YT YouTube2Text

YouTube2Text → Transcripts → FULL 3 HOUR COURSE: Automate Your Business With Claude AI (2026)

FULL 3 HOUR COURSE: Automate Your Business With Claude AI (2026) — Transcript

by KJ Rainey · 34,465 words · 4,678 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Welcome to the most in-depth AI
0:02automations course on YouTube. By the
0:04end of this video, you're going to be
0:06extremely confident in implementing AI
0:08into [music] literally any business
0:10model. And I can only say that because
0:12I've spent the last 5 months using the
0:14exact templates I'm giving you today to
0:16automate systems in [music] 10 plus US
0:19businesses. So yeah, this will not be a
0:21light one, but it is legit. In part one,
0:23we're going to cover the principles of
0:25AI systems and then building systems
0:27with AI so that you finally understand
0:30how this technology works on a
0:32fundamental level and you can actually
0:34build valuable stuff with it. And then
0:35in part two, I'm going to give you my
0:37exact automations [music] project
0:38template. I'm going to give you my
0:40workflow setup, my agent setup, and my
0:42building process. All of which took like
0:445 months to create so that you can start
0:46building automations with this stuff
0:48today. And there's going to be three
0:49different levels to the automation. So,
0:50beginner, intermediate, and advanced.
0:52That way, you can learn from start to
0:54finish how to build these complex
0:56automations that create insane value.
0:58And then finally, in part three, I'm
0:59[music] going to break down multiple
1:01real running automations that I've built
1:03across my businesses and 10 plus other
1:06businesses so you [music] can see how
1:07this stuff creates value right now as we
1:09speak. So, I cannot wait. I have never
1:12been [music] this excited for a video
1:14that I've made. So, lock in, grab some
1:16coffee, and let's get right into it.
1:18Welcome, my friends, to part one of this
1:19course, the principles module. Now, if
1:22you really just want to skip ahead and
1:24go get the tactics and all my processes
1:26and frameworks, go ahead. It's in part
1:28two, execution. But I highly recommend
1:32you watch this. This is my secret. This
1:34is really the sauce. This is what I do
1:36not think you will find anywhere else on
1:38YouTube, at least at this moment. I hope
1:40this will be the most valuable video you
1:42watch this month, maybe even this year.
1:44That would be amazing. I pray for that.
1:46But let's jump right into it. What is
1:48AI? How do we understand it so that we
1:50can actually use it? The cool part about
1:53AI is it's like hiring a brand new
1:56employee. So to get the most out of an
1:58employee, you have to understand them,
2:00their strengths and weaknesses, and how
2:02they work. But here's the difference.
2:05When you understand how, let's say,
2:06Claude works, you don't just unlock the
2:08leverage of one employee, you unlock the
2:10leverage of thousands, infinite Clouds.
2:14So if you don't understand Claude, none
2:16of them are that useful. But as soon as
2:18you do understand AI, in this case
2:20Claude, you unlock thousands of
2:22usefulness points from him and pennies
2:25on the dollar. Like I don't I don't
2:27think we realize us normal, right?
2:30Average people, not kings, not powerful,
2:32famous, whatever. We've never had this
2:35much leverage available to us ever in
2:37the history of humanity.
2:40But you have to seek to understand.
2:43One of my favorite verses is this. The
2:45beginning of wisdom is this. Get wisdom
2:48though it cost you all you have. Get
2:50understanding. And so I don't just want
2:53to make shiny kind of cool looking stuff
2:55that doesn't work for anyone. I want to
2:58understand this technology. I want to
3:00get it so that I can leverage it and
3:02create amazing things in this world. And
3:04obviously these verses are talking even
3:06deeper about the wisdom of the Lord, the
3:08fear of the Lord. So you haven't if you
3:10haven't if you haven't read the Bible
3:12yet, highly recommend that you do. And
3:14that is the true wisdom is fear of the
3:16Lord. But it applies in this scenario as
3:18well. Don't just follow the shiny
3:20videos. Understand the technology. Don't
3:22be lazy. And you'll be crazy how much
3:25everything changes. But that being said,
3:27do not worry. I do not care about the
3:29deep, deep technicals. I don't want to
3:32build my own AI model. I want to be the
3:35driver. I want to be the Max for
3:36Stappen, the F1 driver, the fastest man
3:39in the world who can use these machines.
3:41I don't care about building them. I care
3:43about using them. Um, so I want to learn
3:45just enough to drive like crazy, nothing
3:48more. I don't care about building
3:49engine. I want to drive that sucker. But
3:51in order to drive, you do need to
3:53understand the machine that you're
3:55driving. So, what does it mean to
3:57actually drive AI? What is being an F1
4:00elite driver when it comes to AI? And I
4:03think it's as simple as this.
4:05the ability to give AI the exact inputs
4:08it needs to get the exact outputs you
4:11want. No more, no less. We'll talk about
4:14this throughout this entire course, but
4:16this is the goal. So, get used to me
4:17referring to this a lot. All right, what
4:20is AI? How does it even work? Uh,
4:23disclaimer, I am ignorant on many
4:25things, and this is a very simplified
4:27version, but like I said, it helps me
4:29drive, which is why I'm sharing it with
4:31you guys. And the way I see AI is like
4:34autocomplete on steroids. It is not a
4:36magic genie, although it does feel like
4:38it sometimes. It is a guessing machine.
4:41It is getting very good at guessing.
4:43It's getting better and better. But
4:44essentially what happens is you give it
4:47words which have meaning in real life.
4:50It guesses about what you mean and then
4:53it guesses about what you want and then
4:55it tries to give you that output. it's
4:57guess. And this can be in the form of
4:59words, code, commands, etc. But if you
5:03say, I want a dope website, it's going
5:05to guess what you mean by dope, what you
5:07mean by website, and it's going to try
5:09to give you a dope website. And so the
5:12better these models get, um, the better
5:14the outputs are going to be. But your
5:17job is this, the context. And so if you
5:20tell AI like, I did it, it has no clue
5:24what you're talking about. If you say,
5:25"I cooked it," it starts to understand a
5:27little bit more. If you say, "I finally
5:29cooked it." Or, "I finally cooked
5:31pasta." Or the final full context, "I
5:34finally cooked pasta for my late
5:37grandmother's secret pasta recipe. I got
5:39it right. My family loved it and they
5:41all teared up around the dinner table."
5:43That has meaning. That has full context.
5:46It's technically the same as I did it.
5:49But now we all know what you're talking
5:51about and the full meaning behind it.
5:53And that's the skill of AI, helping it
5:56guess better.
5:59Now, how it actually guesses is actually
6:01really cool. And if you understand this,
6:02you this will make so much more sense
6:04with AI. And there's a video called
6:06large language models explained briefly.
6:09It is highly technical. Um, I don't
6:11really care that much about this, but if
6:12you want to go watch it, I've watched it
6:14a few times. But what happens is you
6:16give it a sentence, right? Paris is a
6:18city in blank. based on the model that
6:21they've trained it on, like trillions
6:23and trillions of data. It gives it
6:25probabilities for what the right answer
6:28it thinks is. And so it's going to say
6:30France. And so this is how AI works. It
6:33says Paris is a city in France. It takes
6:36that guess. Then it feeds all of that
6:38back into AI and it guesses again. Paris
6:41is a city in France. And and then it
6:44takes all that and feeds it back into
6:45AI. And so the way it spits out every
6:48word is by reputting in the entire batch
6:53of words. That's why our goal is so
6:56important. We do we want to give AI just
6:59enough context but not too much is
7:02that's how we waste tokens and we dilute
7:04our outputs.
7:05So yeah, AI just does this. It guesses
7:07each word and then it plugs it all back
7:09in and guesses the next one. And it's
7:11getting really good at it, which is why
7:12it feels magical. But here's a super
7:15dumb down example of what we're doing.
7:17Um, this is how I understand it. Let's
7:19say your friend says, "I'm going home to
7:21walk by blank." Well, what do you reason
7:25he's going to walk? Based on your
7:27training data, your life, you're going
7:28to think probably dog. Right? You're
7:31going to say, "You know what? He's going
7:32home to walk his dog." Now, let's say
7:34that he says, "I'm going home to blank."
7:37You don't know what he's going home to
7:38do. There's not enough context. And so,
7:40you're like, "You know what? I have no
7:42clue. But he's going home. I know that
7:44much. Now, what if he gives you too much
7:47context? I'm going to the pet store. I
7:49got a cat leash, lizard food, and some
7:51dog stuff. I love my pets. Haha. I try
7:53to take care of them. Haha. I'm going
7:55home to walk my blank. And you're like,
7:57dude, I don't know. Is he walking his
7:59lizard, his cat, his dog? Who knows? But
8:01this guy's kind of weird, right? This is
8:03how most people treat AI. They feed it
8:06their entire like life story to try to
8:09make a website and it's just way too
8:11much context. Um, okay. Yeah, that
8:14wasn't so bad, was it? Right. The deep
8:16AI stuff is over. Even though I know
8:17that was not deep at all. This is what I
8:20care about. I want to know how does what
8:22I just kind of showed you change what we
8:25do. I'm not in the business of making a
8:27bunch of technical crap. I'm in the
8:28business of providing value. How do us
8:31as business owners, as entrepreneurs,
8:33use this stuff? What does this short
8:36engine lesson change how we drive AI?
8:39That's what I want to do. I want to
8:41change your behavior so you get better
8:42outputs. Well, now you should understand
8:44that this is why our goal is so
8:46important. Give the exact input, the
8:49sweet spot for what you want to get out.
8:51And this is a skill. It takes time.
8:52That's what I want to teach you. Number
8:54two, knowing where you want to go,
8:58knowing the output you want.
9:01Like it's not a binary thing, right?
9:02It's the clearer you get. So the clearer
9:04you are and exactly what you want and
9:07then how you start is extremely
9:09important. Just like how autocomplete
9:11works, the first words you type direct
9:15it down what it's going to give you the
9:16results. So if I say what is the name,
9:20it's going to give you names because you
9:21start off by asking what is the name.
9:24Basically how you start is so important
9:26because it's autocomplete on steroids.
9:29And so if you do not know where you're
9:32going and you don't get a good start,
9:34you're going to get bad results with AI.
9:36But if you have exact clarity on what
9:39you want, if you have examples of the
9:41website you want, if you're clear on
9:42everything you want included,
9:45the better chance you're going to get to
9:46go there. But if you just say, "Hey AI,
9:49give me a dope website." It's going to
9:50give you the same slop it gives
9:52everybody else. So the better output
9:53clarity, the better start you give it,
9:55the better starting prompt, the better
9:57it's going to do. And we're going to go
9:58over how to do all of that in tacticals,
10:00but these are just the principles.
10:02Number three, probably the most
10:03important takeaway here. AI is built to
10:07give you the most probable answer. It is
10:10built to give you the average answer,
10:13which means it is amazing
10:16at things that have the right answer. If
10:19you say, "What is four* 4?" There's only
10:21one right answer. And so AI is amazing
10:24at this. But on the flip side, it is
10:27awful at things where there is no one
10:29right answer. And so if you were to ask
10:32a lot like, "Hey, does this look good?"
10:34If you asked me this, I'd be like, "Heck
10:35the freak no, bro. I would not hang this
10:36on my wall if you gave me a million
10:38dollars." But this painting sold for 137
10:41mil, by the way. So, to some people,
10:44this is a dope painting. But my
10:46definition of a dope painting is not
10:48this, which is why it's really hard to
10:51do kind of subjective stuff with Claude
10:54is because we all have different
10:55meanings of the words that we're giving
10:57Claude. And based on how he was trained,
11:00he has different definitions of what a
11:02dope website looks like. And it might
11:04mismatch from your definition of a dope
11:06website. And so what does this translate
11:09to? Well, AI is incredible at pattern
11:12recognition, at research, at
11:14summarization, at searching, converting
11:17or repurposing existing kind of
11:19subjective stuff, sorting big volume,
11:22big data, repeating tasks accurately,
11:25rough drafts of creative task. It's
11:27great at that. and basic coding. Um, not
11:29the problem solving or not the like
11:31creative side of coding, but kind of
11:32just like the the objective side. Um,
11:35really anything that you can easily
11:37explain to it what right looks like and
11:39what wrong looks like with very little
11:41room for discrepancy, it is good at. But
11:44it is awful at things like creative
11:46writing. Really, anything super
11:48creative, anything meant to be
11:50experienced by humans,
11:55anything that needs to be fresh or new,
11:57it's bad at because it's literally
11:59giving you the average of everything in
12:01the past. So, it's hard for it to be
12:03new. So, music, art, content, writing,
12:07etc., although it can trick you into
12:09thinking that it's good because it's
12:12putting out the amalgamation of past
12:15creative work in the past. Um, so how do
12:19you actually like think about this in
12:20the right way when it comes to
12:21automations and and plugins? Well, it's
12:24this.
12:25When you ask yourself, will AI be good
12:27at blank? You want to imagine a slider.
12:30So, the closer the slider is to math,
12:33the better AI will be. the closer it
12:35gets to art, the worse AI is going to
12:38be. And here's the secret sauce. You
12:42don't want to rate the macro tasks. So,
12:44if you were to ask me, KJ, is AI good at
12:46YouTube thumbnails?
12:48I can't answer that as far as the macro
12:51because there's so many different um
12:53sliders for each subtask. And so, here's
12:57the real here, guys. This if you get
12:59this man, it's balling. break down the
13:02big task into smaller task and then rate
13:06each step. And also I'm going to show
13:08you guys how AI will do this for you,
13:10which is crazy in the in part two. But
13:12if you were to ask me, KJ, is is AI good
13:15at creating YouTube thumbnails? Here's
13:17what I would tell you. For for finding
13:19out thumbnails and videos on topic, yes.
