52:37is kind of like what it was flagging.
- 3:52:39So, overall here, I will say that Astra
- 3:52:41gave me a better output. It just, you
- 3:52:43know, Fable messed up one sheet
- 3:52:45completely and Astro was cheaper here.
- 3:52:48So 12 bucks, 12 and a half bucks
- 3:52:50compared to about $50, which is insane.
- 3:52:51$50 for that output from Fable. Crazy.
- 3:52:54It did take 17 minutes whereas Astra
- 3:52:55took 30 minutes. But anyways, right now
- 3:52:57we are sitting at 2 to2 Astra and Fable.
- 3:53:00All right, moving on to number five. We
- 3:53:02have meeting analysis. So I asked it to
- 3:53:04look through all of my leadership
- 3:53:05meetings and my syncs with John and
- 3:53:07looking at other meetings that are
- 3:53:08relevant to the team. And I wanted it to
- 3:53:10understand or I wanted it to tell me
- 3:53:12what are the biggest pain points, three
- 3:53:14biggest pain points and one of the
- 3:53:15highest value automations that we could
- 3:53:17build to remove some constraints and
- 3:53:18help us scale faster. So Fable gave me
- 3:53:21the biggest pain points of me being the
- 3:53:22single production and review engine
- 3:53:23which is something that I've talked
- 3:53:24about 25 of 58 meetings. Certification
- 3:53:27delivery does not scale past one cohort.
- 3:53:29So we're trying to figure out what that
- 3:53:30will look like. The critical constraint
- 3:53:31of the business is also a large cost
- 3:53:33center. 52 students blah blah blah. And
- 3:53:35then we also have demand collapsed and
- 3:53:36every fix is blocked by unknown work and
- 3:53:39untrusted data. Weekly leads fell blah
- 3:53:41blah blah. And yeah, all this makes
- 3:53:44sense as far as our three biggest pain
- 3:53:45points. And then the one automation it
- 3:53:46came up with was an AI customer success
- 3:53:48layer for theert one live record per
- 3:53:50student built from circle activity
- 3:53:51mastery check outputs and attendance.
- 3:53:53Students upload case readouts to an
- 3:53:54endpoint instead of DMing pat weekly due
- 3:53:56dates and reminders so non-submitters go
- 3:53:57out through Gmail or beehive. I think
- 3:53:59this is something that I've wanted to do
- 3:54:01and that we talked about a lot and that
- 3:54:03makes a lot of sense based on all these
- 3:54:05conversations. And and I also wanted to
- 3:54:06note here that Fable 5.1 analyzed 58
- 3:54:09meetings and if we go over to Astra,
- 3:54:10Astra analyzed 79 meetings. So it did
- 3:54:13get the context of 21 more meetings in
- 3:54:15here. But anyways, let's take a look.
- 3:54:17The first one was that growth depends
- 3:54:18too heavily on a few acquisition
- 3:54:19channels. Yep, same pain point I think
- 3:54:21as Fable pointed out. The second one is
- 3:54:22that the customer journey requires too
- 3:54:24much personal intervention. This
- 3:54:25appeared in undefined AI plus onboarding
- 3:54:27call reser during certification and
- 3:54:29leadership still has to reconstruct
- 3:54:30status and clarify who owns what. Okay,
- 3:54:32interesting. So these are different pain
- 3:54:34points than Fable 5.1 found. Now as far
- 3:54:36as the automation I would prioritize it
- 3:54:38came up with the same thing. A customer
- 3:54:40success workflow which would check each
- 3:54:42student's onboarding status, unanswered
- 3:54:43questions, available progress, identify
- 3:54:45missteps, prepare follow-up, all makes
- 3:54:47sense. Now in this case I think Astra
- 3:54:50won again like before I even thought
- 3:54:52about the cost which is absolutely
- 3:54:53absurd. Astra still won because I liked
- 3:54:56its recommendation about the leadership
- 3:54:58stuff and ownership. And I think that's
- 3:54:59something that you really want to make
- 3:55:01sure you have figured out within a team.
- 3:55:03But anyways, if you look here, Astra was
- 3:55:065 1/2 bucks and Fable 5.1 here was 46
- 3:55:08bucks, which just makes no sense to me.
- 3:55:10I'm I'm very very confused on how it
- 3:55:12cost that much. I had it look at the
- 3:55:13session logs. I also did a SL usage and
- 3:55:15this one cost 46 bucks even though the
- 3:55:17active agent time was just 6 minutes. So
- 3:55:20I'm very confused on where those tokens
- 3:55:22went. Like I don't understand why. It
- 3:55:23also analyzed less meetings in Astra.
- 3:55:25But anyways, that's pretty interesting.
- 3:55:27This winner is definitely Astra. So now
- 3:55:29we are sitting at 3 to2 Astra to Fable.
- 3:55:32All right, so moving on to number six.
- 3:55:33This one's super interesting to me. I
- 3:55:34gave it this folder which has over 150
- 3:55:37GB of resources and recordings from our
- 3:55:40AIS live event in July. And I asked them
- 3:55:42to make us basically a recap, like an
- 3:55:45energetic reel, 60 seconds that shows
- 3:55:47like a recap of the event that we can
- 3:55:49use to promote the next ones and things
- 3:55:50like that. Now, I'm really impressed
- 3:55:51with both of these outputs. Let me just
- 3:55:53say because with how much information it
- 3:55:55had to look through and how it had to
- 3:55:57create a story and figure out how to put
- 3:55:59everything together with music, I think
- 3:56:01they both did really really well. So,
- 3:56:02here is Fable 5.1.
- 3:56:04>> Hello. Hello, AIS Live. My Let's Let's
- 3:56:07get some energy going in here.
- 3:56:09>> Oh my god, I'm so excited for this. You
- 3:56:11guys are throwing out a fantastic event.
- 3:56:13>> Literally like logged off yesterday just
- 3:56:15buzzing. I had goosebumps. Let's Let's
- 3:56:18get you guys paid.
- 3:56:27This was an experience that didn't feel
- 3:56:29like a webinar.
- 3:56:43[music]
- 3:56:47It's been a great few days.
- 3:56:50>> [music]
- 3:57:02[music]
- 3:57:04>> Okay, so that was Fable 5.1. Let's take
- 3:57:06a look at Astra's version. This has been
- 3:57:08so energizing.
- 3:57:15[music]
- 3:57:19Yesterday I did a workshop where I went
- 3:57:22all in on [snorts] every little tool
- 3:57:23that I use. [music]
- 3:57:30Cloud code is my primary driver and then
- 3:57:32I built like my entire second
- 3:57:34brain/personal agent on top of it.
- 3:57:38>> I could go to bed, come back and it'll
- 3:57:39still be working, you know, 10 hours
- 3:57:41later to try to achieve that goal.
- 3:57:49>> [music]
- 3:57:52>> It's great to be here. You guys have
- 3:57:53done a a fantastic event. Everything's
- 3:57:55been super smooth.
- 3:57:59You guys have been such an amazing
- 3:58:00audience today.
- 3:58:03[music]
- 3:58:07Now, this one, like I said, is really
- 3:58:08tough. I think that they both did
- 3:58:09different things well. I think that
- 3:58:11Asher did a better job of putting like
- 3:58:13live clips in there. I think that Fable
- 3:58:15did a better job of kind of like gearing
- 3:58:17it up for the next event in October.
- 3:58:19There were a few moments where there
- 3:58:20were some things that were off. Like
- 3:58:22Fable had a few moments where the the
- 3:58:23timing and the voices were felt like not
- 3:58:26very synced. Astra had one weird moment
- 3:58:28where it was showing an a video of Russ,
- 3:58:30but it said Devon Karns and it was just
- 3:58:31like that was clearly off. But I am
- 3:58:33going to give the slight edge here to
- 3:58:34Astra. But I will say I'm very very
- 3:58:36impressed by both of those outputs. I
- 3:58:38thought it was given the vague goal,
- 3:58:40they both did a really good job of
- 3:58:42exploring and creating a story. But
- 3:58:44anyways, this one is going to go to
- 3:58:45Astra because you can see this stuff
- 3:58:46here. Astra took 30 minutes whereas
- 3:58:48Fable took 50 and Astra cost 16 bucks
- 3:58:51and Fable cost 26 bucks here. Now, next
- 3:58:54we did sizzle reels. I basically gave it
- 3:58:56an example of a scissor reel that I
- 3:58:57really liked or sort of like a sass I
- 3:58:59don't know a motion graphic a motion
- 3:59:00design video that I liked and I told it
- 3:59:02to analyze what was good about it and I
- 3:59:04told it to use that as inspiration and
- 3:59:06create us one for AIS. So let me show
- 3:59:08you guys these outputs. First one is
- 3:59:09Fable 5.1.
- 3:59:15[music]
- 3:59:24[music]
- 3:59:32[music]
- 3:59:43>> [music]
- 3:59:50[music]
- 3:59:58>> Okay, now let me play you Astro's
- 4:00:00output. Heat. Heat. N. [music]
- 4:00:13[music]
- 4:00:19[music]
- 4:00:32Okay, so Astro wins this one. I mean, I
- 4:00:35will say that music is a little bit
- 4:00:36annoying. The but besides that, it took
- 4:00:39screenshots from AIS. Like, it went in
- 4:00:40here and it grabs pictures of the actual
- 4:00:42courses. Where was the other one? It
- 4:00:44took a picture here, which I thought was
- 4:00:45just like I love that idea that it's
- 4:00:47thinking, okay, why don't I actually go
- 4:00:48get real proof and screenshots from the
- 4:00:50thing I'm actually building a reel about
- 4:00:52and put it in there. I also love this
- 4:00:54like these 3D designs. I don't know, it
- 4:00:56just felt like it had depth. You can see
- 4:00:57the shadows here as it kind of, you
- 4:00:59know, zooms in. I thought that this as
- 4:01:01far as the energy that it had was much
- 4:01:04better and much higher energy than the
- 4:01:05first one from Fable 5.1. Now, let's
- 4:01:07take a look at the cost and time. This
- 4:01:09one was about similar time, but Fable
- 4:01:125.1 was, you know, a little over half.
- 4:01:14So, 16 bucks compared to 26 bucks. But
- 4:01:17once again, I still think that in this
- 4:01:18case, I'm going to take that output from
- 4:01:20Astra as the winner. So, moving on to
- 4:01:22number eight. This one's a little more
- 4:01:23out there as far as I probably wouldn't
- 4:01:24really be doing this every day for work,
- 4:01:26but I had them build me these games. So,
- 4:01:28we have two different games where you're
- 4:01:29a pistachio and you're trying to escape
- 4:01:31the kitchen. So, I'm just going to go
- 4:01:32through these real quick and you're
- 4:01:33going to tell me which one you thought
- 4:01:35was, you know, sort of like which model.
- 4:01:37So, you can see here it's it's uh it's
- 4:01:39pretty smooth. I can double jump. I'm
- 4:01:41collecting these like serial bits and I
- 4:01:42have to hit these checkpoints to get to
- 4:01:44different spots in order to get out of
- 4:01:46the kitchen. The physics are real. Like
- 4:01:47I'm bumping into these things and I can
- 4:01:49like, you know, hit the spoon and
- 4:01:50everything like that. It's pretty easy,
- 4:01:52but you know, I I I missed a cereal, but
- 4:01:54I'm able to escape and win the game in
- 4:01:57about 43 seconds. Okay, now here is the
- 4:01:59other version. This one This one
- 4:02:01definitely looks a little bit more I
- 4:02:02don't know. It looks a little bit more
- 4:02:03premium, doesn't it? Um, this obviously
- 4:02:05this wall isn't there, but it's just
- 4:02:06because we want to be able to see like
- 4:02:07this. But this one feels a bit more
- 4:02:10premium, I'd say. It also feels smoother
- 4:02:12when I'm actually like controlling this.
- 4:02:13It is guiding me very specifically,
- 4:02:15though. I don't know. I can double jump,
- 4:02:17I think. Oh, no. I can't double jump.
- 4:02:18The physics seem to be real still. I can
- 4:02:20bump into stuff. If I can move things, I
- 4:02:22can fall into there. Um, we've got a hot
- 4:02:25stove. Oops. It's interesting though,
- 4:02:26they both decided to have like the same
- 4:02:28flow as far as you jump up on the drawer
- 4:02:31and then you you know, you jump up on
- 4:02:32books to start, you're now on the stove,
- 4:02:34you're going over to the sink. I mean,
- 4:02:36that's how most kitchens are, but still
- 4:02:38the the actual flow is pretty similar.
- 4:02:40And now we escaped in this one in 46
- 4:02:42seconds. So, similar amount of time. I
- 4:02:44did get all of the crumbs in this one.
- 4:02:45They were designed very similarly, but I
- 4:02:47will say I think that as far as like
- 4:02:49feeling realistic and you know physics,
- 4:02:51I liked this version better and this
- 4:02:53version was Astra. Anyways, look at
- 4:02:56these stats. They were very, very
- 4:02:58similar. So Astra was eight bucks
- 4:03:00cheaper, N bucks cheaper, and it also
- 4:03:02took about 40 bucks longer. So very
- 4:03:04similar stats. Nothing blew anything out
- 4:03:06of the water here, but I would say that
- 4:03:07I'm giving this one to Astra. Once
- 4:03:09again, that one's subjective. Maybe some
- 4:03:11of you guys liked Fable 5.1 more, but
- 4:03:13what I think is really important is that
- 4:03:14you're applying your own taste and
- 4:03:15you're finding your own use cases
- 4:03:16because some of these use cases, you
- 4:03:18guys might be like, "Well, I actually
- 4:03:18liked Fable 5.1 here." Cool. Then you
- 4:03:20just figured out maybe you use Fable 5.1
- 4:03:22for that use case, even though some
- 4:03:23other people might use Astra for that
- 4:03:25use case. It's all about your own use
- 4:03:26cases and your own experiments. Okay, so
- 4:03:28moving on to number nine, we have an
- 4:03:30actual SAS, like kind of a mini SAS, and
- 4:03:32it's an evaluation app. So you should be
- 4:03:35able to put in like automations and then
- 4:03:36run them against some golden data set
- 4:03:38and see the score and understand like
- 4:03:40what you need to fix and then you can
- 4:03:41rerun them and you can generate reports
- 4:03:43and things like that. So you can see
- 4:03:44here this is the first version. This is
- 4:03:46Fable 5.1. I think that right away this
- 4:03:48is a little bit more of an intimidating
- 4:03:50interface. It's very very wordy but we
- 4:03:52can see all of these tests. We can see
- 4:03:54our different automations here on the
- 4:03:55left as well. We can add a new one. So
- 4:03:57if we wanted to put the name, a
- 4:03:59description, and then actually you know
- 4:04:01say how how do you actually trigger this
- 4:04:02automation? Is it Python or do you hit
- 4:04:04an API? And then from there, let's just
- 4:04:05put in some information. You would
- 4:04:07actually put in the URL or you would put
- 4:04:09in the Python script. You would put in
- 4:04:10the test cases and then you would write,
- 4:04:12you know, how to grade it. So that is
- 4:04:13going to be very valuable. And it also
- 4:04:15has all of this stored on the back end
- 4:04:17because it has to store all of these
- 4:04:18actual um runs and it has to store the
- 4:04:21code and it has to, you know, show the
- 4:04:23improvements and things like that. So
- 4:04:24here we can see this support ticket one.
- 4:04:26Um I think they ran two tests. You can
- 4:04:28see that there's also like a visual
- 4:04:29breakdown of what this actually does. So
- 4:04:30this is a very deterministic script. We
- 4:04:32can see the code. We can see the setup.
- 4:04:34We can see the test cases. So, this is
- 4:04:36essentially our golden data set. There's
- 4:04:38only 12 examples. You probably need more
- 4:04:39than that. But this was all generated to
- 4:04:41actually test the system, right? And if
- 4:04:43I actually go in here and hit run, what
- 4:04:44does this do? You put the name, you put
- 4:04:46what changed, you put any tags, and you
- 4:04:48compare it with a different version if
- 4:04:49you want. And then I just go ahead and
- 4:04:51start the test. And so, this came back.
- 4:04:52You can see, I mean, there were no
- 4:04:53changes. So, we still got two out of the
- 4:04:5510, or sorry, two out of the 12 didn't
- 4:04:57hit, right? It also went super super
- 4:04:59fast because I, like I said, this is
- 4:05:00super simple. But then you can generate
- 4:05:01a report. So if I come in here, we can
- 4:05:03see all of the different tests that
- 4:05:05we've ran and it would show us like
- 4:05:06what's still broke and it would show us
- 4:05:08the improvements we make. So this is
- 4:05:10something that, you know, I actually do
- 4:05:11want to keep building out so we can do
- 4:05:12this for, you know, our our students and
- 4:05:14basically give them a better way to
- 4:05:15evaluate their agents and things like
- 4:05:17that when they're building them out
- 4:05:18because eval obviously we all know are
- 4:05:20super super important and it's also good
- 4:05:22to be able to see them visually. So this
- 4:05:23was Fable 5.1's version. Let's look at
- 4:05:26Astra's version. So I do like the
- 4:05:27interface of this one more as from a
- 4:05:29design perspective. So, I think that
- 4:05:30this one feels I don't know, it just
- 4:05:31feels a little bit more premium. It's
- 4:05:33also less wordy, so it's less
- 4:05:35intimidating right away. I think we have
- 4:05:37a test library, so we can see all of
- 4:05:38these different versions. We have run
- 4:05:40history. We also have reports over here
- 4:05:41that we can look at. So, we can see
- 4:05:43customer support agent. Um, I like how
- 4:05:45this report is structured. Let's see if
- 4:05:47we look at this next one. We can see how
- 4:05:49it changes over time. This one feels
- 4:05:51more branded. Like, this feels like
- 4:05:53something I'd rather get in front of
- 4:05:54some eyes and get some feedback on
- 4:05:56compared to Oops, I just closed out of
- 4:05:58the other one. compared to this version.
- 4:06:00This one just feels way more like a PC
- 4:06:02compared to this I feel like is, you
- 4:06:04know, something we can almost get out
- 4:06:05there and start testing against. So, if
- 4:06:06I come into an automation real quick,
- 4:06:08let's say we go to the support ticket
- 4:06:09router and I want to go ahead and see it
- 4:06:11visually. I like that a lot. We can see
- 4:06:13all the previous runs and I can go ahead
- 4:06:14and run this eval again and we can see
- 4:06:16it running a bit better. We're seeing it
- 4:06:18actually go through the test cases. I
- 4:06:19like this experience way, way more. And
- 4:06:22look at this. Astra came back and cost
- 4:06:25half the price of what Fable spent in
- 4:06:27order to do this. It did take a little
- 4:06:28longer, 2 hours 25 minutes compared to
- 4:06:30an hour and 20 minutes, but it just so
- 4:06:32much better for less cost. So, in this
- 4:06:35case, I'm giving this to Astra once
- 4:06:37again. So, Astra is now up, I believe, 7
- 4:06:39to2 on Fable. All right, moving on to
- 4:06:41number 10. We now have this Herk brain
- 4:06:44visual. So, this is a local host where I
- 4:06:46can see basically my brain. We see
- 4:06:48different clusters over here. Um, this
- 4:06:49seems more like YouTube videos. This
- 4:06:51seems like things are actually going on
- 4:06:52in the business. Oh, wait. We can see
- 4:06:54down here claude memory video knowledge
- 4:06:56and we can remove things or put them
- 4:06:57back in business wiki skills and agents
- 4:06:59projects and this is dynamic. It's
- 4:07:02really cool because we can see only what
- 4:07:03needs attention. We can see show all
- 4:07:05labels. We can see 612 notes 2500
- 4:07:08connections. We can analyze things. We
- 4:07:11can talk to it. We can search for
- 4:07:12things. Let's see if I search for um
- 4:07:14personal OS personal assistant openclaw.
- 4:07:18We can open up the specific things. And
- 4:07:19that is a pretty cool visual. So we can
- 4:07:21actually start to see in here inside of
- 4:07:23the brain. And as we click on different
- 4:07:25things, we see all of the connections
- 4:07:26that are linking into it. So all the
- 4:07:28relationships. So that is a pretty cool
- 4:07:30experience. It does feel pretty smooth.
- 4:07:31And I can control this. Uh not too well.
- 4:07:34Like I feel like I have to click on
- 4:07:35something to make that the center, but I
- 4:07:38I think that that's pretty cool. So that
- 4:07:40was Fable 5.1's version. Let me now open
- 4:07:42up Astra's version. Okay. So this one's
- 4:07:45a little bit different. Um I can still
- 4:07:48click on all of these things. These
- 4:07:50nodes are smaller. I don't like how that
- 4:07:53feels. I don't like this interface
- 4:07:56at all, like nearly as much. I have the
- 4:07:58same sort of vibe. I can click on
- 4:07:59business strategy or AI and ideas,
- 4:08:01videos and content, um people, projects,
- 4:08:04book and writing, all knowledge. Um this
- 4:08:07interface is more overwhelming for sure.
- 4:08:09Like I I definitely like this one more.
- 4:08:11I wanted just something simple where I
- 4:08:12can see the brain a bit more visually.
- 4:08:14And I was honestly expecting Astra to
- 4:08:16come in here and kick Fable's butt here,
- 4:08:18but I think that I like Fables more. And
- 4:08:20now let's look at the cost. Astra took
- 4:08:2240 minutes. Fable took 24. Astra and
- 4:08:25Fable basically cost the same. Fable was
- 4:08:27a little bit cheaper, but I am going to
- 4:08:28give this one to Fable. I think that
- 4:08:29Fable gave me more of what I was looking
- 4:08:30for. You know, once again, kind of a
- 4:08:32subjective grading criteria, but I think
- 4:08:34Fable won here. So, what does that put
- 4:08:35us at now? We're at 7 to3 Astra to
- 4:08:38Fable, and we have five more use cases
- 4:08:40left. So, Fable could still take the
- 4:08:42lead. All right, guys. So, this one is
- 4:08:44an HTML explainer doc. What I did is I
- 4:08:46gave it my YouTube video, the 25
- 4:08:48Grockbot concepts explained. And I told
- 4:08:50it to explain it to me simply simply
- 4:08:53simply and visually. And I told it to
- 4:08:54take screenshots. So, I said basically,
- 4:08:56give me a doc that I can read through
- 4:08:58and it feels like I actually watched the
- 4:09:00entire tutorial. So, we have all these
- 4:09:01different parts. We can see the
- 4:09:02screenshot to start. This one obviously
- 4:09:04is is kind of wordy up front, but that's
- 4:09:06okay. We see how the team fits together.
- 4:09:08So me, these executives, these
- 4:09:10operators, we can see all these
- 4:09:11different plugins and computers and
- 4:09:13skills. So that's a good little
- 4:09:14explanation at first. Now we move into
- 4:09:16the concepts. So we're taking
- 4:09:17screenshots and it is being analyzed.
- 4:09:19This is Fable 5.1 by the way. We can see
- 4:09:21the executive bots. We can see the name
- 4:09:22and job label. These so far are coming
- 4:09:25through well as far as these screenshots
- 4:09:27being cropped and annotated, which is
- 4:09:29really nice. It's taking a lot of
- 4:09:30different screenshots. We have the
- 4:09:31buttons here. We have all this. I think
- 4:09:33that this is coming out pretty well and
- 4:09:36it's showing us what we need to see from
- 4:09:37the actual video. Now, I do think this
- 4:09:40is pretty wordy. Um, I think that but it
- 4:09:42is getting very specific. Like, I do
- 4:09:44think I could give this doc to some
- 4:09:45people and they would be able to feel
- 4:09:47like they actually watched this video.
- 4:09:49It's it's cropping things. It's zooming
- 4:09:50in. It's annotating. I think that this
- 4:09:52is doing a really good job. It's even
- 4:09:54showing all of this. Yeah, I mean, this
- 4:09:55is doing a really good job here. So,
- 4:09:57let's go over to Ash's output. This one
- 4:09:59already, it looks a bit better as far as
- 4:10:00the HTML. And now let's see if the
- 4:10:03screenshots are good as well. So we have
- 4:10:05one here. It's pointing at a bot. It is
- 4:10:08pointing at the description, the
- 4:10:09executives. It's pointing at these
- 4:10:12different buttons. I will say Fables was
- 4:10:15more detailed for sure. Like this one
- 4:10:18was designed better, but Fable's
- 4:10:20screenshots were better. Like Fable took
- 4:10:21more screenshots and made more
- 4:10:23annotations. And so far, I think that
- 4:10:26that was a better learning experience.
- 4:10:28Even though this one, like I said, looks
- 4:10:29a little bit more I don't know. it it
- 4:10:30feels prettier or better. But if you
- 4:10:33were really thinking about could you
- 4:10:35hand this doc to a complete beginner and
- 4:10:37have them actually go through and feel
- 4:10:39like they're learning, I think that
- 4:10:40Fable's going to take this one. But
- 4:10:42let's look at the cost. Like I still am
- 4:10:43going to give the output to Fable, but
- 4:10:45when we look at the cost, Astra was much
- 4:10:48cheaper here. So the question is like,
- 4:10:49you know, if you used Astra and had it
- 4:10:51spend another, you know, 20 bucks or so,
- 4:10:54would it be better than Fable's output
- 4:10:55right here? And I would argue that yes,
- 4:10:57it would be better. So it was way more
- 4:10:59efficient here, but it's, you know, its
- 4:11:01V1 wasn't as good as Fable's V1 in my
- 4:11:04opinion. So Fable's going to take this
- 4:11:05one. So now we are at 7 to4 Fable or 7
- 4:11:08to4 Astra to Fable. Okay, so moving on
- 4:11:10to number 12, we have this sort of like
- 4:11:11painting example in Canva, which is a
- 4:11:14bit silly, but it does highlight the
- 4:11:16browser use and the vision really well
- 4:11:17cuz GBT6 Astra had to look at this
- 4:11:19picture, open up Canva, understand how
- 4:11:21to navigate with the different tools and
- 4:11:22things, and build this out, which it
- 4:11:24did. And I think that this does look
- 4:11:26pretty good, especially when we go over
- 4:11:28to Fable real quick and I open up
- 4:11:29Fable's output. If I go to the painting
- 4:11:31and we open up the browser here, this is
- 4:11:33literally what it gave me. And this was
- 4:11:35the reference image. I mean, this is
- 4:11:37just this is laughable. This is
- 4:11:39laughable. I I was expecting Fable 5.1
- 4:11:41to do something much better, but it like
- 4:11:44I don't know what exactly it tried to
- 4:11:45do, but this is obviously just very bad.
- 4:11:48So, no question Astro wins this one. And
- 4:11:51Astra was also cheaper. So, it's pretty
- 4:11:53ridiculous here how much Fable spent and
- 4:11:56how long it worked just to give me that
- 4:11:58crazy output. So, Astra wins. Okay, so
- 4:12:01moving on to the next example number 13.
- 4:12:03What I did is I gave Fable and Astra
- 4:12:06this doc which said, "Hey, these are
- 4:12:08this is a new course. I want you to put
- 4:12:10it into the free community as a draft
- 4:12:11and I want you to use browser use in
- 4:12:13order to do this." So, they basically
- 4:12:14opened up school. They went to the
- 4:12:16classroom and they put in these draft
- 4:12:17courses, Astra AIOS and Fable AIOS and
- 4:12:20they put in their different courses. So
- 4:12:22they were basically instructed to have
- 4:12:23the description, the key points, the key
- 4:12:25quotes, and any resources. And they had
- 4:12:27to put the video in here. And so this is
- 4:12:29Astra's output. I told them to do just
- 4:12:30the first five. And this looks really
- 4:12:32good. It had to navigate the interface.
- 4:12:33It had to basically come in here, make a
- 4:12:35new page, it had to put in the video, it
- 4:12:36had to put in the text. And it did it
- 4:12:38really well and saved everything here as
- 4:12:39a draft. Now, if we go to Fable's
- 4:12:41output, there are no videos here. This
- 4:12:44one arguably might have been structured
- 4:12:45better as far as like this looks cool
- 4:12:47and these are numbered whereas the other
- 4:12:49one with Astra, it wasn't like numbered
- 4:12:51as well. Like it just it's longer. But
- 4:12:54the big big issue here is that Fable ran
- 4:12:56into an issue and for some reason
- 4:12:58couldn't upload the video. There was
- 4:12:59something issue with having it download
- 4:13:00locally and whatnot. You know, I gave
- 4:13:02them everything it needed. I gave them
- 4:13:03the videos. I gave it the links. This is
- 4:13:05something that Ashstra did way better
- 4:13:06and I will say I've done this multiple
- 4:13:07times in my school and I've used 5.6 six
- 4:13:10soul with browser use and just in
- 4:13:11general codecs and the GBT models are
- 4:13:14significantly better with browser use.