13:21For analyzing, comparing which elements
13:23are best for thumbnails, yes. For making
13:25a bunch of mock-ups, yes. Um for taking
13:28the actual pictures, no. uh for
13:31composing elements, no. For finishing
13:33touches, no. These are all things that
13:35really influence the taste of the
13:36thumbnails. And so it's good for half of
13:40the thumbnail process, but not good for
13:42the other half. And that's where you can
13:44really get crazy results with AI by
13:47understanding what it's good at and what
13:48you should do. Um, so like I said, I'm
13:51going to show you guys like literally in
13:53the setup I'm giving you for this entire
13:55like course. I have Claude rate what it
13:58thinks it's good at and what it thinks
14:00you should focus on. So I'm literally
14:02going to help you guys do this with the
14:03Claude MD I'm going to give you later.
14:05But this is so important. Feel free to
14:07watch that back if you need to get this.
14:09This is the leverage with Claude is by
14:11getting that. Now number four because of
14:14number three because the fact that AI is
14:17built on averages it is not inherently
14:20creative
14:22caveat it can do some creative domain
14:25things just not great. So if everyone's
14:30using what AI considers to be a dope
14:32website then it's average. It's no
14:35longer creative. And so AI is creative
14:39if if you were the only person using AI,
14:42but then everyone uses it and it's no
14:43longer cool. It's like trends. It's kind
14:46of like uh in school when all the cool
14:49kids would like do some sort of thing
14:51and then the like not cool kids like me
14:53would catch on and start wearing, you
14:55know, the silly bands or wearing like a
14:57certain certain type of socks or
14:58something and then all the cool kids
15:00quit doing it because the uncool kids
15:01like me started doing it. that is
15:04happening exponentially fast with AI
15:07because it's just copying people. It's
15:08not really creative. Now, this this is
15:13so important. Please lock in for this.
15:16By learning how to prompt AI, you can
15:19get better outcomes that you can get
15:22more alpha. So, there's extreme alpha
15:25meaning like edge or value from learning
15:27how to context AI well. So, if you just
15:30say, "Build me a dope website," you will
15:32get bang-on mediocre average. But if you
15:37get better and better at contexting,
15:39better at driving the machine, then you
15:42can get better results that are further
15:44on the curve to being like better,
15:47right? Just imagine there's one side of
15:49this and this is like a super amazing,
15:52you know, valuable website. By doing
15:54better context, you can get better
15:56results up to a certain point. Do I ever
16:00think AI will be better than like the
16:02world's best website designer? No. No.
16:05Never. But you can get pretty good. You
16:07can get better than most people with AI.
16:10Um, and my favorite quote is this. I
16:12came up with it. I don't know. How can
16:13AI truly think outside the box when it
16:15is the dang box? Um, everything that we
16:20do as humans is an experience. Like this
16:22is why people drink beer, right? It's
16:23not because it's like freaking
16:25hydrating. It's because it makes you
16:27feel a type of way. We go to movies
16:29because we want to feel like the
16:31experience. And so since AI really is
16:34just math, it's really bad at creating
16:37things that make us feel stuff. That is
16:40the human element, the experience. And
16:42so an old tweet of mine is AI has a has
16:45about as much taste as a white rapper.
16:47Now, it's just a joke, guys. If you're a
16:49white rapper, don't get offended. But um
16:52it's a really good analogy because what
16:54do most white rappers do? They try to
16:56say as many words as possible, pronounce
16:58every word right, and it just lacks a
17:01lot of taste versus other rappers that
17:04have a more culture than us um white
17:07people. They're able to make songs that
17:09you like you sometimes you don't even
17:11know what they're saying, but they they
17:12feel epic. And so that's the problem
17:16with AI is um it lacks taste, but you
17:20can help it with context get a little
17:23bit better taste. Just know that it
17:24maxes out. And so one thing that I feel
17:27like I run into a lot is I hit a wall
17:29with AI where it's like it gets way
17:31better and then all of a sudden I just
17:33cannot get it to go past this certain
17:35level of like good. And that's because I
17:38feel like I'm hitting the wall where I
17:39just need to implement human taste. So
17:42cool. This is what this course is going
17:43to teach you to do, by the way. is to
17:44teach you to go from here being like
17:46bang- on average with AI to getting like
17:48really good results that honestly 99% of
17:51people can't get even with um AI. So
17:55here is the reminder the true value of
17:58AI does not lie in what it can do
18:01necessarily but it lies in what it can
18:03take off your plate so that you can do
18:06the stuff that really matters. The value
18:09is in the opportunity cost. So, I can
18:11spend all day, every day, really like 50
18:14plus hours on this mural board you're
18:16watching, not automated at all. I did
18:18all of it myself because this is what
18:20matters. This is the taste. This is why
18:22you're watching this video right now
18:24instead of some other AI slob is
18:26because, man, this KY dude is freaking
18:27weird. He's funny uh sometimes because
18:30he's so stupid and his mural boards are
18:32pretty entertaining even though they're
18:34kind of lame. Um, this is the power of
18:37AI. It does all this other stuff in my
18:38life so that I can make these silly
18:40mural boards. And guess what? It's kind
18:42of working, guys. Like, did uh this
18:44video actually has over 300k views now.
18:45So, love you guys. You guys are amazing.
18:48Um, all right, number five. You have to
18:52be in the loop. So, AI is getting really
18:54dang good at guessing, but it's still
18:57guessing. So what happens is if you're
19:00not in the loop when you give AI an
19:03input and it spits out an output and
19:05then that output becomes AI's next input
19:08and then that becomes another output
19:09which becomes the next input. AI is
19:11going to loop itself into slop oblivion.
19:14This is why I do not like the like
19:17slashgoal or slash loops or whatever.
19:20AI will get better over time but right
19:23now you have to be in the loop to to fix
19:26the trajectory. So you might start off
19:28in the right direction, but AI can make
19:31a bunch of slop and then get way off way
19:33off track. And the more complex your
19:36systems are, the more this is going to
19:37happen. And so do you have to use this
19:39for like, you know, oneshotting some
19:41silly little thing? No. But if you're
19:43building real systems or applications
19:45that people are going to pay money for,
19:47you got to be in the loop. You have to.
19:49And these companies that are saying that
19:51they're not in there and that these AI
19:53agents are creating everything are are
19:55just honestly lying. So there has to be
19:58a human in the loop. And when you do
19:59that, AI becomes insanely magical.
20:02Number six, AI lacks what matters most.
20:06It does not have God-given agency.
20:09Despite the name agent, it does not have
20:11agency. It does not have God-given
20:14creativity. Now, dispute me if you want,
20:16right? You believe whatever you want. I
20:19believe that God created us in his
20:21image. God is the ultimate creator. When
20:23I look at the trees, when I look at
20:24creation, I am in awe of how good our
20:27God is. The fact that we're breathing
20:29right now is a gift. And he has given us
20:32that gift to create inside of us. That's
20:35why people are able to create rockets to
20:37Mars and create amazing paintings and
20:40and videos and uh Oceans 11 movies, you
20:43know, god-given creativity. And AI does
20:47not have that. It just copies past
20:50examples of that. It is a watered down
20:53version of the beautiful state of of
20:57humanity. [snorts] And so it will never
20:59be able to replicate you. Keep in mind
21:02that. Now, if you don't put in the work
21:03and you don't do anything useful with
21:05your life, then like, yeah, AI is going
21:06to replace you. But AI can never replace
21:09these mural boards. And I'll prove that
21:10in the future because AI is not me. Um,
21:14AI is a blob of all previous humans
21:16data. And when you mix all colors
21:18together, what do you get? kind of like
21:20just dark brown or black. Um, but you
21:24were completely opposite, right? You are
21:25uniquely one of one. And so, uh, yeah,
21:28don't ever forget that. Do not be
21:29discouraged. AI cannot replace you. Um,
21:32so yeah, hope that was encouraging. All
21:34right. Now, what we just finished the
21:36first principal module. Things are about
21:38to get a little bit deeper after this.
21:39I'm not going to lie. This was kind of
21:40just a little teaser. Now, we're going
21:42to head into the principles of actually
21:44automations, which I cannot wait to show
21:47you guys. This is like literally this is
21:49my secret sauce. Um, but all this is to
21:52once again turn you into the 0001%
21:55driver. So, let's go over to principles
21:57of automation. I will see you guys over
21:59there. Let's get it. All right. Welcome
22:01to the second principle module.
22:02Principles of systems and automations.
22:05Now, apps, plugins, skills, automations,
22:08all these shiny things you hear about,
22:10they are all just systems. And when you
22:13learn how to think in terms of systems
22:15and leverage, almost anything becomes
22:18possible. Elon is a great example. That
22:20dude lives and breathes systems and
22:22leverage. And uh yeah, he's a
22:24trillionaire. So let's let's get a
22:25little closer to that. Why don't we? And
22:27we become real entrepreneurs when we
22:29stop thinking in terms of tactics and
22:32identities like, "Oh, I'm a YouTuber.
22:34I'm a copyriter. I'm a growth operator."
22:36And instead, we start thinking in
22:37problems, in systems, and leverage. And
22:40that's what I want to share with you
22:41today. You want to make money? We do
22:43that by creating value. We create value
22:45by solving problems. You create more
22:47value by solving problems easier,
22:49faster, better, cheaper. How do we do
22:51that? By building leveraged systems,
22:55plugins, automations, software. Man,
22:57when you learn how to do this,
22:58everything changes. And hopefully you
23:00see that on my channel. My videos have
23:01been blowing up all because I stopped
23:03thinking like a tactic hunter and I
23:05started thinking, how do I solve
23:07problems with the leverage of AI? I'm
23:09going to give you guys my secrets today.
23:11Okay, so let's get into it. Principle
23:14number one, systems. A system is just
23:16this. It's inputs that get processed
23:20into an output. And this is all
23:23happening inside of an environment and
23:25we have feedback from running the system
23:28to improve the system. And so it's
23:30really that um simple. The best way to
23:32look at it in my opinion is examples. So
23:35you've have let's say we're going to
23:37cook some lasagna. The output we want is
23:39an incredible tasting lasagna. That is
23:41the desired output that we're trying to
23:43get to. That is what we're building the
23:45system for. We do that by taking
23:48ingredients, we process them by like
23:50mixing them together, cooking them,
23:52doing whatever. This happens inside of a
23:55kitchen and uh after we follow the
23:57process, we get an output. How do we get
24:00a better output? Well, we do feedback,
24:02right? We taste the lasagna and we say,
24:04"hm, needs more salt or it's a little
24:06undercooked." And we can improve the
24:08output by changing the inputs. So
24:11getting better ingredients or changing
24:13the process, right? How we process them.
24:16And this all happens inside of the
24:18kitchen we're working in. And so I used
24:20to watch this show with my wife called
24:22like I think it was Next Level Chef. And
24:24if they like took an L, they had to cook
24:27in this disgusting basement. And if they
24:29won, they got to cook in the
24:31state-of-the-art kitchen. And so that
24:33shows how the environment can really
24:35play a factor on your output.
24:39What the heck does it have to do with
24:40anything, KJ? Why are you talking about
24:41lasagna? Well, this is how I think about
24:44life. This video you're watching right
24:46now, if you're watching it on YouTube or
24:47school, this is how I made this video is
24:50I have a system for creating these
24:52things. I have a desired output that I
24:55want to put out. I have inputs um that I
24:58put into the system. So, what is the
25:00thing that I'm giving? what is the like
25:02problem I'm solving? Where is the
25:04existing market demand? These are
25:06inputs. And I take those inputs and I
25:09come up with a title, thumbnail, intro,
25:12body segments. I record this using a
25:14certain tool, you know, editing tags,
25:16upload, and then my feedback are what
25:19you guys think, right? The views, the
25:21likes, the comments, the uh CTR, average
25:23view duration, CPM, and based on those
25:26metrics, I make changes, right? add this
25:29to the title, less saturation, talk
25:31slower. Y'all got y'all know I talk like
25:33a crackhead. I'm sorry. I'm working on
25:35it. But, um, yeah, this is a system. And
25:38same with my finance plugin, which has,
25:40you know, generated me tons of amazing
25:42leads and helps a lot of people. This is
25:44happening in the in the environment of
25:46Claude. So, it happens in Claude. So,
25:48how Claude works determines how I make
25:50the plug-in. And the output is a fully
25:53customizable private finance dashboard
25:56plugin. And so it adds these these kinds
25:59of things of value, right? Saves people
26:01money. That's awesome. And so I built a
26:04system to do that, right? We input the
26:06user transactions. It processes them
26:08using the plug-in. So this is the system
26:10I built to be the process. And uh yeah,
26:13the iterations are me improving the
26:14system based on your guys' comments.
26:17Pretty cool. Now, let's get a little bit
26:20deeper. I've told you guys I'm gonna
26:21give you guys the real stuff, but it's
26:23going to be kind of deep. It's my first
26:25ever application that I'm still building
26:27called Signal. And we have real paying
26:29users on this. Shout out to the person
26:31who bought the yearly plan yesterday.