- 4:13:16So this one I'm going to give to Astra
- 4:13:18and honestly the cost and the time
- 4:13:20wasn't too different here you know um
- 4:13:22very similar and very similar Astra was
- 4:13:24a little bit cheaper and Astra's results
- 4:13:26were much much better so this one's
- 4:13:27going to go to Astra as well. So we are
- 4:13:29now at the place where the score is
- 4:13:31Astra has nine and Fable has four. So
- 4:13:34Fable can no longer catch up. But let's
- 4:13:35look at the last two outputs as well.
- 4:13:37Okay, so for number 14, we did a website
- 4:13:39and basically what I wanted to happen is
- 4:13:41I gave it this URL and I said, "Hey, I
- 4:13:43want you to basically help me clone this
- 4:13:44website. I like the structure. I like
- 4:13:46the feel and I want you to make this for
- 4:13:48Perkform, which is the fake, you know,
- 4:13:50um, can company." So, this is a really
- 4:13:53cool scroll dynamic. I know you guys,
- 4:13:55some of you guys hate this sort of site.
- 4:13:56That's not the point. The point is,
- 4:13:58could Astra and Fable analyze this site
- 4:14:01and make one that has the same feel and
- 4:14:02the same vibe for perk? So, here is
- 4:14:05Fable 5.1's version. And it loads up. We
- 4:14:07have this, you know, 3D dynamic element.
- 4:14:09Coffee reformed. We have like some, you
- 4:14:11know, DNA looking stuff here. We keep
- 4:14:13scrolling down. We've got these coffee
- 4:14:15beans falling. We've got the can come
- 4:14:16back. We've got all of this going on. I
- 4:14:18would say that it's doing a decent job
- 4:14:20here. It's even got the huge like zoom
- 4:14:22out. It's doing a decent job of making
- 4:14:24this feel like the other site. You know,
- 4:14:26obviously there are some issues here,
- 4:14:28but I think it's doing a decent job.
- 4:14:29Like for oneshot prompt, although
- 4:14:31sometimes I get into these issues with
- 4:14:33the scroll. I mean, that's a bit of a
- 4:14:34bug, right? when you hit the bottom of
- 4:14:35the site, something weird goes on there.
- 4:14:37So, there's obviously some bugs, but
- 4:14:38it's not terrible for a oneshot. And
- 4:14:40now, let's take a look at Astro's
- 4:14:41version. We have the big can here, which
- 4:14:43is a little bit dynamic. There's
- 4:14:44something going on with this text,
- 4:14:47right? Like the coffee is out of bounds
- 4:14:49for some reason. Keep scrolling down. I
- 4:14:51do like this animation. I mean, that's
- 4:14:53pretty cool. It's a It's a coffee bean.
- 4:14:54There's definitely layers in the back.
- 4:14:56And by the way, this one, I didn't tell
- 4:14:57it to use any skills. I just said, "Hey,
- 4:14:59I want the same vibe. I want you to
- 4:15:01analyze the, you know, the original
- 4:15:02site, this one right here." and make me
- 4:15:05one like this. Right. So, anyways, we're
- 4:15:07coming through the little sort of like
- 4:15:08DNA thing. Coffee and protein together,
- 4:15:10two drinks in one can. I like that
- 4:15:12animation. This one overall does feel
- 4:15:15smoother than Fables. I do think that as
- 4:15:19far as a first pass. I like the way this
- 4:15:21one feels. We get stuck at the bottom
- 4:15:22and we'd have to scroll all the way back
- 4:15:25up to the top. Now, I do think that
- 4:15:26Fable did a better job as of actually
- 4:15:29just recreating what we saw in the
- 4:15:31original site, which was the prompt.
- 4:15:33Now, I do get stuck. There's a little
- 4:15:34bit of a there's bugs like this. Like, I
- 4:15:36get stuck. But as far as recreating
- 4:15:37this, Fables felt more similar to this,
- 4:15:40I think. So, we will give this one to
- 4:15:41Fable. Um, it kind of needs it as well,
- 4:15:44but I still think in general, when it
- 4:15:45comes to design, you guys have probably
- 4:15:47seen throughout this video, and you've
- 4:15:48seen my other videos. In general, I do
- 4:15:49think I like Astram more for design, but
- 4:15:51that was like, hey, show me what you can
- 4:15:53do. Take this URL, build me something
- 4:15:55crazy in one shot. It was kind of a
- 4:15:57weird prompt, right? But in this case, I
- 4:15:59still think that Fable won. But here's a
- 4:16:01look at the cost and the time for this
- 4:16:02example. All right. And this is the last
- 4:16:04one. Number 15. We have YouTube 12-month
- 4:16:07review and strategy. So, it basically
- 4:16:09had to look through the past year of my
- 4:16:11YouTube stuff and give me some stats.
- 4:16:12So, this is interesting. I mean, these
- 4:16:14are big docs. The one on the left is
- 4:16:15Fable. It's 25 pages. The one on the
- 4:16:16right is Astra. It's 15 pages. So, let's
- 4:16:18just slowly scroll through and see
- 4:16:20what's happening here. I like how Astra
- 4:16:22right away has, you know, this chart on
- 4:16:24new viewers and regular viewers. Whereas
- 4:16:26over here we have, you know, the year
- 4:16:28and numbers. We have a monthly review.
- 4:16:29We have total views across the past um
- 4:16:32ooh interesting. So actually what they
- 4:16:34show me first on the left side is from
- 4:16:36January to now. Whereas over here I'm
- 4:16:38getting sort of like the actual past 12
- 4:16:40months. So those stats are going to be a
- 4:16:41little bit different. It's showing me
- 4:16:42content type. It's showing me the
- 4:16:43monthly scorecard. It's showing what my
- 4:16:45best content actually does. It's showing
- 4:16:47me some packaging stuff. It's showing me
- 4:16:48what to worry about. So it says news
- 4:16:50brings the views but does not bring the
- 4:16:51subscribers. It's saying um the news
- 4:16:53packaging is drifting towards hype. The
- 4:16:55audience is saying so it's saying agency
- 4:16:57mechanics content has stopped working on
- 4:16:59YouTube. That definitely feels to be a
- 4:17:01bit true for this audience. Now it's
- 4:17:03deep diving into some packaging. So it's
- 4:17:04analyzing my my titles over here.
- 4:17:07Whereas over here when we have the
- 4:17:08packaging um what I would keep and
- 4:17:10improve the evidence retention first 30
- 4:17:13seconds who you are reaching and how. So
- 4:17:14it's looking at the age it's looking at
- 4:17:16the gender it's looking at the
- 4:17:17geography. It analyzes what the comments
- 4:17:19are asking me to do. So then at the end
- 4:17:21they kind of like put all this stuff
- 4:17:22together and they give me the next 90
- 4:17:24days and they give me some things to
- 4:17:25look at. Overall, they're very similar.
- 4:17:28I do like the way that this was packaged
- 4:17:29more by Astra. It just feels a little
- 4:17:31bit easier to read and it also feels
- 4:17:32more visual. So I'm going to give this
- 4:17:34one to Astra as well. And you can see
- 4:17:36here though, Astra did spend more and
- 4:17:38take longer. So 42 minutes compared to
- 4:17:4017 and 27 bucksish compared to 9 bucks.
- 4:17:43So Astra wins, but it was longer or
- 4:17:45slower and more expensive here. But
- 4:17:47anyways guys, that was 15. That was 15
- 4:17:48different use cases here that we scored
- 4:17:50and looked at the cost and time. Astra
- 4:17:53won 10 of them and Fable won five of
- 4:17:55them. But let's take a look at the total
- 4:17:57numbers here. I had both Astra and Fable
- 4:18:00give me these stats to make sure they
- 4:18:02were consistent. And you can see the
- 4:18:03numbers are exactly the same. Fable was
- 4:18:059 hours 35 minutes and 45 seconds of
- 4:18:07total run time here, costing us 513
- 4:18:11bucks.36.
- 4:18:12And Astro was 11 hours 19 minutes 24
- 4:18:15seconds and cost us 326 bucks.98. So in
- 4:18:19total Astro was $186
- 4:18:22cheaper but took an hour and 43 minutes
- 4:18:24longer to run across these 15 different
- 4:18:26use cases. Now I will say this was 15
- 4:18:28use cases. In general my experience has
- 4:18:31been that Astra feels quicker and feels
- 4:18:33more efficient. And also it doesn't give
- 4:18:35us that 5 hour window in codec and it
- 4:18:38doesn't give us that fable limit. You
- 4:18:39know we can just use Astra the whole
- 4:18:41week. So, I do like that we also get
- 4:18:43more inference with Astra in general
- 4:18:45because of the codec subscription. So,
- 4:18:46right now, I'm heavily leaning towards
- 4:18:48Astra for my my day-to-day and my use
- 4:18:50cases. But the other thing I want you
- 4:18:51guys to keep in mind is that GBT 5.6
- 4:18:53Soul is so so good still. It's so
- 4:18:55capable and for the majority of my
- 4:18:57knowledge work, I could use that just
- 4:18:58fine. And that even honestly might be
- 4:19:00overkill for the majority of stuff that
- 4:19:01I'm doing. So, it's not like you need to
- 4:19:03use Astra only for all your use cases.
- 4:19:05And it's not like you need to go grab
- 4:19:06Fable 5.1 for all your use cases. I'm
- 4:19:08still a very firm believer in they both
- 4:19:10have different strengths and weaknesses.
- 4:19:11They both have value and that's why I
- 4:19:13love doing things like this to find out
- 4:19:15what are the different use cases and how
- 4:19:16do they, you know, behave in different
- 4:19:18scenarios. And so yes, I am going to be
- 4:19:20using codecs more right now because
- 4:19:22that's where it currently is. But this
- 4:19:23stuff shifts so fast and I will always
- 4:19:25keep both subscriptions or multiple of
- 4:19:27both subscriptions so I can keep testing
- 4:19:28them out as things change, as my use
- 4:19:30cases change. Who knows what might
- 4:19:32happen. All right, so that was Codex
- 4:19:33versus Cloud Code and it was also Astra
- 4:19:35versus Fable and Astra and Fable are the
- 4:19:37models there. So we talked about how the
- 4:19:39harness cloud code or codeex is
- 4:19:41different than the actual model inside.
- 4:19:43So that's one thing to be aware of as
- 4:19:44far as like the difference between
- 4:19:45enthropic and open AAI and whatnot. The
- 4:19:47model is different than the harness. But
- 4:19:49besides that, I also think it's
- 4:19:51important to understand that these tools
- 4:19:53are all going to start to sort of merge
- 4:19:54into the same standards. At least I
- 4:19:56think so. Here's one really good
- 4:19:57example. Thoric from cloud code came out
- 4:19:59and tweeted and said hey we're adding
- 4:20:01support for agents.mmd inside cloud code
- 4:20:03which as we know that is what codeex
- 4:20:05looks at for sort of like the system
- 4:20:06prompts or the instructions and the
- 4:20:08point I'm trying to make here is just
- 4:20:09that is cloud code changing the harness
- 4:20:10a little bit so now everything is being
- 4:20:12even more tool agnostic so don't worry
- 4:20:14stay flexible and speaking of staying
- 4:20:16flexible I want to tell you guys about
- 4:20:17some of the mistakes and some of the
- 4:20:18lessons that I had learned when I was
- 4:20:20trying to be as flexible as possible and
- 4:20:22selling automations to tons of different
- 4:20:24businesses tons of different clients
- 4:20:26because if that's something that you are
- 4:20:27motivated to do after watching this
- 4:20:29course or maybe you're trying to start
- 4:20:30an agency already. Then let me talk
- 4:20:32about some of those lessons and the
- 4:20:33mindset shift of how you approach
- 4:20:35actually helping a business grow with AI
- 4:20:37rather than just trying to sell them
- 4:20:38some template or some automation that
- 4:20:40you tried to copy off a YouTube video.
- 4:20:42This is how you can actually provide way
- 4:20:44more way more value to clients and that
- 4:20:47means that you're going to get bigger
- 4:20:48retainers and you're going to get more
- 4:20:49referrals and you're going to get better
- 4:20:51case studies. It's just going to help
- 4:20:52your business in the long run. So
- 4:20:53excited to share with you guys what I
- 4:20:56learned here. So right now I'm at this
- 4:20:57event in Montenegro called Workless AI
- 4:20:59and there are a bunch of AI content
- 4:21:01creators, AI experts and a bunch of
- 4:21:02community members. And on the main day I
- 4:21:05closed off the day with this
- 4:21:06presentation which I'm about to share
- 4:21:08with you guys. So hope you guys enjoy.
- 4:21:09Hope you find it insightful. I'll see
- 4:21:11you guys in there. This is called Still
- 4:21:13Relevant Next Year. And it's actually
- 4:21:14kind of funny because last night I kind
- 4:21:16of switched up what I was going to talk
- 4:21:17about a little bit because yesterday in
- 4:21:19the VIP sections and I talked to so many
- 4:21:21of you guys and I just kind of had a
- 4:21:23feeling that this would resonate a bit
- 4:21:24more because whether you are I mean I
- 4:21:26think a lot of you guys are building
- 4:21:27kind of like an AI consultancy and AI
- 4:21:29agency or working with clients. Um but
- 4:21:31even if you're not I think that this is
- 4:21:32the way that you should be thinking
- 4:21:33about using AI within your own business.
- 4:21:36And um yeah so let me just although also
- 4:21:39um I hope this picture that's
- 4:21:40Montenegro, right? Okay. That'd be
- 4:21:42awkward if it wasn't. I I wasn't sure.
- 4:21:44It was Codeex. Threw it in there. So,
- 4:21:46good to know that that's Montenegger.
- 4:21:47Okay, cool. So, let me start with a
- 4:21:50little story. So, I'm I'm 23 years old
- 4:21:53at this point. I just quit [snorts] my
- 4:21:55full-time job. I just closed this deal
- 4:21:57with a client and I deliver something
- 4:22:00and and he messages me on Slack and he's
- 4:22:01like, "Hey, can we get on a call?" And
- 4:22:03this is what he says to me as soon as I
- 4:22:04hop on the Zoom. He goes, "We are not
- 4:22:06getting the value that we paid for." And
- 4:22:08this really made me think because,
- 4:22:11you know, I started to question like,
- 4:22:12did I just make a huge mistake? Am I not
- 4:22:14actually able to build these solutions?
- 4:22:16Am I not going to turn this into a
- 4:22:17business? And so I realized that the
- 4:22:19reason that this happened was because,
- 4:22:21well, first of all, I made this video
- 4:22:23which went viral. It was like a personal
- 4:22:24assistant. And so he saw that, lots of
- 4:22:26people saw that and they wanted this.
- 4:22:29But I made three big mistakes here. And
- 4:22:31that's pretty much what I want to talk
- 4:22:32about today are these three big
- 4:22:33mistakes. So the first one is that what
- 4:22:36I built for him, the personal assistant
- 4:22:38was not the constraint in his business.
- 4:22:41The second one, I guess you guys saw the
- 4:22:43third one. The second one is that I
- 4:22:45didn't choose a KPI.
- 4:22:47And the third one is that I guessed on
- 4:22:48the price. So like I said, that's what I
- 4:22:51want to talk about today. I had a big
- 4:22:53realization here before I started to
- 4:22:54scale my business and I was doing
- 4:22:55everything in the agency myself all the
- 4:22:57way from the lead genen all the way to
- 4:22:59the, you know, invoicing, all of that
- 4:23:01kind of stuff. I realized that our job
- 4:23:03is to be the ones to diagnose and prove.
- 4:23:07You know, it's one thing to just be kind
- 4:23:09of like you take the ticket, you develop
- 4:23:11whatever they're asking for. You kind of
- 4:23:12evolve from that and then you start to
- 4:23:13be the doctor and you start to diagnose
- 4:23:15the pain that they're feeling. But then
- 4:23:16you're still not done yet because you
- 4:23:18have to prove the value. Because if you
- 4:23:20don't prove the value, then how are you
- 4:23:22going to win more business? And I've got
- 4:23:24another quick story to tell you guys
- 4:23:25about in just a sec. But first, I wanted
- 4:23:27to start off with if you guys don't know
- 4:23:28me, a little bit about my background. My
- 4:23:30name is Nate. graduated from the
- 4:23:31University of Iowa in 2024 and I was
- 4:23:33studying business analytics and
- 4:23:35marketing. So don't come from a
- 4:23:36technical background although I did work
- 4:23:38with data quite a bit. So I've got a
- 4:23:39good understanding of data which was
- 4:23:41very helpful in automation. So can't lie
- 4:23:43about that. Um from there I graduated
- 4:23:45and I was working at Goldman Sachs as an
- 4:23:47analyst and I was automating a lot of
- 4:23:49stuff and what that helped me realize
- 4:23:50was that automation is not new at all.
- 4:23:54It feels new because the word AI has
- 4:23:55been put in front of it and that's what
- 4:23:56a lot of business owners are kind of
- 4:23:58waking up to. And it's funny because
- 4:24:00when I first started running my agency,
- 4:24:01I used to joke with my friends and you
- 4:24:03know some of my my partners when I
- 4:24:04started when I brought on some partners
- 4:24:06that we're not even an AI agency. We are
- 4:24:08a data infrastructure agency and then
- 4:24:09after that we're just a boring
- 4:24:11automation agency. And I think that's a
- 4:24:13really good realization to have because
- 4:24:15that
- 4:24:16the whole reframe of diagnosing the
- 4:24:18problem is that people might come to you
- 4:24:19and they want an AI agent because they
- 4:24:20saw a YouTube video that went viral or
- 4:24:22because they saw something on LinkedIn,
- 4:24:23but your job is to figure out what do
- 4:24:25they actually need? Um,
- 4:24:28and remember how I said the thing about
- 4:24:30you have to prove the value. I was
- 4:24:32building a lot of automations at Goldman
- 4:24:33and actually you know that's on the next
- 4:24:35slide. I'll talk about that in a sec.
- 4:24:36Um, sold my agency for so my agency once
- 4:24:40we scaled it past $100,000 a month. It
- 4:24:42was purely because I was way more
- 4:24:43passionate about education. Um, but the
- 4:24:45reason we were able to get to $100,000 a
- 4:24:47month is because the minimum to work
- 4:24:48with us was a $20,000 a month retainer.
- 4:24:50And the only reason we were able to do
- 4:24:52those retainers is because of this exact
- 4:24:53kind of like three-step framework of
- 4:24:55working with clients. Um, I've got a
- 4:24:57YouTube channel. It's pretty much the
- 4:24:58top of funnel. That's where I am doing a
- 4:25:00lot of the education. And then my free
- 4:25:01community called a automation society is
- 4:25:03where we try to get people into and try
- 4:25:06to build community. So, real quick, I
- 4:25:08wanted to start off with like why did I
- 4:25:10bring up the Goldman thing? What did
- 4:25:11that teach me? Number one, like I said,
- 4:25:13boring is beautiful. Deterministic
- 4:25:14automations. They're easier to build.
- 4:25:17They are easier to evaluate and they are
- 4:25:19less risky ultimately because they're
- 4:25:21deterministic. You know the inputs and
- 4:25:23you know the outputs. Two is that big
- 4:25:25companies I was you know Hostinger is
- 4:25:27doing a lot of stuff with AI which is
- 4:25:28which is awesome but typically when I've
- 4:25:29worked with companies and when I was at
- 4:25:31Goldman companies just kind of move
- 4:25:33slower. They have a lot of change
- 4:25:34management issues and obviously in a or
- 4:25:37an institution like Goldman in that type
- 4:25:40of vertical there's a lot of regulation.
- 4:25:43And then the third thing is that you
- 4:25:44need to spotlight your value. So quick
- 4:25:46story here. I remember I was building a
- 4:25:49lot of automations. Like I said, I built
- 4:25:50a few like weekly report things. I built
- 4:25:52a few dashboards and I was on a call
- 4:25:55with my entire team. So probably 20 25
- 4:25:57people. And I was showing them this
- 4:25:58automation, you know, on a Zoom call. So
- 4:25:59your name is right there under your
- 4:26:01face, right? And I get off the call. It
- 4:26:03was also my job to send out the meeting
- 4:26:04minutes, which obviously we know AI can
- 4:26:06do now. But I sent out the meeting
- 4:26:07minutes and I got a response from one of
- 4:26:09the managers that said, "Good job,
- 4:26:11Arthur." I'm like, "Who is Arthur?" my
- 4:26:13name is right there and I sent you the
- 4:26:15meeting minutes. And that just made me
- 4:26:16realize that if you're not explaining to
- 4:26:19the business owner, to the stakeholder
- 4:26:21what your automation actually did, then
- 4:26:23it's hard to get credit for it. So, you
- 4:26:24could potentially be in a situation
- 4:26:25where you're making the business tens of
- 4:26:28thousands of dollars, but they don't
- 4:26:29actually know that or acknowledge that
- 4:26:30because you didn't prove it. And it's
- 4:26:32not bragging, it's just proving what
- 4:26:34your systems actually did. So, those are
- 4:26:36three big lessons that I had that I
- 4:26:37tried to carry with me through my AI
- 4:26:39automation agency journey.
- 4:26:42Now think about it like this. So truly
- 4:26:45the only way for a business to scale is
- 4:26:48to remove constraints. And it's very
- 4:26:50simple at a high level. Obviously
- 4:26:51there's a lot of things that you know a
- 4:26:53lot of nuance once you get into the
- 4:26:54weeds of that. But that's at the end of
- 4:26:55the day if you want a business to scale.
- 4:26:58You have to remove constraints.
- 4:27:00You might get a lot of people coming to
- 4:27:01you like I did that says, "Hey, we need
- 4:27:03an AI agent that does X, Y, and Z
- 4:27:05because they saw a viral video because
- 4:27:08their board is putting pressure on them
- 4:27:09to have an AI strategy because their
- 4:27:11competitors are posting on LinkedIn
- 4:27:12saying, "Look at this AI agent that we
- 4:27:14built that's doing all this for us." And
- 4:27:17so that makes me feel like what they're
- 4:27:18doing is they're actually just buying
- 4:27:19relief. They're buying the ability to
- 4:27:21say, "Yeah, we work with an AI agency.
- 4:27:23Yeah, we have an AI strategy. We have
- 4:27:24this agent going to production. We have
- 4:27:25blah blah blah." But that's not really
- 4:27:28what you're trying to provide. You're
- 4:27:29not trying to provide relief. if you're
- 4:27:30trying to provide real business value.
- 4:27:32So, the whole idea is to reframe. I've
- 4:27:35done
- 4:27:37probably 300 or more discovery calls and
- 4:27:40what I've realized is that they come to
- 4:27:42you and they say something, but that's
- 4:27:43almost never what they want. And so,
- 4:27:45your job is to figure out why did they
- 4:27:47book in the call because they booked in
- 4:27:48for some reason, right? They had they're
- 4:27:50feeling pain or they're feeling
- 4:27:51pressure. And so, you're have you have
- 4:27:52to figure out what is that actual pain.
- 4:27:54You kind of have to take a step back
- 4:27:55there. There's pain somewhere and that
- 4:27:57is your job. You are paid to solve the
- 4:28:00pain. And the more pain you can solve,
- 4:28:01the more money that your business will
- 4:28:03make.
- 4:28:06I think this might have might have died
- 4:28:08on me. Okay, so the reason I put this up
- 4:28:10there is because think about if you were
- 4:28:12to work with an agency that was going to
- 4:28:14help you run ads. It's super super
- 4:28:16clear. Like you know what is your budget
- 4:28:18for ads? $10,000. And you in a direct
- 4:28:21correlation because of you ran those
- 4:28:23ads, you made $50,000 in sales. That
- 4:28:26like the agency doesn't even have to
- 4:28:27prove that to you. It's very clear, but
- 4:28:29it's not super clear when it comes to
- 4:28:31automation. So, you have to make it
- 4:28:32clear.
- 4:28:35So, those are the three things I want to
- 4:28:36talk about, right? Constraints, KPIs
- 4:28:38stands for key performance indicator,
- 4:28:40and then how do you actually price this
- 4:28:42stuff so that you can clearly show this
- 4:28:44is how much you paid and this is how
- 4:28:46much you're going to make your business
- 4:28:47back.
- 4:28:49Okay, so starting off with the
- 4:28:50constraint,
- 4:28:52the only way for a business to scale is
- 4:28:54to remove constraints, to directly
- 4:28:56attack them.
- 4:28:57And at a very high level, your business
- 4:28:59is either going to be supply constrained
- 4:29:00or demand constrained. Meaning supply
- 4:29:03constrainted, too many customers coming
- 4:29:04in and you cannot service them. Demand
- 4:29:06constrained, you need business. You have
- 4:29:08no people coming in. So I really like to
- 4:29:11think of this as a pipe.
- 4:29:14The water coming into your pipe is
- 4:29:16revenue. It's cash flow. And you
- 4:29:19obviously want that pipe to be clean, no
- 4:29:21leaks, and you want lots of water to go
- 4:29:23through. So the top image we see is a
- 4:29:26pipe that has a clog in the middle. And
- 4:29:27this is probably a supply constrained
- 4:29:29business because they're having a lot of
- 4:29:31water come through, but the water isn't
- 4:29:32making it all the way to the end. Maybe
- 4:29:33it's leaking out. It's not coming out
- 4:29:35the other end of the pipe. The bottom
- 4:29:37one is one where there's no water in the
- 4:29:38pipe. And right now there's no clogs.
- 4:29:40There probably are. There there always
- 4:29:42more clogs. But right now this this
- 4:29:44business on the bottom is demand
- 4:29:45constrained and they need more water
- 4:29:46inside their pipe.
- 4:29:48So let me tell you another quick story.
- 4:29:50I had a med spa owner that came to me
- 4:29:53and she wanted a lead generation AI
- 4:29:55agent. She wanted to make more money and
- 4:29:57I saw the stat the other day. Um,
- 4:29:59couldn't be outdated by now given how
- 4:30:01fast the space moves. But the stat was
- 4:30:03basically that over 50% of all AI
- 4:30:05automations are like lead genen or sales
- 4:30:07or marketing automations. And I think
- 4:30:10that makes sense because what is the
- 4:30:11pain a lot of us are feeling is that we
- 4:30:13want to scale our monthly recurring
- 4:30:14revenue. And usually the first thing you
- 4:30:16think is like oh more leads, more
- 4:30:18clients. But in this case, the med spa
- 4:30:20owner wanted more leads. And when I
- 4:30:22really tried to dig into the pain there
- 4:30:24and figure out where the constraint was,
- 4:30:26the constraint was she ended up telling
- 4:30:29me that they had lots of leads coming in
- 4:30:31consistently, but what happened was they
- 4:30:33weren't showing up to the appointments.
- 4:30:35And then all of the people that had
- 4:30:37showed up to an appointment, there was
- 4:30:38no follow-up. So the LTV was extremely
- 4:30:40low, meaning there was a decent amount
- 4:30:41of water coming in the pipe, but it was
- 4:30:43just leaking out before it ever made it
- 4:30:44to the end. And so we realized together
- 4:30:47on this call that it would be much more
- 4:30:49valuable to the business to build
- 4:30:50reactivation systems, to build follow-up
- 4:30:53reminders on their appointments so that
- 4:30:55people actually started to show up more.
- 4:30:57And that's just a perfect example of the
- 4:30:58reframe that you have to do in order to
- 4:31:00really diagnose what the true pain is.
- 4:31:02And something interesting here is I
- 4:31:05think about a slide that says this later
- 4:31:06too. There's just a lot of power in
- 4:31:07silence. So when I asked her the
- 4:31:09question of like why did you book this
- 4:31:11call? Like what is truly the pain that
- 4:31:13you're feeling here?
- 4:31:15I stopped talking. She thought for a
- 4:31:17couple seconds, responded, and then I
- 4:31:18didn't respond yet because I I had a
- 4:31:20feeling that that wasn't true. And when
- 4:31:22you have that silent pause, they think
- 4:31:24and they feel like they need to talk
- 4:31:25more. And I'm not big into like the
- 4:31:27sales tactics, right? Like I I don't
- 4:31:28have the sales background, but that was
- 4:31:31super powerful. I would pause and they
- 4:31:32would just keep talking. And your job on
- 4:31:34the discovery calls is to talk as much
- 4:31:36as least as possible. Your job is to sit
- 4:31:38there and listen. And that's how you're
- 4:31:39able to diagnose. So that's one quick
- 4:31:41example, right? But that happened all
- 4:31:42the time. Now, if you guys were here
- 4:31:44yesterday on the VIP VIP sessions and I
- 4:31:47came around to the table, some of us dug
- 4:31:49into this, some of us didn't, but there
- 4:31:50are two questions that I love to ask
- 4:31:51business owners or even your own
- 4:31:53business, right? Like if you're trying
- 4:31:54to help automate your own business and
- 4:31:55remove constraints, that works [snorts]
- 4:31:57super super well, especially when you
- 4:31:59pair it with silence. So, the first one
- 4:32:02is if a business is
- 4:32:04supply constraint, if you had 10x the
- 4:32:07business tomorrow, what would break
- 4:32:09first? You ask them that and then you
- 4:32:11just be quiet because they'll think
- 4:32:12about the flow in chronological order
- 4:32:14from start to finish. And the key word
- 4:32:17there that I put in all caps is first
- 4:32:18because whatever breaks first is the
- 4:32:21first clog. Now, whenever you diagnose
- 4:32:23and you fix that first clog, there's
- 4:32:25going to be another one. There always
- 4:32:26is. And that's the beauty of it. And
- 4:32:27that's how you get on a retainer because
- 4:32:29you just keep solving constraints. You
- 4:32:30keep scaling the business.