26:32You're amazing. Um, but these users have
26:35a problem that requires human depth and
26:38perspective. Therefore, they want a
26:40YouTube video. So, they don't just want
26:42to ask AI, they want to see what a human
26:44has created around this problem. A good
26:47example is how to speak. So, I'm
26:50learning how to speak better for you
26:52guys to make these videos better. And
26:54that is something that is hard for AI to
26:56explain because well it can't speak. And
26:59so I wanted to solve this problem for
27:01myself of not knowing how to find the
27:04best videos. And so I give AI this
27:06input, right? So context on my problem.
27:10The output I want is the best videos to
27:12solve this problem. And the process is
27:16signal. The main you know the
27:17application is it it does context to
27:20query reasoning. it API calls um the
27:23correct parameters. Anyway, I won't get
27:25into the details, but basically I made
27:27this system to give me an output and now
27:29people pay me to use this system. Um and
27:32obviously this is just one feature
27:34inside of Signal, but this is the main
27:36core value prop and you guys are going
27:39to learn how to do this today. So um
27:41this course that you actually I decided
27:43to get this for free so no one even no
27:45one bought this. The course that you're
27:47watching is my system for creating
27:50systems which is kind of meta but that's
27:53why this course is I think so valuable
27:55and is this you are the input this
27:58course is the process so by you watching
28:01this course and doing taking the action
28:02steps this is all happening on earth by
28:04the way laws of physics you are able to
28:07then go make valuable systems that you
28:09can sell make tons of money change your
28:11life or business whatever and the
28:12feedback is what you guys tell me in the
28:14comments and the emails in the school I
28:16read everything. So, thank you guys for
28:18your feedback. Um, so couple useful
28:21notes on systems while we're here. The
28:24process section in a system is usually
28:27just a bunch of smaller systems chained
28:29together. And so, think about Tesla for
28:32example. The inputs are raw materials.
28:34The outputs are finished Tesla cars. But
28:37inside of that macro system, there's a
28:39lot of small systems for how they make
28:41the wheel wells and just all this other
28:43stuff. Another example is I have a
28:46thumbnail system inside of my YouTube
28:48system. So how I make thumbnails is a
28:51system inside of the bigger system of
28:53how I make YouTube videos.
28:55Number two, learning systems thinking is
28:58mandatory in the AI era. This is the
29:01only way computers thinks in and out.
29:04And so the better you understand
29:06systems, the better you will be with AI.
29:10Three, as powerful as systems thinking
29:11is, it will not magically give you
29:15creativity, passion, or grit required to
29:18make a system actually work. If so,
29:22every single software engineer would be
29:24a billion-dollar founder. But they
29:25aren't. And so, that's kind of the thing
29:27with AI is now that we can all be like,
29:30you know, mediocre junior developers,
29:34writing the code isn't the value really.
29:37It's knowing what to code. That's always
29:39been what determines who makes a billion
29:41dollars. And so this is important, but
29:44you still have to have the judgment.
29:46Number four, desired output, desired
29:49output, desired output. We're going to
29:51go in detail on this in a second, but do
29:53not cope with a cool process. Do not
29:56cope with a cool assembly line. What do
29:59people care about? They care about the
30:01car. The output is what matters. Okay?
30:04And sadly, this is 99% of the AI space
30:06right now. Um, and the secret is like
30:08don't don't be cool, be useful, produce
30:11value. That's my goal at least for my
30:13life. So, principle number two,
30:16leverage. Getting more output per input.
30:19This is uh entrepreneurship 101. If you
30:22don't know this one, man, you are
30:24missing out. And this is going to be
30:25hopefully life-changing for some of you
30:27guys. So, by adding different forms of
30:29leverage into your systems, you're able
30:32to get more for less. This is why
30:36everyone's freaking out about AI. It's
30:38not because the clankers are cool. It's
30:40because of the leverage that it gives
30:42us. So instead of you having to spend 10
30:45days building a website, you can send
30:48off two prompts. So this much input,
30:51insane output. Um that's why AI is so
30:55important. But in order to get the most
30:57out of AI, you have to understand what
30:58leverage is. And this is how we make big
31:01money, right? We talked about this in
31:02the last module. So, for probably the
31:05first time ever, because sadly our
31:07school system really fails us as
31:08entrepreneurs, right? I want to cover
31:10different forms of leverage and what
31:12they look like incorporated into
31:14systems. Then we'll talk about AI
31:16leverage and then we're going to
31:18incorporate them in part two of this uh
31:20entire course execution. So, yeah, it's
31:22going to get juicy.
31:25These are the three forms of leverage
31:26that I kind of bucket leverage into
31:28which helps me a lot. The first is
31:31labor. So this is using someone else's
31:33energy towards your goals. This is the
31:35oldest form of leverage, right? Back in
31:38the day when the Egyptians built the
31:39pyramids, they had tons of, you know,
31:41slaves building them. Oh lord, what is
31:42this? Quit this. Um, that is labor
31:46leverage. Other human beings working for
31:48your goals, your agency, working towards
31:51your outputs.
31:53Um, and the way I like to think about
31:55the leverage gained is energy times
31:58skill times other forms of leverage. And
32:01so if you just have an employee who's
32:02just like moving atoms around, right?
32:05Like moving things around your room, not
32:07that much skill. But if you have an
32:08employee who's incredibly skilled at say
32:11like, [snorts] you know, I don't know,
32:14editing something like shout shout out
32:15to Caleb who's editing this video. If he
32:18edits a video, he might take him an
32:20hour. A bad editor would edit the same
32:22video and they would get less views. So
32:24same input, wildly different outputs for
32:27Caleb because he has skill. And this is
32:29why people hire skilled employees. And
32:32then if those skilled employees have
32:34other forms of leverage, well then
32:36you're getting a lot of leverage out of
32:38hiring this person, right? So if you
32:40hire someone who's incredibly good with
32:41AI, like also shameless plug, we will
32:45build AI systems for you if you don't
32:46want to do this yourself. Shout out to
32:48my agency 780. Book a call with us.
32:50We'll build custom software for you. If
32:52you hire us, you get the leverage
32:54because we are using other forms of
32:56leverage. So what are those other forms?
32:59Well, the second form is copies, which
33:02sounds lame, but this is the the secret.
33:04This is the sauce. So the biggest
33:08leverage we get as as humans is copies
33:10of previous systems or solutions. And so
33:14entertainment, right, that is a solution
33:17to boredom for humans. Information is
33:20copies of different instructions for
33:22humans of how to think and act. Code is
33:25instructions for computers of how to
33:27act. AI is like code 2.0. So, it's code
33:31that can reason. So, it's kind of like a
33:33like a a just a a weird form of like
33:35labor leverage, too, because it can
33:38solve problems that it wasn't
33:39specifically coded for, which is crazy.
33:43But you accessing these models is just
33:45you accessing a copy of Claude. So
33:48someone built this model and now we all
33:51benefit as long as we pay the claw
33:53overlords $20 a month, right? And the
33:56last form is tools and machines. So
33:58these multiply physical effort. So
34:01because someone came up with the
34:03invention of a tractor, we can all
34:05benefit from having tractors as long as
34:07we pay the money. Of course, once again,
34:09that's the whole leverage right now. The
34:12cool part is digital copies leverage. So
34:16the upfront input cost of creating these
34:19copies is massive. So writing a book, me
34:22recording this video is going to take,
34:24you know, 50 hours to make and then more
34:26to record it. A lot of input, but the
34:30cost of distribution is a fraction. So
34:33this video you're watching right now
34:35could be seen by a billion people
34:38without me doing any extra work. So if
34:40YouTube is like, "Man, KJ's cooking,
34:42dude. like let's send this to everybody
34:44on Earth. I do no extra work and a
34:47billion people get this information
34:49versus back in the day if I had to teach
34:52you one-on-one, I'd have to give this
34:542-hour long talk to every single person.
34:58It is so unleveraged, but that's the
35:00power of copies. And so, the person that
35:03invented the tractor unlocked a ton of
35:05leverage. He can sell the tractor. Me
35:07making this video, uh, Kendrick
35:09recording Good Kid Mad City, right? He
35:11recorded the album once but now they can
35:13copy it and everybody can benefit from
35:15listening to that epic album. So yeah,
35:18writing the book, writing the software,
35:19training the AI model, this is all
35:21digital copies. So the effort um and
35:26input goes like this. So to go from
35:29zero, so to go from nothing to having a
35:32thing that works requires all the
35:34effort. That's literally it. And then
35:36distributing it requires so little
35:38effort in comparison. Um this is what
35:40this is like the 0ero to one effect. And
35:42if you want to see this kind of shown in
35:44a a more typical way. This is it right?
35:49This explains exponential growth is
35:51because to get something that works to
35:54build a system that gets the right
35:56output that is the hard part. Dist
36:00distributing it scaling it up that is
36:02when it gets a lot easier. It's still
36:03not easy but it is a lot easier in
36:06comparison. Um, and this is also why
36:09people so so many people fail with AI is
36:11because I will show you guys how to use
36:13AI to build the system, but AI is kind
36:16of most useful in copying what you've
36:19already built in getting more of the
36:22thing that you've already designed or
36:23built, whatever. Um, the third is money.
36:27So money just allows you to buy future
36:31instances of one and two. So with money
36:34you can hire people. with money, you can
36:35buy more copies, you know, whatever. Um,
36:38or you can lend to people for them to
36:42buy more one and two. And so if you use
36:44capital, if you use money to buy a
36:46stock, well, you are benefiting from
36:50someone else buying these things. So
36:52when you buy into the SpaceX IPO,
36:55you are giving Elon more money to buy
36:58these things and to make more of these
36:59things. And so you benefit from it if he
37:01succeeds. This is how the world works.
37:04Ah, I hope you guys are like, is this
37:07clicking? This is what no one's taught
37:09you. Seriously, this is it. This is
37:12getting past all the hype and looking at
37:14like, how do billionaires become
37:16billionaires? This is it, guys. Now,
37:19just knowing this alone is not going to
37:20make you a billionaire, clearly, or else
37:22I'd have more money. But this is the
37:24start. This is how you play the game.
37:26I'm explaining the rules to the game so
37:27that we can play it. Now, here's a goofy
37:30example that really helped me get
37:32leverage.
37:33This is Chick-fil-A. If you're not from
37:35America, man, um I I'm sorry for you
37:38because America is the greatest country
37:40on earth. But Chick-fil-A is one of the
37:43best food that we have in in America.
37:45And what's funny is my friend Luca is
37:47from London. He's actually staying with
37:48us right now. We've been taking him to
37:50Chick-fil-A almost every day because how
37:51much he loves his chicken burger. Um
37:56back on track, they use leverage in
37:58every single form. So this system to
38:02create this chicken sandwich is a
38:04they're copying a system. The recipe,
38:07the inputs, the type of chicken, how
38:09long they cook it, what ingredients that
38:12is a system that they've copied across
38:14the country. They how they build the
38:16stores is a system. The application,
38:18right? Code, software copies. Um the
38:22labor, how they train the labor using
38:24information copies,
38:27it's all leverage. And because they have
38:29all these forms of leverage, guess what?
38:32They make $22 billion with a B selling
38:36fried chicken. But it's because they
38:38spent 70 years nailing one simple output
38:42better than anyone else, a good chicken
38:45sandwich at scale. So yeah, they have
38:48scaled up like crazy and now they're
38:50doing 20 billions of dollars a year with
38:52chicken. So anytime you go to
38:53Chick-fil-A, remember this example and
38:56think about all the leverage. I actually
38:58used to have a friend who worked on the
38:59Chick-fil-A software, but he worked on
39:01the software that they had in in the
39:04store for their like assembly line of
39:06chicken. Like that's how freaking
39:07leveraged Chick-fil-A is. They're
39:09crushing it. So, it's freaking crazy.
39:12How does this all tie back to AI? Well,
39:15there has never in the history of
39:17humanity been this much leverage
39:20available to the average bloke like me
39:22and you. So because of AI, we can now
39:25achieve massive leverage like people um
39:28who always had all the power before. We
39:30can do this with AI. All right, but how
39:34how do we actually leverage AI um in a
39:38system?
39:39[sighs] So let's first talk about how do
39:41we apply stuff? Now, quick warning here.
39:44Uh this is this is a serious one. I'm
39:46about to contradict a lot of the stuff
39:48you see online and I'm I'm not saying
39:50I'm right. I I could be wrong, but this
39:52is how I view the world. This is what is
39:53this is what has served me. And uh I'm
39:56gonna piss off probably some people
39:58because I'm gonna kind of call out some
39:59some YouTubers and teachers, but I do
40:01not want to do this and like I'm better
40:02than them. It's more of like I feel like
40:05you guys are being lied to and I would I
40:07want to show you guys the truth as far
40:09as I know it. Um I could be wrong, but
40:12just just a disclaimer here. Um now I'm
40:16talking about this through the lens of
40:18value creation. So, if you care more
40:20about just building stuff for fun,
40:22right? Like you're just having a good
40:23time, disregard what I'm saying, okay?
40:26This is only through the lens of solving
40:28problems. If you're just like having a
40:30good time playing around with AI for
40:32fun, keep doing it, man. I'm not here to
40:34shame you on that. That is how you
40:35learn. But we're going to talk about
40:37value creation. Uh one of my favorite
40:38quotes from Elon is they asked him what
40:41a a young person should try to do, and
40:44he's like, "Try to be as useful as
40:45possible, and that's what I want to do."
40:47So step zero in creating a system is
40:50deciding what system to build in the
40:52first place. And this is quite honestly
40:54the hardest step. If this step was easy,
40:57every single software engineer would be
40:59a billionaire. But they aren't because
41:00they can't figure out what to build. And
41:03so Elon says the most common error of a
41:06smart engineer is to optimize a thing
41:08that should not exist. Same with
41:10YouTubers. You can replace this with any
41:12single title. Um, yeah. Here's the
41:16evidence, right? Around 90% of all
41:19startups and tech startups ultimately
41:20fail. Crazy.
41:23Here's the problem. Once again, guys, I
41:26am a sucky entrepreneur. I don't have a
41:27million dollars. Um, so I can't even
41:29speak on this really myself, but I'm
41:31trying to learn, but I have one heristic
41:34that has served me really, really well
41:36this year, which allowed me to do these
41:38amazing things. Um, so 500k views in
41:41three months in a brand new niche, first
41:43live app with real paying users and it
41:45actually works and a community with on
41:4720,000 people now since I made this
41:49mural. Um, and I know that these aren't
41:51great indicators of like actually value
41:54creative problem solved, but I can't
41:56really show you guys the the DMs and
41:58emails I get, but hopefully you're one
41:59of the people who I have created value
42:00for, so this is proof in itself.
42:03Anyways,
42:04this is my heristic. This is my
42:06threshold for evaluating whether or not
42:08I should build a system. And that is
42:11this.
42:13Is it valuable enough to me that even if
42:15no one else buys it or uses it, it was
42:18still worth the input to create it? And
42:21so this is it. If the value it adds to
42:23my life is greater than the upfront cost
42:27of creating the thing, so not just money
42:28but like time, um, then it's good. And
42:32so basically, is the 0ero to one cost
42:35worth it? even if it stays at one. So,
42:37if it doesn't go from one to a thousand,
42:39is it still worth it? And that helps me
42:41focus. Um, and this helps me verify the
42:44idea from a few different angles. So,
42:46first off, it helps me do ideas that
42:50have very low upfront time and capital.
42:53Um, so it's cheap and fast to go from
42:56zero to one because guess what, guys? I
42:58don't have a million dollars. So, I
42:59don't have time and money to invest in
43:02ideas that may or may not work.
43:04And then second, only ideas that I would
43:06actually use survive. And what's cool is
43:09the bigger the problem is in my life,
43:12the more capital and time I can allocate
43:14to solve it because the more value it
43:16adds to me, well, the higher cost it can
43:19be and it still be worth it as far as my
43:21heristic goes.
43:23Um, next, even if I have an idea that I
43:26think is dope, because it's built for
43:28me, if I don't use it after a week, then
43:30I know that it's trash and I just don't
43:32have to build it any further. And
43:34lastly, I only build it if I think the
43:37skill and insight gained would be worth
43:39it as well. So obviously building these
43:41courses doesn't really do much for me
43:44besides you know submitting cementing my
43:47like knowledge I guess
43:49but I learn the skill even greater by
43:52doing these courses which is why the
43:53only courses that I make are things that
43:54I want to learn the skill in. And what's
43:57funny about this heristic is when you
44:00only make stuff that you would actually
44:02use other people use it too. And so my
44:04videos get views and my app gets paying
44:06users because I only made it for me. uh
44:09which is kind of kind of funny. So
44:10that's the first step in deciding what
44:13system to build. Now step one once you
44:16know what to build we have to get
44:18extreme clarity on the output of the
44:21system and so you have to get very
44:23detailed what do you want to see and
44:26then also what is the deeper why that
44:28you even want that output and so um
44:30we'll go about this in the the execution
44:32like you know I'll show you guys how to
44:34get AI to interview you to answer these
44:37questions but just remember the market
44:40cares about output not your process what
44:44comes out of the machine that is what
44:47matters. I'd rather have a boring
44:49process and an incredible output than I
44:53then I would have an incredible process
44:55boring output which this module alone
44:57should be proof of this. So far we've
44:59not we've not touched Claude once. I've
45:01not built a single vibecoded slob
45:03website. Um but I care about changing
45:06the way you think. I care about this
45:08video might be boring but the outcome
45:10that you get from watching it should
45:11hopefully be incredible. And that's what
45:13I want to do. Um, there's some more
45:16stuff about this if you want, but
45:18basically like avoid these videos about
45:21notion, obsidian, like AI operating
45:23system setups because they're only
45:25building it to get views. They don't
45:27actually use these things. And if they
45:29do, the outputs that come out of them
45:30are typically not great because they're
45:32just trying to build a cool process, not
45:35the outcomes. Um, and fundamentally, a
45:38teacher can only really teach what they
45:41have done themselves. This is the
45:43problem I have at business college. Most
45:45of my teachers have never owned a
45:47business. Um, not to say that they can't
45:49teach good stuff, but I I want to listen
45:52to people who have done what I what I
45:54want to do, right? So, step two, you
45:59achieve the desired outcome by any means
46:01possible. I know this is going to piss
46:03off some people. What I mean by this is
46:06if you want to be a good YouTuber, do
46:08not use AI to make your YouTube videos.
46:12If you cannot get a video that gets a
46:15lot of views without AI, the chance of
46:17you do getting a video with views with
46:18AI is extremely low. Um, there's two
46:22types of people who are using AI. And
46:25the first type is a process worshipper.
46:28So, they say, "Hey, look, my YouTube
46:29channel is 100% automated by AI. It's
46:32crazy. It's insane. Editors are dead.
46:35YouTube has changed forever." and uh the
46:38video that they actually make in their
46:40video gets like [clears throat] no
46:41views. It's all about the process. How
46:44cool the machines are. But those who are
46:46output focused,
46:48the process sucks. It's boring. Guys, if
46:51you if you watch me make these mural
46:52boards, man, it is boring. It takes 50
46:54plus hours, but it gets an amazing
46:58value. And once again, I can't show you
47:00the actual value because views are not
47:02value, guys. You you could slap a
47:04grandmother and get a million views um
47:05and not change anybody's life. But it is
47:08hard for me to quantitate life change.
47:10So hopefully you personally have
47:11experienced value through my creations
47:13and that enough is proof alone. And once
47:16again, I am not the end all be all guys.
47:17I suck at this. I'm just trying to
47:18explain how I view the world. Do not be
47:21process blind. This is what I mean by
47:23this.
47:24Most people look at the the like
47:26assembly line. They're like, "Wow,
47:28AI did the whole video. That is sick."
47:31But the actual outcome is not cool.
47:34versus a really boring process but a
47:37great outcome and people are like,
47:39"Dude, gross. Get with the times, man.
47:41You don't have automated thumbnails."
47:44This is what process bindless looks
47:45like. Please don't fall for it. Um,
47:47another good example. I absolutely love
47:49Nate, but I feel like his videos can be
47:52misleading. Um, I don't think he's doing
47:53it on purpose, but he is saying, "Look
47:57how great AI editing is, guys. I'm
47:59getting 200k vid views on this video."
48:03But here's the thing. He's getting those
48:05views because it's about it's in an AI
48:06niche.
48:08If you were to go offer this editing,
48:10like if you were to use this editing
48:12style and offer it to people, you would
48:14not get paid anything for it. So, it is
48:17the output is only cool because AI made
48:20it, but it's not good editing. And so,
48:23if you were to use this outside of the
48:24AI niche, it's going to flop for you.
48:27So, anyways, I really pray that this
48:29makes sense. Um, we want to care about
48:32the outcome, not the process. What is
48:35actually happening? Long story short,
48:38there is no point in automating in using
48:42AI inside of a system that doesn't get
48:45desired output. There's no point in
48:47scaling a system that doesn't work. And
48:49so, the highest leverage thing that you
48:51can do is by putting in maximum effort.
48:54So, doing things manually until you
48:57actually get the outcome you want and
48:59then using AI. So, I've designed over
49:02like 2,000 thumbnails in my day. It used
49:04to be my my full-time job. So, the only
49:07reason I know how to use AI in my
49:09thumbnails is because I know how to make
49:11good thumbnails without it. And so, that
49:12is the kind of lie I want to free you
49:15guys from. You have to get the output
49:16first. Um, you can use AI to help you
49:19learn, but you have to actually do the
49:21work. And so phase zero is basically
49:24this system doesn't work. Phase one is
49:26hey this system worked but it man it
49:28takes a lot of work from me. Um and then
49:32phase two which we'll talk about in a
49:34second is uh or phase three is actually
49:36decreasing your input. So that's step
49:38three. Once the system works that's when
49:41you add leverage. So you make the system
49:44work get the output you want. Write the
49:47email write the marketing thing. um make
49:49the video uh build the application
49:53whatever get the output and then you add
49:55in AI leverage
49:57um and so phase one system doesn't work
50:00phase two hey look it works but we have
50:02to do a lot of effort this is where you
50:05add in all of the leverage so now the
50:08system works and I don't have to do much
50:11here's a simple example I used to sell a
50:13high ticket program and I used to sell
50:16the program myself on the calls
50:18And so I learned how to get people to
50:22close into my program. So pay me money
50:24for my program. And then I turned what I
50:27learned into a script that I hired
50:30salespeople to run for me. So I used to
50:32have to take every call and I would make
50:34money. But then I would take a fraction
50:37of the calls and I would still make the
50:39same money because I added leverage of
50:41labor and I average added leverage of
50:44copies. meaning like my system for
50:46taking sales calls for my offer. Really,
50:48really cool. And so, yeah, guys, um I
50:52know that was dense. I hope that you're
50:53still watching this video, but hopefully
50:55this is helpful. I really pray that it
50:56is. Um but now we can actually go do
51:00this. We're going to spend the rest of
51:01this course. So, in part two execution,
51:03we're going to actually do this. Um the
51:06next course is going to be like adding
51:08more like AI leverage, like the next
51:10principle module, but the entire part
51:12two is going to be doing. Don't worry,
51:13we're going to be doing this. So, yeah,
51:16phase 4 is adding more leverage. We
51:18won't talk about this much, but
51:20basically, you just bring money in and
51:22you uh scale. So, in recent news, SpaceX
51:25just IPOed, and they're hitting phase 4
51:28like crazy. So now that they figured out
51:31how to launch rockets that actually get
51:33to space, they are saying, "Hey, give us
51:36a lot of money and we'll launch a bunch
51:38of rockets because we'll buy more people
51:40and we'll buy more copies and uh yeah,
51:42we'll make more rockets." So yeah, that
51:44is phase four scaling, which is really
51:45cool. This is why as a software
51:48developer, as a founder, once you hit
51:50that one, the VC gang piles in. So once
51:54you get a working system, everybody
51:56wants to invest in your stuff because
51:58they want to see that they want to see
51:59it scale. Um, yeah. So pretty cool. So
52:04how do we leverage AI leverage? I know
52:06this was a lot, but hopefully I if if
52:08this was helpful, please, you know,
52:10leave a comment or send me an email. So
52:12I'll make more of this if it is helpful.
52:13If this is just boring, I'll go back to
52:15doing shiny stuff, but just let me know.
52:18Now what? Well, we're going to get into
52:20building systems with AI. So, like we
52:22just talked about, the real leverage in
52:25adding AI into a system is phase three
52:28where you're adding in leverage after it
52:30already works. But the cool part about
52:32AI is that it can reason and so it can
52:35help you get to phase two. It can help
52:39you build the system, but it's a
52:40different way of thinking about it. And
52:42so, that's what we're going to talk
52:42about in the next module. So, I'll see
52:45you guys over there. Uh, let's get to
52:47it. All right. Welcome to the third
52:49principle module. How to build systems
52:51with AI. This is what I've been waiting
52:53to get to. We had to cover that other
52:55stuff first, but man, this is going to
52:57be epic
53:00because AI leverage is weird and it's
53:03hard to explain, which is why you've
53:05probably never heard it explained
53:07before. And I am not some philosopher
53:10genius, right? But I be talking to the
53:12clankers, trust me, bro. 8 to 10 hours a
53:15day for months, I'll be talking to the
53:17bots. And all I want to share is my
53:20perspective that I have found truly
53:22useful, meaning has created insane value
53:25in my life and business. So yeah, let's
53:27get into it. Why why is AI leverage so
53:30weird? Well, number one, it can reason
53:33and so it's it kind of feels like labor
53:36leverage a little bit. Like it can think
53:38a little bit, but it's still just math.
53:41And so some stuff it royally drops the
53:44ball on and you're like, "What in the
53:46world? How did you not get that right?
53:49Um, so it's like labor leverage, but
53:51it's technically copies leverage because
53:54someone built this model and you're like
53:56copying this like fake employee. And so
53:59basically, if you treat it like a human,
54:01you're overselling it. It doesn't have
54:03perspective and it it lacks the human
54:05touch. But if you treat it like just a
54:08computer, you're underelling it because
54:10it can reason. Um, and so a good example
54:14of why this thing is not, because a lot
54:16of you guys probably like, well, it's
54:17going to become, you know, sentient and
54:18all this. I don't believe that crap.
54:20Why? Because look at this. Look at this
54:22meme. Would AI have ever decided to put
54:26this meme right after what I just said?
54:28No. But I'm a human. I'm from Alabama. I
54:31uh my granddad dips. I have perspective.
54:35I have taste. Therefore, I wanted to hit
54:37hit you with this meme to prove that
54:39man, stuff is weird. And that's probably
54:41why you're watching this video. If
54:42you're still here, it's because I'm not
54:44going through a stupid AI slop slide
54:46deck. I'm giving you some like some raw
54:49crap. Um, but KJ1, it'll transcend. It
54:53will be creative. It'll have
54:54perspective. Ah, we're all screwed. I
54:56don't think so. Because every model is
54:59going to make stuff like that model. So,
55:01if we're all using Claude and we all ask
55:03it to do a task, doesn't matter how much
55:05context we give it, it's going to do the
55:07task like Opus 4.8, right? And as
55:10humans, we like new stuff.