- 4:32:32[clears throat] So that's really good.
- 4:32:33If there's supply constraint, if they're
- 4:32:35demand constraint, you basically just
- 4:32:36flip it. You say, "What would you do to
- 4:32:3810x the business coming in tomorrow?"
- 4:32:39How do you get 10x the amount of water
- 4:32:41into your pipe tomorrow? So, whatever
- 4:32:42they're already doing to get leads or
- 4:32:44maybe they're not doing anything to get
- 4:32:45leads, building systems to support that
- 4:32:48initiative. And
- 4:32:50that doesn't sometimes what you'll
- 4:32:52realize is as an AI consultant is what
- 4:32:54like a lot of us are calling it right
- 4:32:56now. Your job is just as much to tell
- 4:32:58them, hey, this is the problem that I'm
- 4:33:00seeing and we can solve this with no AI.
- 4:33:02So, sometimes the problem could be it's
- 4:33:04a very service-based business. they're
- 4:33:06supply constrained and so the key is to
- 4:33:08just hire more people. Now, yeah, maybe
- 4:33:10you could come in here and say, you know
- 4:33:11what, let's build you some recruiting
- 4:33:12agents and some screening agents. That
- 4:33:14helps the hiring process. That removes
- 4:33:15that constraint. But that's a really
- 4:33:17great way to think about it as well is
- 4:33:19that an AI consultant or someone that's
- 4:33:21AI native, it doesn't mean that you use
- 4:33:22AI for everything. It means that you
- 4:33:24know just as well when not to use AI
- 4:33:26because that perspective is even more
- 4:33:27valuable in a world where AI is being
- 4:33:29shoved down everyone's throats
- 4:33:32and then you just be silent.
- 4:33:34Okay. So, moving on to the second one
- 4:33:36here. We've got the KPI.
- 4:33:38So, the whole idea here is that you
- 4:33:41before you accept any payment, you want
- 4:33:44to choose one number that you're trying
- 4:33:46to move and you're not choosing this
- 4:33:47number, you're choosing it with the
- 4:33:50stakeholder.
- 4:33:51If you think back to that example
- 4:33:52earlier where the business owner said to
- 4:33:54me, "Hey, we're not getting the value we
- 4:33:55paid for." It's because we didn't do
- 4:33:57well, all of these things, but this one
- 4:33:59very specifically because I had no way
- 4:34:00to prove that a personal assistant was
- 4:34:02helping his business. He came to me, his
- 4:34:04pain was, I feel busy. What is the
- 4:34:06solution in my mind? A personal
- 4:34:07assistant. When we really dug into it,
- 4:34:10the reason he was so busy is because he
- 4:34:11was handwriting all his proposals. So,
- 4:34:12the better automation there would have
- 4:34:14been for me to help him build an AI
- 4:34:15agent that would automate the proposals
- 4:34:17or at least get them 90% of the way
- 4:34:18there. He checks it, he approves it. If
- 4:34:20we would have aligned on a number
- 4:34:22beforehand, then I know what I'm working
- 4:34:24towards. I know the result that I'm
- 4:34:26trying to drive for the business.
- 4:34:29Because productivity is really
- 4:34:30interesting.
- 4:34:32Productivity is actually moving the
- 4:34:33needle. And you can't move the needle if
- 4:34:36you don't know what you're trying to
- 4:34:37move it towards. Because a lot of us
- 4:34:39feel like you might have had a day where
- 4:34:41you sat down on your at your desk all
- 4:34:42day, eight hours straight, let's say.
- 4:34:45But maybe you watch a lot of YouTube
- 4:34:46tutorials. Maybe you played around with
- 4:34:49image generation for an hour and you
- 4:34:50felt productive because, you know, you
- 4:34:52you clock off and you you sat there for
- 4:34:54eight hours, but how many of those hours
- 4:34:55were you actually doing what you need to
- 4:34:57do to move that number forward? And for
- 4:34:59us, you know, a lot of us, you know,
- 4:35:01it's a business, so there's a lot of
- 4:35:02different north stars, but knowing what
- 4:35:04you're building towards that quarter or
- 4:35:05that day or whatever it is helps you
- 4:35:07actually be a lot more productive. And
- 4:35:09if your agents know your goals, then
- 4:35:11they can help you plan your days better
- 4:35:13to be more productive. So, something
- 4:35:15that I would like to ask is, okay, let's
- 4:35:16say we've scoped it down. We've we
- 4:35:18understand the constraint, and I forget
- 4:35:21the example I use. Um,
- 4:35:24anyways, let's say we we've scoped it
- 4:35:26down to a certain automation. Then I
- 4:35:28could ask in a perfect world, what does
- 4:35:29this look like for your business?
- 4:35:31If we push this into production and it's
- 4:35:34working perfectly, what do you want to
- 4:35:36see? And from there, you try to drill
- 4:35:38into one certain metric. So maybe it is
- 4:35:41an AI agent that sets appointments.
- 4:35:44Maybe they say, "Okay, yeah, I mean,
- 4:35:45right now we're getting like five
- 4:35:46appointments a week from our manual
- 4:35:48setters." In a perfect world, our manual
- 4:35:50setters have those 10 hours a week back
- 4:35:53to do other things, and we're booking in
- 4:35:55like 10 appointments a week. And so now
- 4:35:57I know exactly what is a success to
- 4:35:59them. So there's that's completely
- 4:36:01objective. I can go work towards that
- 4:36:02and I can prove that. So if someone says
- 4:36:04to me, "Hey, we're not getting the value
- 4:36:05we paid for. Look at the scope. This is
- 4:36:07what we agreed on. This you said that
- 4:36:08you wanted 10 appointments a week from
- 4:36:10this automation and I objectively
- 4:36:12delivered that to you." I probably would
- 4:36:14say a little bit nicer though. Okay. So
- 4:36:17that's a trackable and objective metric
- 4:36:18because a lot of this stuff is very
- 4:36:20subjective. I want a personal AI
- 4:36:22assistant so I can feel more productive,
- 4:36:23so I feel less busy. There's no way that
- 4:36:25I can actually prove that. There's no
- 4:36:26way I can prove to the business owner
- 4:36:27that he is now less busy because of what
- 4:36:29I built. So, they can't be subjective.
- 4:36:31So, like I said, you have a baseline
- 4:36:33like five weekly appointments. You've
- 4:36:35got a target like 12 weekly
- 4:36:36appointments. And now you know what
- 4:36:38you're working towards. Of course,
- 4:36:39you're going to have to leverage some of
- 4:36:40their expertise because if you know not
- 4:36:42much about the process and you try to
- 4:36:44set a baseline target of like, you know,
- 4:36:4535 appointments a week, that's very
- 4:36:47tough. And now you're not going to be
- 4:36:48able to prove that.
- 4:36:50But the idea is that each automation is
- 4:36:51having something to work towards. And
- 4:36:52like I said, if this is not you working
- 4:36:54with a client, before you go out and
- 4:36:57decide to dedicate some some of your own
- 4:36:58internal resources or budget to a
- 4:37:00project, you should probably figure out
- 4:37:01what is that one metric you want to
- 4:37:03move. A lot of times there are, you
- 4:37:05know, it it drips into other metrics
- 4:37:07too, but if you can choose one, it makes
- 4:37:08the decision a lot easier. And then you
- 4:37:10can figure out how is that actually
- 4:37:11impacting the bottom line of the
- 4:37:13business.
- 4:37:14And that's how you prove, like I said,
- 4:37:16that your automation helped.
- 4:37:18Okay, the third piece now is on the
- 4:37:20pricing side. Now, this is really
- 4:37:22interesting.
- 4:37:24I really don't think there's like a
- 4:37:26standard universal right way to price.
- 4:37:28Everyone's doing it a bit differently.
- 4:37:30Even the big firms are changing up their
- 4:37:32pricing. I've tried almost everything
- 4:37:33that you could think of. I've tried
- 4:37:35performance-based. I've tried hourly.
- 4:37:36I've tried retainers on hourly. I've
- 4:37:38tried value based. I've tried just lots
- 4:37:40of stuff.
- 4:37:42Um, the way I feel is that you should
- 4:37:46not be charging hourly. I think that a
- 4:37:49lot of people when they're getting
- 4:37:50started, I hear, you know, I keep
- 4:37:52getting nos to my price. I hear that
- 4:37:54there's imposter syndrome. And so, I
- 4:37:56think hourly to start is a really good
- 4:37:58way to get your foot in the door and win
- 4:37:59some trust or even start free. I think
- 4:38:01it's a really good way to start, but I
- 4:38:04don't think you should be building a
- 4:38:05business around your hourly pricing. And
- 4:38:07it's it's really simple. And especially
- 4:38:08if you start to think about the way that
- 4:38:10AI is supposed to make us do more in
- 4:38:11less time. So, it's all about the
- 4:38:14incentives. All right, real quick, guys.
- 4:38:17I mentioned this at the end of the
- 4:38:18presentation, but I attach this
- 4:38:20completely free PDF, which is like a
- 4:38:2230-page resource on pricing your AI
- 4:38:24solutions, which you can access for
- 4:38:25completely free inside of my free school
- 4:38:26community. The link for that is down in
- 4:38:28the description. You just hop in there,
- 4:38:30you'll click on all YouTube resources,
- 4:38:31and you'll find that pricing masterclass
- 4:38:34guide in there. So, if you want it, it's
- 4:38:36there. Let's get back to the video.
- 4:38:38Let's say you've got two developers on
- 4:38:40your team and you're providing services
- 4:38:41to businesses. You've got one developer
- 4:38:44named Sarah and she's a killer. She gets
- 4:38:46something done in 1 hour. And you've got
- 4:38:48another developer named Dan. He's good,
- 4:38:50but he's a little slower, three hours.
- 4:38:52And let's say for the sake of the
- 4:38:53example, they produce the exact same
- 4:38:55quality of solution. You would then bill
- 4:38:57for Sarah's work, your better developer,
- 4:39:00you would bill $150 to the client or I
- 4:39:02don't know, however you want to, you get
- 4:39:04the point, right? $150. And now Dan,
- 4:39:06because he took three hours, you get
- 4:39:09more money. And those incentives are
- 4:39:10completely backwards. Why would you make
- 4:39:13more money for moving slower and on a
- 4:39:16worse developer? Right? It's all about
- 4:39:19incentives. It's all about value based
- 4:39:21pricing. That this is where we got to
- 4:39:23and this is what I believe at the moment
- 4:39:25is the best way to be doing this. And
- 4:39:27what that means is yes,
- 4:39:30the same automation could be different
- 4:39:32amounts of valuable to different
- 4:39:33businesses.
- 4:39:36So you have to kind of take that into
- 4:39:38account at least when you're going very
- 4:39:39custom and you're consulting.
- 4:39:41This is the way we did it, right? Oops.
- 4:39:44A client should understand how they will
- 4:39:45receive 10x their investment. That's
- 4:39:47kind of the golden rule that I like to
- 4:39:49to give, right? Because if you can walk
- 4:39:52the stakeholder through the math,
- 4:39:55whether that is a business you're trying
- 4:39:57to help or whether that's you internally
- 4:39:59trying to win a budget because you're
- 4:40:00trying to, you know, build AI projects
- 4:40:02inhouse,
- 4:40:04you need to prove that this investment
- 4:40:05of their money will get them a 10x on it
- 4:40:08because that makes it a really hard
- 4:40:10offer to say no to. So the way that we
- 4:40:13would do this is basically take 10% of
- 4:40:15the projected annualized value of the
- 4:40:17price of the the automation and that's a
- 4:40:21gear price. So an example here, let's
- 4:40:23say you're building a customer support
- 4:40:25agent and the manual rep, the human rep
- 4:40:28spends about 10 hours a week responding
- 4:40:30to these tickets and this person is paid
- 4:40:3340 bucks an hour. So the cost of the
- 4:40:34business 40 bucks an hour for customer
- 4:40:36support. 40 bucks an hour and the rep is
- 4:40:40does this about 10 hours a week. That's
- 4:40:41400 bucks a week. And if you annualize
- 4:40:43that, that's about $2,100 over the
- 4:40:46course of the year, the first year
- 4:40:47projected annualized revenue.
- 4:40:50And you take 10% of that. Now, 10% 20%,
- 4:40:53I play with it a little bit, but let's
- 4:40:55just say 10% for the sake of example,
- 4:40:56that's $2,80
- 4:40:59here. Because what you have to assume,
- 4:41:02they don't always,
- 4:41:04but you have to assume that they're
- 4:41:05going to ask you,
- 4:41:08can you please tell me how you got to
- 4:41:09that number? Because a lot of people
- 4:41:10will come in and they're kind of like
- 4:41:12maybe they're shopping around. They're
- 4:41:13trying to get a quote. They don't even
- 4:41:14know how much to allocate a budget
- 4:41:15towards. So if you assume that they're
- 4:41:18going to ask you, "How did you get to
- 4:41:19that number?" And you can confidently
- 4:41:20walk them through the math and then you
- 4:41:22just be silent and you don't devalue
- 4:41:23your work. I'm kind of jumping ahead. I
- 4:41:25know I got some slides here that cover
- 4:41:26this, but anyways, you have to figure
- 4:41:28out what does the manual process cost
- 4:41:30the business. And so in this example,
- 4:41:32right, I I was able to understand, okay,
- 4:41:34this rep cost the business 40 bucks an
- 4:41:36hour. They're not always going to tell
- 4:41:37you salary. They're not always going to
- 4:41:38tell you the hourly weight rate, but
- 4:41:42there's other ways to figure out what
- 4:41:44does the manual process cost the
- 4:41:45business. It could be opportunity cost.
- 4:41:46Maybe they've been having a different
- 4:41:49funnel or trip wire that they've wanted
- 4:41:51to push out for two quarters, but they
- 4:41:52haven't because they are too busy,
- 4:41:54right? There's there's other types of
- 4:41:55cost besides just like the hourly cost.
- 4:41:57Maybe you're building a voice
- 4:41:58appointment, uh, a voice agent because
- 4:42:01the business misses calls all the time
- 4:42:03and every missed call could potentially
- 4:42:06be um, a driveway, right? A driveway and
- 4:42:09that driveway could win the business
- 4:42:11$10,000. So, there's other ways to
- 4:42:12figure out what is the cost of the
- 4:42:15manual process. You just have to ask
- 4:42:17good questions and then be silent and
- 4:42:19listen,
- 4:42:21like I said.
- 4:42:23So, the key thing to remember here is
- 4:42:25that cost doesn't justify price.
- 4:42:27value justifies price. And
- 4:42:30this was kind of tough for me to wrap my
- 4:42:32head around at first because the first
- 4:42:33thing I thought of as a very new
- 4:42:35business owner was I need to protect my
- 4:42:37margins here. I need to make sure that
- 4:42:38I'm making this much money or this
- 4:42:40percent of the deal and I'm
- 4:42:42[clears throat] keeping that much to
- 4:42:42account for taxes to account for, you
- 4:42:44know, what I have to pay my employees to
- 4:42:46account for our API cost or our service,
- 4:42:48whatever it is. I was thinking about the
- 4:42:49margins, but I think that obviously you
- 4:42:52still have to think about the margins so
- 4:42:53you're not running your business into
- 4:42:54the ground, but the cost doesn't justify
- 4:42:55the price. And a really good example
- 4:42:57that I heard from Jonathan Stark who
- 4:42:59spoke at our event a couple weeks ago,
- 4:43:01he was saying like let's say you're
- 4:43:02paying someone to do your lawn and you
- 4:43:04pay them a hundred bucks for the lawn
- 4:43:07and he's been come, you know, he's been
- 4:43:08coming to you for for years. All of a
- 4:43:10sudden he buys a new pickup truck and
- 4:43:12that's an expense to his business. So he
- 4:43:13has higher costs on his end and then he
- 4:43:15comes to you and wants to charge you 200
- 4:43:17for the exact same job for the exact
- 4:43:18same lawn. That's not something that
- 4:43:19people will typically agree to because
- 4:43:21the cost doesn't justify the price. The
- 4:43:22value justifies the price.
- 4:43:25So, like I said, you have to pretend and
- 4:43:27you have to assume that as soon as you
- 4:43:28say the price, they're going to come
- 4:43:29back to you and ask, "How did you get
- 4:43:31that number?" And then you just have to
- 4:43:32walk them through the math. Now, what's
- 4:43:34really important there is you don't want
- 4:43:36to devalue your work. So, let's say
- 4:43:37someone comes to you, you scope out the
- 4:43:38right constraint, you figure out the
- 4:43:39cost of the business, and you give them
- 4:43:41a 10% number. Let's say it just comes
- 4:43:44out to 15,000. And they were basically
- 4:43:46planning 7,000 to be the maximum they
- 4:43:48were going to spend on AI this quarter.
- 4:43:51You don't want to get into this
- 4:43:52negotiation of devaluing your work and
- 4:43:54saying, "Ah, you know, this would be
- 4:43:5515K, but for you, we'll do it for 7K."
- 4:43:58You know, if you're really trying to get
- 4:43:59your first client and you can get that,
- 4:44:00then I would say do it reps over cash
- 4:44:02upfront when you're getting started. But
- 4:44:05you don't want to devalue your work.
- 4:44:06What I would do in this situation is I
- 4:44:08would say, "Okay,
- 4:44:10this build, this solution is $15,000
- 4:44:12worth of value to the business and you
- 4:44:14have a $7,000 budget. Let's just drill
- 4:44:16down the functionality a little bit and
- 4:44:18treat this as a V1. we'll give you the
- 4:44:20$7,000 worth of this automation and then
- 4:44:22you know next quarter when you've got
- 4:44:23the budget or once this automation's in
- 4:44:24production and we're actively earning
- 4:44:26you back time and money then we'll look
- 4:44:28at the next one because I think if you
- 4:44:30if you set the impression right away
- 4:44:32that you're willing to bump down your
- 4:44:34price as soon as they push back then
- 4:44:36they're just going to keep doing that
- 4:44:37and you don't want to devalue your work.
- 4:44:41So that's pretty much what I wanted to
- 4:44:44run through today. Um, obviously those
- 4:44:46kind of three main things, there's a lot
- 4:44:47more nuance to get into, which if you
- 4:44:49guys want to chat more, I'd love to to
- 4:44:50chat more.
- 4:44:51>> But think about this with your own
- 4:44:53business. Um, actually, let's just take
- 4:44:56let's just take one volunteer real
- 4:44:57quick. Someone that Yeah. Um, is it
- 4:45:00Allejo?
- 4:45:02>> There we go. Tell us
- 4:45:05in your business, you know what you're
- 4:45:07driving towards, right?
- 4:45:08>> Yeah.
- 4:45:09>> At the highest level, are you supply
- 4:45:11constrainted or demand constrained?
- 4:45:12>> Demand constraint. need more clients.
- 4:45:14>> You need more clients. And how are you
- 4:45:15currently getting those clients?
- 4:45:16>> Uh content on YouTube. Uh it's not quite
- 4:45:20converting. I'm doing webinars and out
- 4:45:24of 20 people that have assisted webinars
- 4:45:26over two months,
- 4:45:28>> uh we've gotten zero appointments set.
- 4:45:30>> Okay. So,
- 4:45:33if you wanted to have 10x the amount of
- 4:45:35business tomorrow, what would you have
- 4:45:36to do?
- 4:45:38>> Uh probably 10x the content.
- 4:45:41>> Mhm. 10x the content. uh seems crazy but
- 4:45:46the methods that I see today would be
- 4:45:48very helpful.
- 4:45:49>> Yeah, volume negates luck, right? But
- 4:45:51you want to 10x the content without
- 4:45:53decreasing the the quality of course. So
- 4:45:56in this case,
- 4:45:59obviously there's a lot more to dive
- 4:46:00into here, but you know the way that
- 4:46:04you're kind of you know your top of
- 4:46:05funnel. What does it seem like you
- 4:46:07should be focusing on automating when
- 4:46:08you get home in a couple days or next
- 4:46:10week?
- 4:46:11Um the problems that I have heard on
- 4:46:14sales calls uh turning those into a
- 4:46:18video.
- 4:46:19>> Okay.
- 4:46:19>> Uh that one is huge is I like forget
- 4:46:22what we talked about
- 4:46:24>> and maybe focus on something in a sales
- 4:46:26call but I'm going to miss something
- 4:46:27else that they said that is an actual
- 4:46:29problem.
- 4:46:31>> Totally. And do you know how you could
- 4:46:33automate that?
- 4:46:34>> Now I do.
- 4:46:35>> There we go. Um now something I saw I
- 4:46:38remember talking to you about was the
- 4:46:39grill me skill, right? Yeah,
- 4:46:41>> if you guys don't know the grill me
- 4:46:42skill, I mean, it's a very simple
- 4:46:43prompt. Essentially, just at the end of
- 4:46:45of questions, ask the AI to interview
- 4:46:49you relentlessly about this topic
- 4:46:51because what that does is it gets all of
- 4:46:53the subject matter expertise in your
- 4:46:54head into the AIOS that you're building
- 4:46:57because once you give it all of that
- 4:46:58nuance, it's able to build an automation
- 4:47:00out of that so much better because the
- 4:47:01way that you would kind of create these
- 4:47:03AI agents to to uh, you know,
- 4:47:06operationalize that process is different
- 4:47:07than someone else might. So obviously
- 4:47:10that was a really quick example. Thank
- 4:47:11you Allejo. I appreciate the theuh
- 4:47:13participation there. But just constantly
- 4:47:15be asking yourself like what is the
- 4:47:16actual constraint? You know how do we
- 4:47:1810x the business? How do we actually
- 4:47:19look through the flow of the pipe and
- 4:47:21the water and figure out where the clog
- 4:47:23is or where the leak is and then just
- 4:47:25directly attack that? And so like I said
- 4:47:27earlier
- 4:47:28when we were working with businesses we
- 4:47:30would always start with our foot in the
- 4:47:32door with one project on a value based
- 4:47:35uh project fee. And from there, we
- 4:47:37started to explain to them,
- 4:47:40we are seeing all of these other clogs
- 4:47:42because once we've removed this
- 4:47:43constraint, there's another one. And
- 4:47:45that's a good thing because still you're
- 4:47:46getting more water through. And that's
- 4:47:48how you're able to get on the retainer
- 4:47:49because not only did you
- 4:47:52find the right constraint, you also
- 4:47:54chose a number beforehand so you can
- 4:47:55objectively prove to them the automation
- 4:47:57helped their business. And then you
- 4:47:59price it in a way where it was a
- 4:48:00no-brainer for them to say yes. And then
- 4:48:02guess what? after you push it into
- 4:48:03production, you come back 1 month, 2
- 4:48:05months, 3 months later and you show them
- 4:48:07that number and now you're able to
- 4:48:09justify asking for that retainer and to
- 4:48:11just say, "Hey, you know, we can move a
- 4:48:13lot quicker if we're not always doing
- 4:48:14the scoping and we're not always
- 4:48:15figuring out pricing, if we can just
- 4:48:17work more and just keep attacking those
- 4:48:19constraints." And that's the only way
- 4:48:20that the business is truly going to
- 4:48:21grow. So, that's pretty much what I want
- 4:48:24to talk to you guys about today. I have
- 4:48:25a full pricing guide. This is like a 29
- 4:48:2730 page doc. Um, kind of just like about
- 4:48:30what I talked about today on the
- 4:48:31valuebased pricing and some stuff like
- 4:48:32that. Um, so go ahead and scan that and
- 4:48:35grab that doc. I'm sure we can include
- 4:48:36it in the the resources later as well if
- 4:48:39you miss it. But yeah, thank you guys so
- 4:48:41much. It's been an awesome day and yeah,
- 4:48:43let's let's have some dinner.
- 4:48:46So, I hope you guys enjoyed the speech
- 4:48:47that I gave as well as that free pricing
- 4:48:50guide resource. I also wanted to say
- 4:48:51that I have a second channel right here
- 4:48:53where you can see this was actually the
- 4:48:54workless event that I just showed you
- 4:48:56guys that speech from. And I have the
- 4:48:57second YouTube channel where I kind of
- 4:48:59show some more behind the scenes stuff
- 4:49:00when I go to AI events or when I do
- 4:49:02other cool things in the space and just
- 4:49:04showing you like the other side. So I
- 4:49:05will link this YouTube channel down in
- 4:49:07the description if you want to check it
- 4:49:08out. But now before we wrap up this
- 4:49:09whole course, let me dive into even more
- 4:49:11about how you actually think about
- 4:49:12pricing these solutions with valuebased
- 4:49:14pricing. So let's hop into the last
- 4:49:16video of the course. All right, so I've
- 4:49:18sold over a 100 AI automation systems
- 4:49:20and I've priced a ton of those wrong.
- 4:49:22I've undercharged, I've underscoped, and
- 4:49:24I've just thrown out random numbers that
- 4:49:26were kind of a guess and I couldn't
- 4:49:28explain when someone actually asked me
- 4:49:29like, "Hey, where'd you get that number
- 4:49:30from?" So, in this video, I'm going to
- 4:49:32talk about everything that I know about
- 4:49:34pricing AI solutions. I'm going to walk
- 4:49:35you through one real build that I sold,
- 4:49:37every single number, and by the end of
- 4:49:39this video, you'll be able to take any
- 4:49:40project, turn the client's own numbers
- 4:49:42into a price that you can actually
- 4:49:43defend, and then get paid in stages, so
- 4:49:45you're never carrying much more than,
- 4:49:47you know, 30 days of unpaid work. And if
- 4:49:49you don't know who I am, my name is
- 4:49:50Nate. I've been teaching hundreds of
- 4:49:52thousands of people how to build AI
- 4:49:53agents and how to implement them into
- 4:49:54businesses as well. And I scaled my
- 4:49:57agency to over $100,000 a month, and
- 4:49:59then I sold it. With our business, we
- 4:50:01got to the point where the minimum to
- 4:50:02work with us was a $20,000 a month
- 4:50:04retainer. And I'm assuming that sort of
- 4:50:06engagement is where a lot of you guys
- 4:50:07want to end up getting to. So, let's not
- 4:50:10waste any time and let's just jump
- 4:50:11straight into today's video. Okay. So,
- 4:50:13this example was an appointment setting
- 4:50:15agent. The business had employees
- 4:50:17manually setting these meetings and this
- 4:50:19was about 20 leads a week, each one
- 4:50:21taking roughly an hour of a human's
- 4:50:23time. And those employees were costing
- 4:50:25the business about 40 bucks an hour all
- 4:50:27in. So 20 hours a week at $40 an hour is
- 4:50:30about $800 bucks a week. And if you
- 4:50:32multiply that by 52, which is 52 weeks
- 4:50:34in the year, that's $41,600
- 4:50:36annualized. And before I said anything
- 4:50:38to the client about price, I walked them
- 4:50:41through the entire solution. You know,
- 4:50:42how the agent would work, what testing
- 4:50:44looked like, how it would change their
- 4:50:45speed to lead, and their lead quality.
- 4:50:47But I also had to ask them a ton of
- 4:50:49questions through what we call the
- 4:50:50discovery phase. Because some clients
- 4:50:52want to talk about price right away, but
- 4:50:54the honest answer to them is, you know,
- 4:50:56in order for me to give you an accurate
- 4:50:57ballpark estimate of what this is going
- 4:50:59to cost, I really need to understand
- 4:51:00more in depth the complexity of the
- 4:51:02system and how it will actually look
- 4:51:04once it's fully integrated into your
- 4:51:06actual business processes. Then I priced
- 4:51:08the build right around 13% of what that
- 4:51:10annual number was, which came out to
- 4:51:125,500 bucks. So, the business would
- 4:51:15basically be paying $5,500 for a system
- 4:51:18that over the course of the year was
- 4:51:19going to give them back $41,600,
- 4:51:22which comes out to about a 7.5 multiple
- 4:51:25on their initial investment. Now,
- 4:51:27typically, I like to say the golden rule
- 4:51:28is that you need to be able to show the
- 4:51:30client that their investment is going to
- 4:51:3210x over the course of the year because
- 4:51:34that math makes it really hard to say no
- 4:51:36to. So, in this specific example, where
- 4:51:38I accounted for that extra 2.5x that we
- 4:51:40were missing that would put us at 10 was
- 4:51:43the baseline. So right now the baseline
- 4:51:44was about 20 leads a week. But as the
- 4:51:47business earned more time back because
- 4:51:49of the system and this whole process
- 4:51:51gets more streamlined that baseline of
- 4:51:53leads per week would probably start to
- 4:51:56increase right it would go to 21 and
- 4:51:57then 23 and then 27 and that's where we
- 4:51:59are earning them even more money back.