55:14Imagine your favorite song ever. What if
55:17you had to listen to it every hour on
55:20the hour for the rest of your life? You
55:22would probably hate that song. And that
55:24is because we like new stuff. We like
55:28unexpected. An AI is expected, which is
55:31why it's very hard for it to create
55:34incredibly tasteful things. Now, can it
55:36do some stuff pretty well? Absolutely.
55:38But when it comes to really good
55:40experience things that humans are going
55:42to use like this video is very bad at.
55:45Um, and remember value is in comparison.
55:48AI slop will always be AI slop. So
55:51although this looks really good or it
55:53used to look good, it became what?
55:56Sloppish. But now this is the new AI
55:58slop. Even though it looks incredible
56:00comparatively, right? And soon this will
56:03be awful and there will be something
56:04new. It's just like why there's there's
56:06trends in fashion, right? It's why we
56:08look at people from the 60s and we're
56:10like, "Oh, dude, what were they
56:11wearing?" It's because it's comparison.
56:14Things change. And so, um, this is why
56:16models are always going to be kind of
56:18average because we all have access to
56:19the same model. Once again, values in
56:22comparison. If you had the iPhone 1 in
56:252007, you were a freaking legend. All
56:28right? Let me tell you. I was in middle
56:29school and people that had this were
56:30like instantly goatated. But if you have
56:34this phone in 236, right, you're going
56:35to be shamed. It's the same phone. And
56:38so the comparison, what's happening in
56:40society is more important than the
56:42actual like thing. So that's why AI is
56:44tough. AI will never replace actual
56:47human creativity. Right? Debate me on
56:50this. I believe it from the Bible.
56:51Right? So God created mankind his own
56:54image. The image of God, he created
56:55them. Male and female, he created them.
56:58We are created to create. Look at God's
57:00beautiful creation. um we get to create
57:03in his image and that's why I don't
57:05think AI will ever fully replace just
57:08that raw spark of life. I don't know. Um
57:11the good news is it will replace
57:14everything else so that we can do more
57:16creative stuff. This is the good news.
57:18People were like, "Oh, AI is going to
57:19take jobs." Bro, so did the calculator.
57:22Did you know they used to have entire
57:24buildings of people who would be like
57:26one calculator, right? They'd all crunch
57:29these numbers. Thank god no one has to
57:31do that anymore because of the
57:32calculator. AI will do the same thing.
57:34It will get rid of some really boring
57:36stuff so we can do more amazing free
57:38will stuff. All right, KJ. Well, what
57:41the freak do we use AI for then if it's
57:43not good at this stuff? Well, I'll
57:45explain to you. Um, the first is
57:47understanding explaining things to you.
57:50And so, as I was making this module in
57:52my office, my editor Caleb and best
57:54friend was editing something. He was
57:57building this insane model, but he was
57:59asking um Claude questions on the
58:03technical tools. So Claude is helping
58:05him learn, but he is still the creator.
58:08Um as I was typing this, that's what he
58:10was doing. Same with you building
58:12systems. Um two, researching AI can help
58:16you find the best tools, the best copies
58:20of the latest information tools that you
58:22can use to build your systems. And so an
58:25example is I was working with one of my
58:27one-on-one clients and we're going to
58:29build this this product for him so that
58:32it will save him $250 a month. And I
58:34asked AI to go look at each one of these
58:37features and tell me which tools should
58:39we use for this. Dude, normally this
58:42would have taken me like all day. It did
58:45this in five minutes. This is the
58:47leverage. It's so it's so good. Um all
58:50right, cool. Number three is reasoning.
58:53So AI can think and it's really good in
58:56poking holes in your plan. It it helps
58:59you explore every options. The only
59:01caveat here is do this after you've
59:04already thought of stuff. This is it,
59:06right? Um so it's like labor leverage of
59:08skill. So it's like having a thinking
59:10partner. It is so good for um and you'll
59:13see in the build template later, but
59:16imagine imagine hiring Claude as a
59:17thinking partner. He's still an
59:19employee. You're still the visionary.
59:21you show the CEO, but he can think with
59:23you and help you come up with the better
59:25outcomes, which is epic. And four, this
59:28is where things get pretty fun. This is
59:31using reasoning in your systems. So the
59:34first three were all about building the
59:36system. This is about it being inside
59:38the system. And so using basic reasoning
59:41at scale in your systems for your users
59:45to then use. So this is kind of like
59:48copies leverage um plus artificial skill
59:51leverage. It is awesome. Good example is
59:54the signal app. And so input user has a
59:58problem that they want to see videos on
1:00:00how to solve. They explain that to
1:00:02Claude or to signal which uses Claude.
1:00:05That's the input. Then I have a skill
1:00:07which I've programmed Claude to do and
1:00:10he thinks about what queries he should
1:00:13run based on the instructions. And so I
1:00:15won't give you guys all that because
1:00:16that is literally like the core of my
1:00:17app. That is the money. Um, and this is
1:00:20the output. So normally if I was like to
1:00:23hire my mom and say every time someone
1:00:25has a problem, take that problem and
1:00:28come up with 10 search terms, she could
1:00:31do it, right? She could totally do that.
1:00:33But she could service one user probably
1:00:35every 20 minutes. AI gives us basic
1:00:39reasoning at insane scale. I know this
1:00:42is a little bit tough to get, but this
1:00:44is the the crazy leverage. Um,
1:00:48uh, so one of my quotes, we'll go over
1:00:50this in a second, but the real leverage
1:00:51of AI is the una application or no, the
1:00:56the creative application of average
1:00:59reasoning. And this is kind of a good
1:01:00example of that. Use case number five,
1:01:04basic reasoning and commands. This is
1:01:05like a a virtual
1:01:08I'm such an idiot, guys. virtual VA, a
1:01:11virtual virtual assistant. Oh my gosh.
1:01:14Um, so basically doing literally
1:01:16anything that is closer to objective. So
1:01:18anything mathreated so you can spend
1:01:20more time doing the subjective. So once
1:01:23again, writing code, right? Not the
1:01:25vision or the deep problem solving, but
1:01:28just the executing, translating your
1:01:30words into code on the screen.
1:01:32Summarizing, so reading stuff,
1:01:34summarizing it. Uh, very average kind of
1:01:36creative stuff. So like writing an
1:01:38email, a YouTube title, it's good at
1:01:40that small task. Pattern and research
1:01:42analysis, running computer commands, all
1:01:45this good stuff. So an example is this
1:01:47is not loading, but this is one of the
1:01:49skills that I actually we'll probably
1:01:51build this plugin in this course at some
1:01:53point. AI does all the research for my
1:01:56videos that I so I I can come up with
1:01:58thumbnails and titles. It comes up with
1:01:59the keywords. It does all the research
1:02:01for uh the titles. It ranks the titles.
1:02:05It has literally given me the titles
1:02:06that have given me my last two viral
1:02:08videos. Epic. That's using all these
1:02:11things mentioned above to save me a lot
1:02:13of time. This is a project for one of
1:02:15our clients and it is a pricing estimate
1:02:19for he he builds custom houses. He
1:02:22inserts the the uh blueprints. We then
1:02:24go through and add in these custom like
1:02:27variables that he wanted in here and
1:02:28then it gives him the takeoff and
1:02:30estimate. This has to take them like a
1:02:32week manually. But with AI, it can be
1:02:35done in a matter of literally minutes,
1:02:37which is crazy. But honestly, anything
1:02:39that you can think of that you would
1:02:41hire an assistant or a junior developer
1:02:43to do, AI can do. It is um absolutely
1:02:46awesome. So, the real leverage with AI
1:02:51is the una application of average
1:02:53reasoning. But if we're honest, this is
1:02:56how it's always been, right? Think about
1:02:58McDonald's. they can hire high schoolers
1:03:01because the process of making a burger
1:03:03is not that hard. Someone made this
1:03:06system so simple and so easy, but they
1:03:09made it they created it in a creative
1:03:11way that gets a lot of value. Um, I know
1:03:14this is trippy because me and you have
1:03:16never had this amount of leverage
1:03:18before, but this is something that you
1:03:20really need to think about. And, uh, the
1:03:22first time I ever used Claude, these
1:03:23were my thoughts. You can pause it and
1:03:25read if you want, but uh, I thought that
1:03:27was pretty cool. And a lot of things
1:03:28have have come out to be true. But a
1:03:31couple things I want to talk about is
1:03:33people will feel so good about doing
1:03:35something that they will cope for doing
1:03:37the right things. So AI makes it so easy
1:03:40to make a website, to make some emails,
1:03:42to do whatever. It gets outputs that you
1:03:45otherwise would have never gotten. But
1:03:48the problem is they're not the right
1:03:49outputs. They don't actually work. You
1:03:51don't actually get clients. You don't
1:03:52actually get views. And that's the
1:03:54issue. Um, everybody will be able to
1:03:56code, but very few will make successful
1:03:58apps.
1:04:00Um, people will spend 95% of their time
1:04:02asking how to build something and 5%
1:04:04asking what they should build. This is
1:04:07bad. It should be the opposite, but I
1:04:08guess that's how it's always been. So,
1:04:10anyways, there's some more like takes in
1:04:12there if you want to check those out.
1:04:14Yeah, what it really did is AI unlocked
1:04:16scalable reasoning, but it did not
1:04:19unlock true creativity.
1:04:22And a lot of times when people approach
1:04:24me for help with AI stuff, they're not
1:04:26getting results because they think it is
1:04:28fully creative. It is not. But it can
1:04:31reason. And so if you give it the right
1:04:34guidelines, which you'll learn how to
1:04:35do, it can do things for you. But you
1:04:37have got to think and create for
1:04:40yourself. And so the alpha is in the
1:04:44creative application of the scalable
1:04:46reasoning. How can you creatively use a
1:04:50very, you know, 200 IQ person with not a
1:04:52whole lot of street smarts? How can you
1:04:54use that at scale um to make value? But
1:04:58to be honest, this is how it's always
1:04:59been, right? Steve Jobs, this was him
1:05:01when when he first made Apple. Whenever
1:05:03they fired him, they almost went
1:05:04bankrupt. They brought him back and he
1:05:06made everything go crazy. The reason I'm
1:05:08showing you guys this is because they
1:05:10had the same engineers. So, the same
1:05:13engineering talent, right? real humans,
1:05:15not just AI, but like real creative
1:05:17humans
1:05:19were not able to create value because
1:05:20they lacked the vision and the agency.
1:05:23When Jobs was gone, things fell apart.
1:05:25When he came back, things went
1:05:26exponential because it's all about how
1:05:29or it's all about what you choose to use
1:05:32your your reasoning on. That's what I'm
1:05:34trying to say. How you actually use the
1:05:36reasoning. So, anyways, here is my
1:05:38optimistic prediction. AI will take over
1:05:41everything black and white. Um, so we
1:05:43can work on stuff in full color. I know
1:05:45that sounds extremely cliche, but AI
1:05:48will take over doing all my research for
1:05:50these video thumbnails so that I can do
1:05:52what? Make a cool thumbnail. Like,
1:05:54that's the cool part about AI. And the
1:05:56problem that you run into with AI is
1:05:58you're trying to get it to do
1:06:00everything. And it can't. If you could
1:06:02tell AI, make me a million dollars, make
1:06:04no mistakes, you'd have a million
1:06:06dollars. But it cannot. You have to have
1:06:08that perspective and that agency.
1:06:11So KJ, how do we do this? Well, like how
1:06:14do we use AI like a non-gubber? Well, I
1:06:17remind you this is the goal, right?
1:06:19Figure out what to give AI so that you
1:06:22can get out what you want. So you have
1:06:24to figure out what you want, then figure
1:06:25out what to give it, figure out what
1:06:27instructions to tell it, and you'll get
1:06:29you get what you want. So to accomplish
1:06:31this, we need to do three things. Well,
1:06:33the first is context structuring. So how
1:06:36do we structure what we give Claude? The
1:06:38second is workspace setup. How do we
1:06:41make it easy for you to work with
1:06:43Claude? And three, how do we understand
1:06:46giving Claude different roles? So the
1:06:48brain of the the model, we can give him
1:06:52different kind of agent setups so he can
1:06:55have different roles and that can allow
1:06:57us to get bigger, juicier outputs.
1:07:00Number one, context structuring. What is
1:07:03context? Well, it's just files and
1:07:06folders. So it's files, the actual stuff
1:07:09you give it, and then it's folders, how
1:07:11you structure and group those files.
1:07:14That is context. I like to break context
1:07:17into two different types. This helps me
1:07:19a lot. So the first is context clues. We
1:07:22covered this in the first module.
1:07:24Basically, you're explaining the
1:07:26situation in more detail to Claude so
1:07:28that he knows what you mean. So if I
1:07:31say, "Build me a dope website," he
1:07:34doesn't really know what dope means. But
1:07:36if I say, "Build me a website for this
1:07:38thing. It needs to look like this.
1:07:39Here's the parameters. Here's examples."
1:07:42Now he knows what dope means to you. He
1:07:45can get a better outcome. And so this
1:07:47example is what we talked about earlier,
1:07:48right? Giving just the right context.
1:07:50The second is instructions. These are
1:07:54this is a specific direction on how
1:07:56Claude should do something the way you
1:07:58want it. This is also referred to as
1:08:00skills, which are now inside of plugins.
1:08:03But guys, it's all just text
1:08:05instructions. It's all just files.
1:08:07That's all that it is. And so, here was
1:08:10my old skill for my second channel that
1:08:12I used to post daily is I have a certain
1:08:15way I like Claude to do my YouTube
1:08:17titles. And so, if I tell Claude, just
1:08:19base Claude, hey, give me a YouTube
1:08:21title, it's going to give me the most
1:08:23average cookie cutter title it can think
1:08:25of. But if I say, "Give me a YouTube
1:08:27title," and I give it some instructions
1:08:30for how it should create them. Now, I'm
1:08:32going to get a result that's closer to
1:08:34what I want. And so, you're e either
1:08:36giving him more context or you're giving
1:08:38him instructions. Um, which are all just
1:08:41once again context, but I like to break
1:08:42it down. It makes things clearer. And
1:08:45so, here's an example from my finance
1:08:47plugin. If you guys have used this from
1:08:48my last video, I gave Claude
1:08:51instructions
1:08:52for how it should create um your finance
1:08:55dashboard when you run my skills. And so
1:08:58there are so many ways to create a
1:09:00finance dashboard, but I figured out the
1:09:02way that I think is very valuable. I
1:09:05gave Claude instructions and I gave that
1:09:07to you so that Claude can use his
1:09:09reasoning and my instructions to build
1:09:11you something epic. So yeah that is the
1:09:13kind of two different ways I break
1:09:15context into.
1:09:17Now
1:09:19context and folder structuring
1:09:21this is how you achieve the goal. So if
1:09:24we want to give Claude exactly what he
1:09:26needs no more no less we have to learn
1:09:29how to structure context and folders.
1:09:33And so this is why folder structuring is
1:09:35so hot right now. There's a meme. is
1:09:37because this is the key to unlocking
1:09:40efficiency and outputs with Claude. And
1:09:42so, a good example of this is I used to
1:09:45have an email system that I ran for one
1:09:47of my marketing clients when I still did
1:09:48marketing full-time. And I used to give
1:09:51Claude this 250 page PDF every time I
1:09:55went to write an email. And I kept
1:09:57wondering, why does he not get the
1:09:58testimonials right? He's not doing the
1:10:00testimonials good at all. And then I
1:10:02realized, whoa, I can break this 250
1:10:05pages into different folders, different
1:10:08segments. And I made a testimonial
1:10:11folder. And so whenever Claude needed a
1:10:13testimonial, he didn't have 250 pages.
1:10:16He had two pages. And he knew where to
1:10:18find them because the folder was labeled
1:10:20testimonials. That is a simple way to
1:10:22get better outputs by structuring. So
1:10:25imagine it like this. I want you to
1:10:27imagine that you hire an assistant and
1:10:29you gave them a thousandpage PDF with
1:10:31random words scrambled all across the
1:10:33pages. So not like there was any
1:10:35coherent like thought to this PDF. It's
1:10:37just a bunch of random segments of
1:10:39random words with random like meaning.
1:10:42That's level zero. Level one. Now I want
1:10:45you to imagine you separate the PDF into
1:10:48um different pages and files that are
1:10:51titled what they contain. So, it's
1:10:53titled like, "Hey, here's a research
1:10:55paper on this. Hey, here's some past
1:10:57examples of what I did here, right? It's
1:11:00still one file, but it's at least, you
1:11:02know, separated a bit." Now, I want you
1:11:05to imagine that you separate those files
1:11:08into folders. And so, you group these
1:11:10things in folders so that now if Claude
1:11:13or your assistant needs something, he
1:11:16doesn't have to look through a massive
1:11:17stack of papers. He goes and looks at
1:11:19the folder. Right now, level three. I
1:11:23want you to imagine that you have
1:11:25organized the text into files. You've
1:11:27organized the files into folders. You've
1:11:29organized folders into different filing
1:11:31cabinets. They're all separated. They're
1:11:33all labeled in ways that are focused on
1:11:36the output you want. And then you give
1:11:38your assistant specific instructions on
1:11:40where everything is. This is how you
1:11:43save tokens. This is how you get good
1:11:45outcomes. This is what is happening with
1:11:47Claude behind the scenes. Now, as these
1:11:50AI models get better and better,
1:11:53you have to do less and less of this.
1:11:56But the skill of knowing how to
1:11:58structure context is one that will serve
1:12:00you, I think, for the rest of time with
1:12:02AI. So, you should get as good as
1:12:04possible at making it as easy as
1:12:06possible for AI to do what it wants uh
1:12:08what you want it to do. So, this is how
1:12:11you become powerful with AI. And I'm
1:12:13going to give you my templates for this.
1:12:14Don't worry. I'm going to give you like
1:12:15my filing cabinets that I've spent
1:12:16months creating, but you have to know
1:12:18why I'm doing this and and why you
1:12:20should learn it. So, do not let it fool
1:12:23you. Um, this is a skill. I'm not going
1:12:25to lie, this is really, really tough.
1:12:26And there are levels to it. I'm not
1:12:29anywhere close to the peak, but if you
1:12:31just understand what I just showed you,
1:12:33you will be in the top 1%, maybe zero
1:12:341%. Because most people are just giving
1:12:37their entire life to Claude in some
1:12:39crazy like carpety wiki and saying, you
1:12:41know, fix my life. It's just it's so
1:12:44Claude is suffering. He's so confused.
1:12:47So yeah, you're about to skip straight
1:12:48to level three just by watching this
1:12:50course, but there are so many levels to
1:12:53this at doing this. Well, right. Um so,
1:12:57uh I don't even know what I was talking
1:12:58about this. Yeah. Uh I'm going to give
1:13:00you my dial context structure setup in
1:13:02this course. You're going to skip to
1:13:04level three. You're also going to get my
1:13:05quad, which will basically set up this
1:13:08entire setup for you. It's going to be
1:13:09epic. That'll be in the next part. But
1:13:12first, Claude MD. What is a Claude MD?
1:13:15What does it do? How do you make a good
1:13:17one? Let's let's talk about it. So,
1:13:19Claude MD is just a markdown file with
1:13:23the name Claude in all caps. It is just
1:13:25a text file. What makes it special is
1:13:28that when you title a file [snorts]
1:13:30Claude in all caps, anytime you open
1:13:34Claude in that folder, it will reference
1:13:36this file first before it does anything
1:13:38else. And so this is useful because now
1:13:42we can spawn in different cloud
1:13:44instances and they sort of have this
1:13:46like memory and instructions for working
1:13:49in that project. So you don't have to
1:13:51prompt it every single time. So really
1:13:54what CloudMD is is it's it's a it's a
1:13:57starting prompt that you don't have to
1:13:59type out. Basically it's a saved prompt
1:14:01that's going to load every time you open
1:14:03that project. Um, and so we we want to
1:14:06put in here important context clues and
1:14:08instructions for working in that
1:14:10project. This will make sense in a
1:14:12second. Now, the best analogy that I've
1:14:14came up with this is a movie called 51st
1:14:17Dates. Now, in this movie, this girl
1:14:19gets in a car wreck and every day she
1:14:22cannot remember what happened any days
1:14:24before. So, every day she spawns in, she
1:14:27has no memory of what happened
1:14:28yesterday. She remembers up to like her
1:14:3018th year of life and then the rest,
1:14:32every day is fresh.
1:14:34And so in the movie, the guy who ends up
1:14:36marrying her, what he does is he creates
1:14:39a little tape which fills her in every
1:14:42morning about who she is, what happened
1:14:44to her, their marriage, um their life
1:14:48together, and then she's like, "Wow,
1:14:50this is who I am." She walks upstairs on
1:14:52the boat and she joins her family and
1:14:54she's like, "Wow, this is, you know,
1:14:56this is my life." And so before he made
1:14:58this tape, she didn't know who he was.
1:15:01She's like, "Who is this stranger?" and
1:15:02it was awful. But he made for her a Claw
1:15:06MD so that when she spawns in, she knows
1:15:08what she's doing here, what her purpose
1:15:10is, and she can go about doing the rest
1:15:12of her life. So yeah, that is the Claude
1:15:15MD. That is the tape. Um, so what do you
1:15:19put in Claude MD? Here's the mistakes I
1:15:21see all the time. Dumping a load of junk
1:15:24in there. Okay, treating it like a
1:15:26long-term memory. Cloud MD is not a
1:15:28long-term memory. It is just for some
1:15:31brief macro context
1:15:34um about what it needs to get the
1:15:37output. Right? So Claude does not need
1:15:39to know how you like your coffee, how
1:15:40many kids you have or every team member
1:15:42on your payroll and all this other junk.
1:15:45Okay, for every prompt, it just needs to
1:15:47know where does it find stuff when it
1:15:49needs something. What are the
1:15:51preferences of working with you? So like
1:15:53what tools do you use? Um your workspace
1:15:55preferences, how does it how is it
1:15:57supposed to name stuff, etc. and then
1:15:59concise macro context on the big goal
1:16:02your system is a part of and who it's
1:16:04for. So for example um the GitHub for my
1:16:07application that me and my developers
1:16:09work on together. The cloud MD for the
1:16:11GitHub explains what type of code that
1:16:14we use our like structure, who does
1:16:17what, what the different features are.
1:16:19It's it's literally like one page long
1:16:21for an entire application because we
1:16:23wanted to keep it concise. Um because we
1:16:25don't want to tell Claude how to do its
1:16:28day. We just want to orient it about
1:16:30where it's at and what's going on and
1:16:32then let him work with us every single
1:16:34day and what we need. So, um I'm going
1:16:37to give you guys my own claude. Like,
1:16:38don't worry, you're going to get mine.
1:16:40But I want you to know why I've made
1:16:42this one the way it is. And you should
1:16:44prune your Claude constantly. You should
1:16:46make this as short as humanly possible.
1:16:49Um most people get Claude to write its
1:16:51own claude. That is dope. Okay, you're
1:16:53going to do that in my course, but you
1:16:55have got to prune it. He will put in
1:16:57there stuff that you do not want in
1:16:59every single instance of Claude that
1:17:00you're going to do from now on. Okay, so
1:17:03just think about it. If you're not going
1:17:05to use it, if you're not going to use a
1:17:06certain line or instruction for every
1:17:08single prompt, then it should not be in
1:17:10your CloudMD. Okay? Just put it in a
1:17:13separate file and refer to it whenever
1:17:14you need it. So, here's an example of um
1:17:17just a a CloudMD. This is a uh YouTube
1:17:22course builder like example thing. So
1:17:24it's like key context of like what we
1:17:26use where where everything is at what
1:17:28the folders are and then the working
1:17:30rules. So here are some logs, here's
1:17:32some separate chats, whatever. So yeah,
1:17:34that's just a basic cloud MD. So
1:17:38context folder structuring. Here's a
1:17:40little like sneak preview of my proven
1:17:43structure. We'll cover this in depth in
1:17:44part two, but let's go over some context
1:17:47structuring. So once again, we want to
1:17:49give the exact inputs to get the exact
1:17:50outputs. That is our goal. To accomplish
1:17:52this, we need to do three things. Well,
1:17:55once again, we just covered context
1:17:56structuring. Now, we're going to cover
1:17:59workspace setup, right?
1:18:02Oh, here we go. Let's start by reminding
1:18:04you of your job with AI.
1:18:07Setting initial trajectory. So,
1:18:09clarifying where you want to go is one
1:18:11of the most important things with AI. We
1:18:13are lazy. We don't want to do this, but
1:18:15this will give you the good results.
1:18:17Then number two, realigning Claude to
1:18:20that trajectory. So making sure that he
1:18:22stays on track, pruning all the slop,
1:18:25keeping him in the right direction. That
1:18:27is your job.
1:18:29How well we have to be able to work with
1:18:32Claude. How we do these two things right
1:18:35here is by improving the context clues.
1:18:38So that part of context giving better
1:18:40instructions in the form of context and
1:18:43improving the structure so that claude
1:18:45can find these things easier and easier
1:18:47and this is extremely hard to do if you
1:18:50cannot see what's going on in your
1:18:52project. And this is why like claude
1:18:55coowwork was made for beginners. I want
1:18:57you to understand this. They have hidden
1:19:00everything from you.
1:19:02They care a lot less about helping you
1:19:04really. Um, and they care more about you
1:19:06being wowed by this technology so that
1:19:08you pay for it, which is, you know, it's
1:19:10fine, but if you've made it this far in
1:19:14this course, you you do not want to be
1:19:15average AI, right? You want to be above
1:19:17average. And so, if we want the insane
1:19:19edge that AI gives us, we need to go
1:19:22deep. We need to not just hide
1:19:24everything so that it's simple. We want
1:19:26to understand how the junk works. We
1:19:28need to see, we need to understand, we
1:19:30need to apply the human perspective to
1:19:33our projects. So, what does this mean?