- 4:52:02Now obviously something like that is a
- 4:52:03projection and you can't just go in
- 4:52:04there and say that you can guarantee
- 4:52:06that. So be careful about making any
- 4:52:08guarantees or trying to tie revenue to
- 4:52:10those specific results or anything like
- 4:52:11that. But the general idea is that as
- 4:52:14the system gets used more, the business
- 4:52:16will grow, which in turn utilizes the
- 4:52:19system even more. So you create this
- 4:52:20really cool like flywheel. And then on
- 4:52:22top of the build, we set up a standard
- 4:52:24maintenance plan. So this was 400 bucks
- 4:52:26a month. And I want to be clear what
- 4:52:28that actually is because a lot of people
- 4:52:30kind of misinterpret this word
- 4:52:32maintenance. So that $400 a month isn't
- 4:52:35for me to bolt on new features every
- 4:52:37month. It's just me guaranteeing that
- 4:52:39the build keeps doing what we agreed
- 4:52:41that it would do. Meaning, if something
- 4:52:43breaks or an API changes or, you know,
- 4:52:45maybe a new model releases or some weird
- 4:52:46edge cases show up that need a little
- 4:52:48tweak in order to keep the system living
- 4:52:50up to the functionality of the scope,
- 4:52:52that would be on me and that's covered
- 4:52:54by the client's retainer fee. But new
- 4:52:57functionality is a completely separate
- 4:52:59conversation. So, with maintenance
- 4:53:00retainers, I always keep the number
- 4:53:02super simple and standard across all
- 4:53:03projects. And at this point in my
- 4:53:05career, our maintenance package was 400
- 4:53:07bucks a month. You just do want to be
- 4:53:09careful though because if you're
- 4:53:10delivering a system that's, you know,
- 4:53:11let's say $30,000. If maintenance on
- 4:53:14that solution is going to cost you and
- 4:53:15your team more than 400 bucks a month,
- 4:53:17meaning you would be like losing money
- 4:53:18on giving away that package, then
- 4:53:20obviously you can't charge that. But I
- 4:53:22think that with well-designed
- 4:53:24automations and well-built automations,
- 4:53:26it really shouldn't be too timeconuming
- 4:53:28for you to maintain them. because once
- 4:53:30again there's a big difference between
- 4:53:32maintaining and adding minor feature
- 4:53:34enhancements and adding you know more
- 4:53:35functionality. Now a big mistake that I
- 4:53:37made in this specific project was that
- 4:53:39once I was in production what I should
- 4:53:41have done was gone back and measured how
- 4:53:43valuable this thing truly was and I
- 4:53:45didn't do that. So I had no after state.
- 4:53:48I had no transformation number and what
- 4:53:50ended up happening is that cost me on
- 4:53:52the next conversation when I tried to
- 4:53:54win more business. So make sure you're
- 4:53:56capturing the baseline up front which in
- 4:53:58this case was you know like 20 leads a
- 4:54:00week and the speed to lead time because
- 4:54:02that was taking the humans you know time
- 4:54:04and then you follow up after a month
- 4:54:06after 2 months and after 3 months and
- 4:54:08you prove to them that those numbers are
- 4:54:09moving in the direction that the
- 4:54:11business wants because of your system.
- 4:54:12And I know that this might kind of feel
- 4:54:14like bragging, but it's not. You know,
- 4:54:16because if you don't put a spotlight on
- 4:54:18those numbers, even if your automations
- 4:54:20really really are helping the business,
- 4:54:22the business owner might not actually
- 4:54:24feel that and won't acknowledge that.
- 4:54:26So, it's really important that you're
- 4:54:27the one surfacing that stuff. Okay, so
- 4:54:29the obvious question is why not just
- 4:54:32bill hourly and be done with it? And I
- 4:54:34want to give a quick shout out here. A
- 4:54:35pricing guy named Jonathan Stark came
- 4:54:37and spoke at AIS Live and he made some
- 4:54:39really amazing points on this topic. So,
- 4:54:41I'm going to talk about some of the
- 4:54:42things that he brought up, but really
- 4:54:44the problem with hourly is that it pays
- 4:54:46you more to be slow. I mean, picture two
- 4:54:49people on your team. You're billing them
- 4:54:50the same 150 bucks an hour. Your best
- 4:54:53developer and your slowest one. Some new
- 4:54:55feature comes in and your best guy
- 4:54:56knocks that out in a day. So, you bill
- 4:54:58one day, but your slow guy takes 3 days.
- 4:55:01So, you bill 3 days. You just made three
- 4:55:04times the money off of a worse, slower
- 4:55:07employee. And if your best guy gets
- 4:55:09faster, then you're going to make less
- 4:55:10money off of him. So getting good at the
- 4:55:12job actually cut your income. It's all
- 4:55:14about incentives. Would you want to hire
- 4:55:17someone who was incentivized to work
- 4:55:19slower in order to get more money out of
- 4:55:21you? Probably not. I wouldn't want to.
- 4:55:23But now think about that in, you know,
- 4:55:252026. You get really good with cloud
- 4:55:26code or whatever tool you're using and
- 4:55:28you can do something in an afternoon
- 4:55:29that used to take you a week. And if
- 4:55:31your hourly pricing, then that afternoon
- 4:55:34because you moved so fast just slashed
- 4:55:36your income. So, I'm not saying hourly
- 4:55:38is never okay. I think for your first
- 4:55:39two or three projects when you've got,
- 4:55:41you know, nothing to point at, not a
- 4:55:42bunch of proof, then billing hourly is a
- 4:55:45safer ask. It's a it's a easier ask for
- 4:55:47the business owner and it's a good place
- 4:55:49to start. Maybe just like a hundred
- 4:55:50bucks an hour. But after that, quit
- 4:55:52billing hourly. By the way guys, if you
- 4:55:54want to access this completely free
- 4:55:55resource, which is like a 27page doc on
- 4:55:58pricing your AI services, then you can
- 4:56:00grab that, like I said, for completely
- 4:56:01free by joining my free school
- 4:56:03community. The link for that is down in
- 4:56:04the description. All you have to do is
- 4:56:05jump in here, click on classroom, go to
- 4:56:07all YouTube resources, and then find the
- 4:56:10resource in there. I promise you guys,
- 4:56:12it's in there. But let's get back to the
- 4:56:13video. So, if you're not pricing off
- 4:56:14hours, then what are you pricing off of?
- 4:56:16Well, there's three things to keep
- 4:56:18straight. So, those three words are
- 4:56:19cost, value, and price. So your cost is
- 4:56:23the floor, the number that you'd walk
- 4:56:24away if you know it was below that.
- 4:56:26Their value is the ceiling, the most
- 4:56:28that this whole thing is even worth to
- 4:56:30them in the business. And the price is
- 4:56:32any number between the two of those that
- 4:56:34you guys can land on. Now on an AI
- 4:56:36build, that gap is pretty enormous
- 4:56:38because whatever it costs you to build
- 4:56:40the thing and harden it, their ceiling
- 4:56:42is a higher they now don't need to make.
- 4:56:45And just remember this, cost doesn't
- 4:56:47justify price. Price justifies cost. And
- 4:56:50Jonathan made a really great example of
- 4:56:52a landscaper here. So let's imagine a
- 4:56:55landscaper mows your lawn every single
- 4:56:56week for a h 100red bucks. Then one week
- 4:56:58he shows up. He does the same job he's
- 4:57:00always done on the same exact lawn he's
- 4:57:01always mowed. And now he tells you it's
- 4:57:03200 bucks because he bought a fancy new
- 4:57:05truck and his costs went up. That's not
- 4:57:08how it works. Nobody accepts that. Cost
- 4:57:10doesn't justify price. Price justifies
- 4:57:13cost. Okay. So the client ceiling is the
- 4:57:16number you're actually pricing off. and
- 4:57:18you get one conversation to find it. So,
- 4:57:20as you're getting to know the business
- 4:57:21and getting to know about the
- 4:57:22automations that they need, you're
- 4:57:24trying to uncover scope just as much as
- 4:57:27you're trying to uncover the value to
- 4:57:29the business. So, you open by letting
- 4:57:31them dump everything. You know, tell me
- 4:57:33everything you know about this, what
- 4:57:34you've tried in the past, what's worked,
- 4:57:36where it keeps getting stuck. And your
- 4:57:38job is to just listen, repeat, and poke.
- 4:57:41And just run that LRP framework to get
- 4:57:42as much information as you can. Take a
- 4:57:44ton of notes, and then you pivot. Ask
- 4:57:46them about what happens after we're done
- 4:57:48with this project. You know, what is the
- 4:57:50best case scenario? And what actually
- 4:57:52changes for the business once this thing
- 4:57:53is successfully up and running? And then
- 4:57:56you run three buckets of questions. Why
- 4:57:58this? Why now? And why me? So why this
- 4:58:01is checking that what they asked for
- 4:58:04actually gets them the outcome that they
- 4:58:05want. Because a lot of the time it
- 4:58:07doesn't. And because they, you know,
- 4:58:08walked in asking for an AI agent, but
- 4:58:10the real fix might be a really simple
- 4:58:12deterministic script and a Slack
- 4:58:14notification. Then you ask why now? And
- 4:58:16this is about urgency because if a
- 4:58:18window is closing or if a competitor is
- 4:58:20breathing down their neck, then that's
- 4:58:22real money to you. And then why me is
- 4:58:24the kind of uncomfortable one, but you
- 4:58:26literally ask them, you know, why
- 4:58:28wouldn't you just do this internally?
- 4:58:29Couldn't you vibe code this? Couldn't
- 4:58:30you hand it to an intern? And the reason
- 4:58:32you want to do this is because these
- 4:58:34questions or that question specifically
- 4:58:36surfaces objections that are already in
- 4:58:37their head. So, either you can hear it
- 4:58:39now while you're on the call with them
- 4:58:41and you can answer those objections and
- 4:58:42handle them or it's going to kill the
- 4:58:44deal later down the line when they're
- 4:58:46reading your proposal or talking with
- 4:58:47their team. And when they give you a
- 4:58:49real answer like, "Yeah, you know, my
- 4:58:50nephew probably could build this." You
- 4:58:52don't argue. You go, "Yeah, he probably
- 4:58:53could get a version of this up and
- 4:58:54running." But what I'd want to know is
- 4:58:56who's going to watch it at 2 in the
- 4:58:57morning when the model updates and you
- 4:58:59know, who's going to help you scale this
- 4:59:00when you have way more throughput than
- 4:59:02you're expecting and when it gets a bit
- 4:59:04more complex? And then what you do is
- 4:59:06you stop and you just let them answer
- 4:59:07because their answer is the actual
- 4:59:09reason that they're looking to hire
- 4:59:10someone. And if you find that you're in
- 4:59:12a situation where they just keep pushing
- 4:59:14you for a number, if they keep pushing
- 4:59:15for a price before you can ask these
- 4:59:17questions and before you can really dive
- 4:59:18in, don't blurt one out. And honestly,
- 4:59:20if they keep pushing you for a price and
- 4:59:22all they want is to get a quote out of
- 4:59:23you so they can compare you across a few
- 4:59:25different other vendors, then I would
- 4:59:26say that's probably a red flag and you
- 4:59:27just want to get out of that engagement.
- 4:59:29Because the hard truth is some people
- 4:59:32are just looking for the cheapest labor.
- 4:59:34they're not really looking for a
- 4:59:35consultant or partner and that's how you
- 4:59:36want to position yourself and there's
- 4:59:38unfortunately not much you can do to
- 4:59:39change that person's mind besides
- 4:59:41telling them hey you know what I bring
- 4:59:42to the table is different I bring a
- 4:59:44different level of expertise and you
- 4:59:46know that's where you really want to
- 4:59:47leverage your proof but ultimately some
- 4:59:48people know that they could go to Upwork
- 4:59:50and get much cheaper labor and that's
- 4:59:51just what they're going to do so don't
- 4:59:53take it personally okay so the obvious
- 4:59:56problem with a lot of this is that a lot
- 4:59:58of clients won't just hand you the
- 4:59:59numbers like they don't want to tell you
- 5:00:01salaries and exact bottom line numbers
- 5:00:03and things like that and and they don't
- 5:00:04have to. So, start with three questions.