1:19:37Essentially,
1:19:38we need a better file viewer. That's
1:19:40literally it. So, if we were able to see
1:19:42what Cloud was doing better, we would
1:19:44work with it a lot easier. And so right
1:19:46now, if you're not using a file viewer,
1:19:48how you would view Claude's files is in
1:19:51the Cloud app, which is pretty
1:19:52inefficient right now, having a folder
1:19:54open and then opening the text file in
1:19:57this ugly, hard to read form. Okay? Or
1:20:01you can do the exact same thing inside
1:20:04of a file viewer like Obsidian or
1:20:06anti-gravity or cursor or VS Code where
1:20:09you can have the exact stuff, right? You
1:20:11can have claude, you can have the file,
1:20:13you have the folder structure in one
1:20:15window and it looks incredible. That's
1:20:18all a good workspace setup is. So this
1:20:20is why we work in an IDE which is means
1:20:22integrated development environment. Um
1:20:24it's really just like it's kind of like
1:20:26a browser for your files like using
1:20:28Google Chrome but for viewing your files
1:20:30and working with cloud. Why are IDE
1:20:33valuable? Well, they have a great user
1:20:35interface. That's the whole point,
1:20:37right? To make this easier for you to
1:20:39work with cloud. And they have custom
1:20:41plugins to make it fun. They integrate
1:20:43with AI models really well. And you have
1:20:46the ability to read, write, and edit
1:20:48files and folders easily. That is the
1:20:50value prop here to give better context
1:20:53to CL. So in this course, we're going to
1:20:56be using Obsidian. This is my personal
1:20:58favorite file viewer. That is because it
1:21:01is just the easiest in my opinion to
1:21:03edit files. It's got a lot of good core
1:21:05plugins. It's got some great community
1:21:06plugins. You can sync this with team
1:21:08members so that they can actually sync
1:21:10with your projects. And it's built
1:21:12around information, not code. And so I'm
1:21:15not doing a bunch of coding. I mean, I
1:21:17do write code like Claude does it for
1:21:19me. I'm prompting Claude. And I find
1:21:21that this is easier than VS Code or
1:21:24something like that because it's able to
1:21:26I'm able to give it better context
1:21:28really. But you can use whatever you'd
1:21:30like. It's all the same thing, right? VS
1:21:32Code, anti-gravity cursor. It's all
1:21:34about you being able to see what Claude
1:21:35is doing and help him do it better.
1:21:37That's it. So, in part two of this
1:21:39course, the execution chapter, we're
1:21:42going to cover my exact setup. I'm going
1:21:43to give you guys literally everything of
1:21:45500 plus hours of building, like how I
1:21:47set it up, how I build, but I wanted to
1:21:49explain kind of why we do this, why I'm
1:21:52giving this to you, and how I made it.
1:21:54So, these are the principles that I used
1:21:56to make my workspace. This is regardless
1:21:58of whatever application you choose to do
1:22:01with. Okay. Number one, stay focused on
1:22:04output. No one cares how cool your setup
1:22:06is. No one cares about your Notion
1:22:08database, and they especially don't care
1:22:11until you've gotten insane outputs.
1:22:13Think about it. Elon's a trillionaire.
1:22:15And when is the first time you ask, "hm,
1:22:17I wonder what his notion setup looks
1:22:19like, or I wonder what his Obsidian
1:22:21setup looks like." Probably not, because
1:22:23people don't care about the process.
1:22:25They care about the value you put out.
1:22:27You care about his cars, right? His
1:22:29rockets. And so, um, stay focused on
1:22:31output. Now, I still want you guys to
1:22:34try stuff out, though. Like, have fun.
1:22:36But resist bells and whistles for the
1:22:38sake of bells and whistles. Make sure
1:22:40it's valuable. If something isn't adding
1:22:42value, meaning isn't helping you create
1:22:45better stuff, delete it. Because the
1:22:47goal is clean thinking and clean inputs
1:22:49to Claude, not a bunch of random plugins
1:22:51and crazy themes, right? Keep it clean.
1:22:54Um, so yeah, that was workspace setup.
1:22:57Now, there's one more thing to cover as
1:22:59far as principles, and this will help
1:23:00you fully understand everything I'm
1:23:02going to give you in the template
1:23:03section, and that is understanding
1:23:05different cla roles.
1:23:08What is an AI agent? We talked about
1:23:09this in the co-workass 2.0 inside of the
1:23:13free school. It is free if you want to
1:23:14check it out. If you're basic, you
1:23:16should probably watch that, but you're
1:23:17already deep, so let's keep going. Um,
1:23:19an AI agent is basically this. It's the
1:23:21brain model. So, the actual code that
1:23:23reasons. It is context. And so how you
1:23:27prompt that code, what context you give
1:23:29him and what instructions you give him.
1:23:32And then it is tools. So what does he
1:23:34has access to do? So he can run
1:23:36commands, he can access different
1:23:38connectors, API calls, and then all this
1:23:41happens inside of a loop. And so if you
1:23:44set a goal and you have the agent
1:23:47continue working until it actually gets
1:23:49that goal, that is kind of what makes it
1:23:51an agent. It's what makes it problem
1:23:54solve and think for itself without you
1:23:56needing to be in the loop. Spoiler
1:23:57alert, you still need to be in the loop,
1:23:59right? Like that is the whole point. But
1:24:01this is what makes an agent. And so, um,
1:24:04like I said, most people see AI agents,
1:24:08but I'll be honest, most of it is hype
1:24:10garbage. Like, these people do not get
1:24:13many valuable outputs. They get a lot of
1:24:15clicks, a lot of attention on how cool
1:24:18their process is, but they're not
1:24:20actually making incredibly valuable
1:24:22stuff. Remember, boring process,
1:24:24incredible outputs is what we want. We
1:24:26do not want incredible process, boring
1:24:28output. We want to be providing value.
1:24:30And so, beware of the people who are
1:24:32building an agent army and paperclip or
1:24:35whatever. No, that's uh that's not not
1:24:37what we mean by this. But why is having
1:24:41these different roles or agents
1:24:43valuable? Like what? Why would you even
1:24:45do this? Well, remember the goal is to
1:24:48get the exact inputs we need to Claude
1:24:50to get the exact outputs we want. And so
1:24:53if you have different outputs you want
1:24:55in different parts of a system, you want
1:24:57to give Claude a different role. So
1:25:00let's say we're trying to heal people,
1:25:01right? Well, we would want to have a
1:25:03doctor Claude for that. So a doctor
1:25:06suited model like a a good brain doctor
1:25:09context and instructions with a claude
1:25:11giving it a stethoscope and surgery
1:25:13tools and then the doctor goal of what
1:25:15its goal is. Let's say we want to uh
1:25:18ride bulls. Well then we would give that
1:25:20claude a cowboy suited model. Probably a
1:25:24uh I'm not even going to make that joke.
1:25:26Probably a brain that's more for
1:25:28trucking through stuff than
1:25:29intelligence. um giving it context for
1:25:32cowboy stuff, cowboy tools, giving them
1:25:34a lasso. Anyways, you get the point. By
1:25:37having separate cloud instances work on
1:25:39different specific systems in your
1:25:41build, you're able to get better outputs
1:25:43more efficiently. So, instead of having
1:25:46all of this inside of one cloud and him
1:25:48getting confused, we separate it into
1:25:50different roles, different instances,
1:25:52different chats, and we're able to get
1:25:54better outputs. And once again, this
1:25:58tech is only getting better and better.
1:25:59And so you do not have to separate
1:26:01everything you do into a new chat.
1:26:03Honestly, you don't anymore. But I want
1:26:05to show you how to max out this skill so
1:26:08that when you get in a situation where
1:26:09you should be separating things, you'll
1:26:11know what to do. Okay, so these are the
1:26:14different roles that I use. I do not use
1:26:17all the hype the hypy tools. Mine is
1:26:19actually really really simple and I'm
1:26:21going to show you guys how to use it in
1:26:22part two. But these are my Claude rolls.
1:26:25So the first role is builder claude and
1:26:27he is using the best model I can get. He
1:26:30needs macro context of the full build
1:26:32which is usually inside of the cloud MD
1:26:34and I run this agent inside of the
1:26:36claude app typically co-work or code
1:26:39doesn't really matter whichever one your
1:26:40your preference is. Um I start a fresh
1:26:43chat every time I hit around 50% context
1:26:46window. And so I start a new clot and I
1:26:48give him the exact same context every
1:26:50time. Um, and you'll see in the
1:26:53execution phase how I do this using my
1:26:55daily logs. That's how I give him kind
1:26:56of some memory. Um, and if I have to
1:26:59start a new chat, like a a deep one, in
1:27:01the middle of a project, I will have him
1:27:04write a handoff prompt, which is very
1:27:06simple. You just basically say, "Hey,
1:27:07write yourself a prompt for the next
1:27:09cloud. Open a new cloud. Give him the
1:27:10prompt." And the goal here is to keep
1:27:13the context clean. This is the main
1:27:16cloud. This is the one where we
1:27:17literally want to give him exactly what
1:27:19we need, nothing more. Um, and he only
1:27:22needs context for the problem you're
1:27:24solving. So, he doesn't have to know
1:27:26exactly every single detail of every
1:27:29feature of your app. If I'm just
1:27:31building the notes section of my app, I
1:27:34only want this Claude to really
1:27:36understand kind of what the macro app is
1:27:39for and how the notes feature fits into
1:27:41that so that he can build the notes
1:27:42feature. I don't want to go in depth on
1:27:44everything else. And so that is a skill
1:27:46but you'll you'll learn it. Okay. Next
1:27:48is the research and VA clot. So this is
1:27:52using a cheaper model typically using
1:27:53like sonnet but he doesn't really need
1:27:56any context besides the initial prompt.
1:27:59And this is just a separate chat that I
1:28:01use for researching stuff explaining
1:28:03things to me helping me understand
1:28:06creating context for the main agent to
1:28:08use. And so let's say I want to use a
1:28:11bunch of examples for my main builder
1:28:13claude in like an email system. I will
1:28:16have this claude read 250 short form
1:28:19scripts and then distill those into a
1:28:22two-page list of email ideas and then
1:28:25I'll give that list to the main claude.
1:28:28So instead of that main claude doing all
1:28:30this work in the middle of our project,
1:28:32I'll do it in a separate cloud and just
1:28:34give him the output. Right?
1:28:36I typically run this claude in Obsidian,
1:28:38which I'll show you guys how to use in
1:28:39the Claudian plugin, or if I'm working
1:28:42in VS Code, I will use the Cloud Code
1:28:44plugin in VS Code. The main value here
1:28:46is keeping the main agents context
1:28:49clean. So, this the goal of this
1:28:51research cla to basically do low do low
1:28:55tiered tasks outside of the main agent
1:28:57so that I keep this one clean as
1:28:59possible. That's it. It's all I'm doing.
1:29:01Um, I don't give him any extra context.
1:29:02It's literally just a claude chat. Like
1:29:04I just open Claude app, do the chat or
1:29:07open Claudian, start a new chat and just
1:29:09ask him questions. And then last one is
1:29:12crash dummy claude. So I build a lot of
1:29:15stuff for you guys and I want to test it
1:29:18as if you were using it. And so I cannot
1:29:20test it using the claw that I'm using to
1:29:22build it. And so I will start a new
1:29:24cloud instance in usually co-work
1:29:27because most you guys use co-work um and
1:29:29I will test it as if I'm you guys. I
1:29:31will have no fresh context. It'll just
1:29:33be the plugin that I've given it or the
1:29:36instructions I'm giving it. And I'll try
1:29:38to use a cheaper model because I know I
1:29:39know a lot of you guys will be trying to
1:29:41do a cheaper model. Also, I use this to
1:29:44test stuff for my application. And so in
1:29:47my application, I'm not using my cloud
1:29:50app. It is API calling typically a
1:29:52cheaper model than cloud. And so I want
1:29:55to use whatever model I'm calling in the
1:29:57API to test it in a fresh chat to see if
1:30:00I get the outputs that I want. Um, and
1:30:02so yeah, the value here is you're able
1:30:04to test your automations in real
1:30:06scenarios that your users will be using
1:30:08it in. Then you can bring back the
1:30:10feedback and give it to your builder
1:30:11claude. This is how I made the co-work
1:30:14starter pack. This is how I made the
1:30:15finance dashboard. This is how I made
1:30:16the second brain was using a crash dummy
1:30:19claude to test stuff and then give the
1:30:21results back to my builder. So that's
1:30:23it. Really simple. And once again, bro,
1:30:26you ain't that G yet. None of us are. If
1:30:29you can't make one clad make you money,
1:30:32you can't make five clouds make you
1:30:33money. Keep it simple. Once again,
1:30:36output over process. Do not do this just
1:30:38to feel cool, right? Start with one clad
1:30:41and work your way up until you really
1:30:43need all these clots. It's crazy. So,
1:30:46yeah, that is all of the three goals.
1:30:48You now understand the principles which
1:30:50underly the entire actionable part of
1:30:52this course. I know this was kind of a a
1:30:54a long one, but I want to congratulate
1:30:56you. Like, you probably understand AI
1:30:58better than 01% of people when it comes
1:31:01to actually creating valuable things.
1:31:03So, uh yeah, congrats. We are done with
1:31:05principles. It is time to have some fun
1:31:08and go execute. So, yeah. Um just know
1:31:11that you need both. Okay? I've been
1:31:13harping on tactics a lot here, but
1:31:15really you have to have both. Principles
1:31:18alone just leaves you in your head. You
1:31:20have to go do stuff. But if you don't do
1:31:22stuff with the understanding of
1:31:23principles, you're going to be confused
1:31:24when you don't get the right results.
1:31:26But if you have both, man, you get
1:31:28results. So, it's time to go put these
1:31:30principles in actions. I'm going to
1:31:31share with you the latest and greatest
1:31:33tactics that I found from uh helping 10
1:31:36plus business owners automate real
1:31:37businesses. 500 plus hours, lots of
1:31:40helpful automations. Let's freaking get
1:31:42it. All right. Welcome to part two of
1:31:44this course, execution. It is time to
1:31:47start building, my friends. I'm going to
1:31:49give you guys my entire building process
1:31:51and the template, how I worked with
1:31:52Claude, all the context structuring,
1:31:54literally everything. It's taken me five
1:31:57months of building and 500 hours of
1:31:59working with AI to actually perfect this
1:32:02setup and workflow. And I want you guys
1:32:04to judge it by the outputs, not how cool
1:32:06it looks. Like, what did it produce? And
1:32:08I managed to accomplish honestly some
1:32:10stuff that I've only ever dreamed of,
1:32:12including my first ever custom paid
1:32:14software with lots of paying users.