- 5:00:06How long does this take you today? How
- 5:00:08many people touch it? And what happens
- 5:00:09when it goes wrong? These are all sizing
- 5:00:11questions. They measure the ceiling, not
- 5:00:13the build. Then you keep stacking proxy
- 5:00:16questions on top, like you, oh, how many
- 5:00:17locations do you have or how many of
- 5:00:19these come in on your busiest days? That
- 5:00:21kind of stuff. And I know some of you
- 5:00:23are on a job marketplace where you have
- 5:00:25to put a number in your proposal before
- 5:00:27you even get a chance to talk to that
- 5:00:28person. And you can still kind of do a
- 5:00:30version of this. The job post almost
- 5:00:32always tells you things like volume or
- 5:00:34headcount or you can figure out some
- 5:00:35complexity there, how often the thing
- 5:00:37happens. You size it off that and say
- 5:00:39your assumption out loud. Something
- 5:00:41like, you know, I've priced this
- 5:00:42assuming that there's about 200 of these
- 5:00:44a month. And you know, if that's off,
- 5:00:46let's hop on a 15-minute call and I can
- 5:00:48adjust the price and we can rework the
- 5:00:49scope a little bit. But that basically
- 5:00:51just turns the cold price into a reason
- 5:00:52for them to hop on a call and talk to
- 5:00:54you. And here's a quick check. If you
- 5:00:56can't land on a number, if you feel like
- 5:00:58you don't understand the value enough to
- 5:00:59accurately give a number, then that's
- 5:01:01probably a signal that you're not ready
- 5:01:02to write that proposal yet. You need
- 5:01:04more information. Okay, so now you've
- 5:01:07got that value number from that total
- 5:01:09annualized value number, first year
- 5:01:11annualized. What I like to do is pick
- 5:01:13somewhere between 10% and 20% of that
- 5:01:16number as my starting point. So you say
- 5:01:18your price, you assume the next sentence
- 5:01:20out of their mouth is, can you walk me
- 5:01:22through how you got to that price? And
- 5:01:23you basically have to answer that
- 5:01:25confidently. You have to say, "Okay,
- 5:01:26yeah, this is a customer support
- 5:01:27automation. It takes a rep an hour a day
- 5:01:29manually, and an hour of that rep's time
- 5:01:31is about 50 bucks, right? Like that's
- 5:01:33how much it's costing the business over
- 5:01:35a year. That equals about $12,000 to the
- 5:01:39business. So 10% of that $12,000 is
- 5:01:42$1,200. So the build is $1,200 bucks. So
- 5:01:45conservatively, you will be 10xing your
- 5:01:47investment of that $1,200 just in the
- 5:01:50first year." And when you go to write
- 5:01:51this up, don't just send one price. What
- 5:01:53I've always done is tiered packages. So
- 5:01:55you can have like a starter, a growth,
- 5:01:57and a scale or how whatever you want to
- 5:01:59call them. And before you get anywhere
- 5:02:00near the options, the top of that
- 5:02:02proposal is three short paragraphs. And
- 5:02:04none of them are about you or your
- 5:02:05pricing. They're about where the
- 5:02:07business is right now in their words
- 5:02:08with their numbers in it, where they
- 5:02:10said they wanted to be and why you're
- 5:02:12the person that can help them get to
- 5:02:14where they want to be because of your AI
- 5:02:16expertise and your systems. So you would
- 5:02:18then take the first year value. Let's
- 5:02:20for now just call it 100 grand to keep
- 5:02:21numbers even. You could say option one
- 5:02:23is 10% so $10,000. Option two is 25% so
- 5:02:2825,000. And option three is 50% 50
- 5:02:31grand. And I'm just kind of throwing out
- 5:02:32rough numbers here. But obviously as you
- 5:02:34move up those tiers you have to have
- 5:02:36different sort of, you know, value. It
- 5:02:38still has to make sense. You have to
- 5:02:40have more functionality or you have to
- 5:02:42have different types of results tied to
- 5:02:43it, right? But the middle one is kind of
- 5:02:46the one that's designed to win because
- 5:02:47the bottom one looks thin, the top one
- 5:02:49maybe makes them wse a little bit and
- 5:02:51the middle one looks more correct. And
- 5:02:52what this does psychologically is really
- 5:02:54interesting because if you give them one
- 5:02:55number, their decision is hm should we
- 5:02:57work with this person? But if you give
- 5:02:59them three numbers, their decision is
- 5:03:00more around how should we work with this
- 5:03:02person? You know, like which one of
- 5:03:03these deals should we take? Okay, so one
- 5:03:06of the most profitable automations that
- 5:03:07I've ever seen was one of the simplest
- 5:03:09because all it did was it took a
- 5:03:10construction crew's daily phone orders
- 5:03:12and converted them into the text format
- 5:03:14that the crew already used. That was it.
- 5:03:16It was super simple. It only saved like
- 5:03:1845 minutes a day, but it helped them
- 5:03:20avoid around $12,000 a month in
- 5:03:22scheduling errors. And that second
- 5:03:24number isn't freed up hours like the
- 5:03:26earlier example was with the appointment
- 5:03:27setter. It's hard dollars that they were
- 5:03:29losing every single month because of
- 5:03:31errors, because of the inconsistency of
- 5:03:33humans. which is why the saving 45
- 5:03:36minutes a day was worth so much more
- 5:03:37here. And the reason I'm telling you
- 5:03:39guys this is because typically a good
- 5:03:40place to start is by thinking about the
- 5:03:42hours you're saving. But as you get a
- 5:03:43little more comfortable and you have a
- 5:03:44little bit more experience behind you,
- 5:03:46you start to think about the bottom line
- 5:03:47impact of the entire business. If you
- 5:03:49think back to the earlier example about
- 5:03:51the appointment setting agent, we
- 5:03:52basically only attributed the value to
- 5:03:54the time we were saving to the time we
- 5:03:55were buying back the business. But what
- 5:03:58if we would have thought about how much
- 5:03:59does a converted appointment actually
- 5:04:01make the business? So, all of these um
- 5:04:03appointments that we're helping set with
- 5:04:04our system, what if each of those closed
- 5:04:06sales was worth $5,000? We could have
- 5:04:09valued that system way higher. And if we
- 5:04:12would have communicated it that way, we
- 5:04:13probably could have charged way more for
- 5:04:15that automation. So, like I said, as you
- 5:04:17get more experience, start to think more
- 5:04:18about that. But once again, it's your
- 5:04:20job to communicate that value because
- 5:04:22the business owner isn't just going to
- 5:04:23see that like that immediately. Okay.
- 5:04:25Now, I want to talk about underscoping.
- 5:04:27This is the number one mistake that I
- 5:04:29made when I first got started. And then
- 5:04:31what this caused was, you know,
- 5:04:32timelines to get pushed and arguing
- 5:04:35about milestones, right? So, here's how
- 5:04:38I structure this kind of stuff so that
- 5:04:39that never happens. Now, this video
- 5:04:41specifically isn't about scoping. That's
- 5:04:42a different topic entirely, but this is
- 5:04:44kind of more about setting up the
- 5:04:46milestones in the correct way. So, what
- 5:04:48I would do, let's say we have a $9,000
- 5:04:50project and we want to split this up
- 5:04:52into major milestones. So, let's say we
- 5:04:54have one milestone at the halfway point
- 5:04:55and one milestone, you know, once this
- 5:04:57thing has been pushed into production.
- 5:04:58And I like to think about those as far
- 5:05:00as like what do we think we could
- 5:05:01deliver in 30 days. So each milestone is
- 5:05:0330 days apart and that's where you're
- 5:05:05going to get paid is essentially every
- 5:05:0630 days. Then what you do is you could
- 5:05:08split this into three payments if we
- 5:05:10have two major milestones. One to get
- 5:05:11started, the second one after the first
- 5:05:13milestone has been hit and the final one
- 5:05:15after the final milestone has been hit.
- 5:05:17But this only works if each milestone
- 5:05:19itself cannot be argued with. It has to
- 5:05:21be so objective. It can't be subjective,
- 5:05:23right? Like there can be no ambiguity
- 5:05:25there, which we've run into a lot. So,
- 5:05:28you know, here's what an objective one
- 5:05:29could sound like. At this milestone,
- 5:05:31there is an AI system in the business
- 5:05:32owner's hands, a PC, a proof of concept
- 5:05:34that they can actually talk to. And when
- 5:05:36the owner sends it a question, the agent
- 5:05:38pulls from the database and responds
- 5:05:39within a minute. All of that stuff is
- 5:05:41provable and it's very simple to write
- 5:05:43down and the client could go prove it
- 5:05:44itself. It's not claiming that the
- 5:05:46database is perfectly optimized yet.
- 5:05:47It's not even saying that all the
- 5:05:48answers are perfect or correct. It's
- 5:05:50just saying that it works and it pulls
- 5:05:52there and it responds and that's
- 5:05:54objective. Now, a subjective milestone
- 5:05:56maybe could be something that's
- 5:05:57obviously easy to argue like, "Oh, the
- 5:05:59inbox agent is working as expected."
- 5:06:01Okay, what's expected? You and the
- 5:06:03client could potentially go back and
- 5:06:04forth on that for weeks and you're not
- 5:06:05going to get paid and it's just going to
- 5:06:06get frustrating. And then what else
- 5:06:08could happen is they start to ask for
- 5:06:09things that weren't on the original list
- 5:06:11because it's so ambiguous and they might
- 5:06:13say, you know, can you add this? Can you
- 5:06:15add this? That's not, you know, this is
- 5:06:17a milestone. I need this functionality.
- 5:06:18And so what you want to do there is if
- 5:06:20they do start to try to scope creep,
- 5:06:21this is actually a great sign because it
- 5:06:23means that they're excited and it means
- 5:06:24that they're already starting to imagine
- 5:06:26working with you more. So the way that I
- 5:06:28handle this, I don't say, "Oh, no."
- 5:06:29Right? Like I say, yeah, it's a great
- 5:06:31idea and I could definitely see how this
- 5:06:33would add value to the system. Let's go
- 5:06:34ahead and throw this on the backlog for
- 5:06:36our version two of the project. And as
- 5:06:38you get more ideas, you can just add
- 5:06:39more stuff to the backlog because I just
- 5:06:42want to make sure that we're hitting the
- 5:06:43milestones that we've already set as
- 5:06:45quick as possible for you guys so that
- 5:06:46we can deliver as much value to the
- 5:06:48business as possible. Okay. Now, let's
- 5:06:50talk about what you do if they see the
- 5:06:52numbers that you've presented your price
- 5:06:54and they don't like them and they're
- 5:06:55saying, "Oh, you know, this is way more
- 5:06:58expensive than I thought it was going to
- 5:06:59be. I didn't really have a good gauge.
- 5:07:00This is not in our budget." So, what you
- 5:07:02want to do is say something like, "Okay,
- 5:07:04it sounds like 20 grand isn't in the
- 5:07:06budget right now. Why don't we just like
- 5:07:08reduce the scope a little bit and start
- 5:07:09with a smaller project? And once this is
- 5:07:11working and winning you guys back some
- 5:07:13time and some more business, then we can
- 5:07:15move on to the next piece together cuz
- 5:07:16we've kind of already got it scoped out.
- 5:07:18And that's how you can avoid teaching
- 5:07:20them that your prices move down every
- 5:07:21time that they frown and you're also not
- 5:07:23devaluing the work that you would be
- 5:07:25doing. You're just reducing the scope
- 5:07:27and that keeps it pretty fair. And the
- 5:07:29last thing that I wanted to address here
- 5:07:30was the other costs that typically go
- 5:07:32along with these systems which are API
- 5:07:34costs or cloud subscriptions and you
- 5:07:36know token usage things like that. Now I
- 5:07:39have always said client's account
- 5:07:41client's card every single time your fee
- 5:07:43is for design consulting building
- 5:07:46testing. Now the tokens are a utility
- 5:07:49bill and utility bills go in the
- 5:07:50client's name. I used to start off by
- 5:07:52running everything under my own billing
- 5:07:54and invoicing them each month and it
- 5:07:55just got super messy. I was babysitting
- 5:07:57the billing. I had to follow up with
- 5:07:59clients. Obviously, I built agents to do
- 5:08:00that, but still, it wasn't fun. And it
- 5:08:03also left them with no idea what they
- 5:08:04were paying for and potentially
- 5:08:06misaligned, you know, some trust. And
- 5:08:09what else you should do is probably give
- 5:08:10them an expected monthly run cost in the
- 5:08:12proposal with the volume assumption next
- 5:08:14to it. Now, obviously, not a guarantee,
- 5:08:16but an estimate of how much this thing
- 5:08:18will typically cost per month when it's
- 5:08:19fully in production and how ideally the
- 5:08:22system starts to get used more and more
- 5:08:24over time, like month over month. So the
- 5:08:26costs are probably going to scale up a
- 5:08:27little bit month over month as well.
- 5:08:29Now, one thing that you do want to think
- 5:08:30about in your pricing is that there are
- 5:08:31typically a decent amount of testing
- 5:08:33costs that go into the system. At least
- 5:08:35if you're doing it right, you're
- 5:08:36spending a lot of money testing before
- 5:08:37you push anything into production for a
- 5:08:39client. And these testing costs could
- 5:08:41genuinely be anywhere from a hundred
- 5:08:42bucks to a few thousand based on how
- 5:08:44rigorous your evals and your QA process
- 5:08:47is, which I think should be pretty
- 5:08:49pretty rigorous. So, I would usually
- 5:08:51just factor that in to our final price
- 5:08:52by just bumping it up by like a,000 or 2
- 5:08:54or $3,000 depending on the size of the
- 5:08:57automation and how much testing you
- 5:08:59think is going to go into it. Obviously,
- 5:09:00the more AI that's in there, the more
- 5:09:02autonomy, the more testing you're going
- 5:09:03to have to do. And the whole thing, the
- 5:09:05way I feel about pricing right now is
- 5:09:07that even the biggest firms, McKenzie,
- 5:09:09Salesforce, everyone's trying to figure
- 5:09:11out this AI pricing thing, and no one
- 5:09:12has the right golden answer. And I think
- 5:09:15that a lot of people might disagree with
- 5:09:16some of the things I'm saying here, and
- 5:09:18that's okay. But I didn't feel like it
- 5:09:19was a great feeling to say, "Hey, you
- 5:09:21know, Mr. or Mrs. client, can you please
- 5:09:22go ahead and give us this API key and
- 5:09:24this one and this one and this one." And
- 5:09:25then before we're even giving them any
- 5:09:27sort of P or showing them any value,
- 5:09:29we're already spending a few hundred of
- 5:09:31their dollars just testing the system. I
- 5:09:33just don't think that's a very good way
- 5:09:34to kick off a partnership. So basically
- 5:09:37what I meant by that is in the testing
- 5:09:39we paid for everything and then when we
- 5:09:41moved everything into production we
- 5:09:42swapped out their API keys for you know
- 5:09:45compute and whatever else it was that
- 5:09:48was costing us. Okay. So if you take one
- 5:09:51single thing out of this into your next
- 5:09:54discovery or sales call, make it this
- 5:09:55one. Before you say any number at all,
- 5:09:57get the client to tell you what the
- 5:09:59problem is costing them in their own
- 5:10:01words. Just have them say it out loud.
- 5:10:03And then once that number is on the
- 5:10:04table in their language, your price is
- 5:10:06just a fraction of a number that they
- 5:10:07have already stated. So what you're
- 5:10:09selling is the result and the build is
- 5:10:11just how it gets delivered. Last thing I
- 5:10:12wanted to call out for you all is that I
- 5:10:13have a road map, the exact playbook that
- 5:10:15I followed to get to $100,000 a month
- 5:10:18and exit my agency when I started as a
- 5:10:20solo founder. So if you do want to take
- 5:10:21that path and sort of like run your own
- 5:10:23oneperson agency to start and then look
- 5:10:25at scaling up, there's a free road map
- 5:10:26down in the description if you want to
- 5:10:28check that out. But anyways guys, that
- 5:10:29is going to do it for this one. I hope
- 5:10:30you enjoyed the video. video. I hope you
- 5:10:31learned something new and if you did,
- 5:10:33please give it a like. It helps me out a
- 5:10:34ton. And as always, I appreciate you
- 5:10:35guys making it to the end of the video
- 5:10:36and I'll see you on the next one.
About this transcript
This page contains the full transcript of Build & Sell with Codex (5+ Hour Course) by Nate Herk | AI Automation, generated from the public captions YouTube serves with the video. The transcript has 76,259 words across 10,410 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.