1:32:16Shout out to who bought the yearly
1:32:17package this morning. and you're an
1:32:18absolute beast. Yeah, you can check out
1:32:20the app if you want to in the
1:32:21description to see just how legit it is.
1:32:23And then also tons of custom plugins,
1:32:25automations, client work. If you want us
1:32:27to build custom software for you and you
1:32:29don't want to do this yourself, then hit
1:32:31us up uh the agency in in the
1:32:32description. But yeah, custom thumbnail
1:32:35software, dashboards, client dashboards
1:32:38for my one-on-one clients, all kinds of
1:32:40amazing creations all through this setup
1:32:43that I really think is incredible. And
1:32:45let's get straight into it. But I'm
1:32:46going to give it to you guys for free.
1:32:48No strings attached. There is one string
1:32:50attached though. If you're someone who
1:32:52skipped the principal section, I want to
1:32:54warn you. You will not get the most out
1:32:56of this until you watch principles. I
1:32:58know that it's kind of boring. It's not
1:32:59shiny, but I cannot explain enough. This
1:33:02stuff did not just come from having a
1:33:04cool template and a good claim MD. It
1:33:06came from understanding principles. So,
1:33:09I cannot recommend enough you go back
1:33:11and watch it if you're someone who
1:33:12skipped it. But with that being said, if
1:33:14you watch principles, man, you are
1:33:16ready. Let's get right into the sauce.
1:33:18So, what is included in this template?
1:33:20What am I actually going to give you
1:33:22today? Well, the first thing is my
1:33:25Obsidian/IDE
1:33:26setup. So, how do I view what I'm
1:33:29working on with Claude? And this allows
1:33:32me to prompt him well to to see what's
1:33:34going on, to see why, what's causing
1:33:36problems, and actually fix the context
1:33:38so that he can get better outputs. The
1:33:41second is how I set up my cloud agents,
1:33:44how I work with cloud, like what does my
1:33:46workflow with an agent look like? Third,
1:33:49probably most importantly, is my folder
1:33:52and context structure. So, how do I
1:33:54structure these projects cleanly to
1:33:57build these robust automations and have
1:33:59Claude work on it without having to kind
1:34:01of start fresh every single time, which
1:34:03leads to some ballin context
1:34:05engineering, which is the whole point of
1:34:07this course. And then last but not
1:34:09least, the workflow. We're going to go
1:34:10over three levels of how I go from start
1:34:13to finish with automations depending
1:34:15whether it's beginner, intermediate, or
1:34:17advanced. I'm going to show you guys all
1:34:18of it. So, even if you're a brand new
1:34:19person, this should be able to help you
1:34:21a ton. And all this with a goal of not
1:34:24to look cool, not to get internet
1:34:26points, but once again to build ballin
1:34:28systems with ballin outputs. This stuff
1:34:31is built for value. I built it for
1:34:33myself, not to share. I built it to
1:34:34build things. And so, I hope you guys uh
1:34:37get a lot of value from it. And also a
1:34:40lot of stuff I'm going to cover right
1:34:41now in this module it will do
1:34:42automatically. But I have to explain how
1:34:44it works so that you can personalize it.
1:34:47You can make this template your own and
1:34:49really get maximum value from it and you
1:34:52can understand like we talked about in
1:34:55the principal module. So first up my
1:34:57Obsidian/IDE
1:34:59setup. Like I mentioned before, I use
1:35:01Obsidian. I absolutely love it. I've
1:35:04tried like 10 IDEs, 10 different
1:35:06softwares, and I always come back to
1:35:08Obsidian just because it is the cleanest
1:35:10and easiest in my opinion. Here is how I
1:35:12set up my Obsidian. So, if you just
1:35:14download Obsidian fresh, this is what
1:35:16these are the settings that I change,
1:35:18but they're always they're going to be
1:35:19personalized in my template already. So,
1:35:21you don't have to change them if you use
1:35:22my template. They're already there. So,
1:35:24let's go ahead and take a tour of my
1:35:26Obsidian and kind of see what it looks
1:35:27like. All right, so here's the Obsidian
1:35:29setup. I'll run you guys through it real
1:35:30quick. We are going to do live builds
1:35:32with this. So you don't have to ma like
1:35:34master this right now, but I'll give you
1:35:36guys a tour. So the only plugins that I
1:35:39really have enabled are some of the base
1:35:41plugins. So bookmarks and search. That's
1:35:44literally it. And then I have Claudian,
1:35:46which is enable you to have Claude on
1:35:49the side. And then also Terminal. So I
1:35:51can spawn into terminal and I can spawn
1:35:54Claude code if I want to work with
1:35:55Claude Code on the right hand side as
1:35:57well. And so if you if you're new to
1:35:59Obsidian, it really is just a file
1:36:02viewer and kind of a note-taking system,
1:36:04which makes it perfect for what we're
1:36:06doing. And if you're brand new, I have
1:36:07tons of other courses about this. It's
1:36:09not going to be a beginner course. Um,
1:36:11but it's very simple. You can close the
1:36:13tabs here. I can close this. And so on
1:36:16my lefth hand side, we have folders and
1:36:18we have the different files. And so I
1:36:20can click in the files. I can edit this
1:36:22stuff. And the way that I actually have
1:36:24this laid out is it has a translucent
1:36:27window on the appearance, which is what
1:36:28makes it look a little cool, but you can
1:36:30turn that off if you want and it'll look
1:36:32just same as normal Obsidian. But I've
1:36:34built this to be very, very simple. I
1:36:37just want to prompt Claude. And um we'll
1:36:39go over the Claude MD here in a little
1:36:41bit, but this Claude MD actually sets up
1:36:44the entire project for you. You don't
1:36:46have to really touch any of these
1:36:47things, but this is my Obsidian setup.
1:36:49The only setting that I will change is
1:36:51if I'm going to be sharing this with my
1:36:53team. Um, also I will check on
1:36:55automatically check for updates. That
1:36:57usually helps. But if I'm going to be
1:36:59sharing this with my team, I will also
1:37:00turn on Obsidian sync and then I will
1:37:03sync this to my Obsidian account and it
1:37:06can share it with my team if I want to
1:37:08if that's a project that I'm going to
1:37:09share. So yeah, let's jump back into the
1:37:11mirrorboard. So yeah, that is the
1:37:12Obsidian setup. And the way that you
1:37:14want to personalize this template to
1:37:16yourself, which I'll show you guys how
1:37:17to download the template here in a
1:37:18little bit. All you have to do is open
1:37:20up the template in Obsidian, make
1:37:23changes, and then whenever you go to to
1:37:25do a new project, you just want to copy
1:37:27the folder of the template and then just
1:37:29like basically you have this folder that
1:37:31is your template. You can change that
1:37:32Obsidian, and then whenever you want a
1:37:34new project, you just copy it and it has
1:37:36all your settings. Um, and if you want
1:37:37to actually see the settings, you just
1:37:39do command shift period in the folder to
1:37:42open up the dot files. So, hidden files,
1:37:45and you can see all your settings in the
1:37:47Obsidian folder. Yeah, you guys can look
1:37:49that up. Second part of what makes this
1:37:51template so special is how I work with
1:37:54Claude. So, what are the different
1:37:55agents that I use and what are their
1:37:57roles? First, Claude is just in the app.
1:38:00So, I'm using co-work.
1:38:02and I will basically open up the folder
1:38:04that I'm working on inside of co-worker
1:38:07inside of code. So that's really it and
1:38:09this is the main agent that I use. I uh
1:38:12use the latest and greatest brain
1:38:13usually because I have a good
1:38:14subscription and then I give it macro
1:38:16context usually inside of the cloud. I
1:38:19run this inside the app like I said and
1:38:21then I will start a fresh chat every
1:38:24completed like subsystem or system or if
1:38:27I hit around 50 to 75% of the context
1:38:30window limit and um if I need to kind of
1:38:33continue working on the same subsystem
1:38:35uh across different clouds I will have
1:38:37him check out the daily logs which I'll
1:38:40show you guys how to use in the
1:38:41execution phase and I'll have him write
1:38:43himself a handoff. So, I'll say, "Hey,
1:38:45Claude, um, we're running out of context
1:38:47in this window. I want you to kind of
1:38:48write yourself a handoff for picking up
1:38:51with the next Claude." Um, and it's
1:38:53super super easy. And the entire goal
1:38:55here, like I said in the principles, is
1:38:58to keep this main claude as clean as
1:39:01possible. I only want the context needed
1:39:03for solving the specific problem we're
1:39:05tackling. So, if I'm working my
1:39:07application and I'm working on the note
1:39:10takingaking feature inside of my app, if
1:39:14I'm working with a brand new cloud agent
1:39:16on that, I only want him to have context
1:39:18around the note-taking feature and
1:39:20whatever he needs to make that great.
1:39:22So, sometimes that includes a little bit
1:39:24of macro context about the whole app,
1:39:26but most of the times he doesn't need to
1:39:28know that. But let's say the notes play
1:39:30into a feature that we have with the
1:39:32actual video player inside of Signal.
1:39:35Well, he needs to know that context. And
1:39:37so, it's all about giving it just enough
1:39:39that they need to solve that problem and
1:39:41nothing more. The second way I work with
1:39:44Claude is with like a VA/ressearch
1:39:48agent. So, what I do is inside of uh
1:39:51Obsidian like you just saw, I will spawn
1:39:53in the Claudium plugin or I will spawn
1:39:55in um the terminal plugin and I will
1:39:57open claude code in terminal. Now, this
1:40:00Claude I just use for doing research. I
1:40:02use it to explore different tools. So,
1:40:04let's say I'm going to connect something
1:40:06through an API. I will use it to search
1:40:08the web and kind of summarize what
1:40:10different tools are out there. I will
1:40:12get it to
1:40:15build context for the main agent. So,
1:40:17let's say that I have I'm building an
1:40:19email system and I want the main agent
1:40:21to take some email ideas and rough them
1:40:23out. If I want to take a ton, let's say
1:40:25I'm going to take like 200 past emails,
1:40:29I'm going to feed them to this VA agent.
1:40:31he's going to make me an idea document
1:40:33from all those emails to then hand to
1:40:35the main agent. So, it's keeping the
1:40:37main agent as clean as possible. And so,
1:40:39um yeah, you guys will learn how to do
1:40:41this with time. But basically, the
1:40:43entire point of this kind of research
1:40:45agent in the Obsidian is it can do these
1:40:49small tasks that I don't want to fill up
1:40:51the main builder agents context with.
1:40:54So, I'm working on a complex feature. I
1:40:56don't want to just start asking
1:40:58questions about this one API
1:41:00documentation in my main thing because
1:41:02I'm just clouding up the context. So I
1:41:04will ask the research agent to just hey
1:41:06I'm working on this. Go find these
1:41:08documentation. Give me a summary. Teach
1:41:11me about this thing that I don't know
1:41:12about. So like you know walk me through
1:41:14it and then I'll go back to the main
1:41:15builder. So super cool. Another thing
1:41:18here is you you have to have cloud code
1:41:20like installed on your machine. So not
1:41:23just the application but cloud code in
1:41:25terminal. And so there's tons of
1:41:27tutorials on that. Literally go to cloud
1:41:29code website, copy and paste their bash
1:41:31command, open up the terminal app and
1:41:33then paste it in and you're good to go.
1:41:35Once you have that, you can open up
1:41:37Claudian or you can open up cloud code
1:41:39in terminal app in Obsidian. So like I
1:41:42said, um this uses a cheaper model. I
1:41:44don't really give it any context besides
1:41:46just the initial prompt that I send in.
1:41:48and it's just for researching,
1:41:50explaining things to me, creating and
1:41:51curating context for the main agent,
1:41:53organizing files, renaming, moving them
1:41:55around the project. And um I use it
1:41:57inside of whichever ID I'm in because
1:41:59it's just so so simple. And the value of
1:42:02this, like I said, is to keep the main
1:42:04agents context clean, creating better
1:42:06context, better systems, you know,
1:42:07better outputs. Cloud number three, test
1:42:11dummy cloud. So, if I'm using an LLM
1:42:15inside of my system, so for example, the
1:42:17plugin I gave you guys for co-work, the
1:42:19plugin I gave you guys for the second
1:42:21brain, the finance plugin, the system
1:42:24I'm creating is going going to use the
1:42:27LLM that you have access to. So,
1:42:29basically, it's using your cloud. And
1:42:31so, I have to make sure that it will
1:42:34work for you on your machine. And so, to
1:42:36do that, I have to have a test dummy cla
1:42:39basically. It's very simple. I just open
1:42:41up a brand new cloud chat in co-work
1:42:44because that's probably the one that
1:42:45most of you guys are going to be using.
1:42:46Um, I'll use a cheaper model and I will
1:42:49just test my plugin. I'll test my skill.
1:42:51I'll test my application, my automation,
1:42:53whatever. And uh, I will not give it any
1:42:56other context besides the context that
1:42:58you would have. I want to test it as if
1:43:00I am a user. And so yeah, it keeps the
1:43:03build claude clean and I'm able to bring
1:43:05notes back in from the crash dummy
1:43:07claude to give to the builder claude to
1:43:10improve the plugin to then test it
1:43:12again. So that is kind of the workflow.
1:43:15Number three, folder and context
1:43:17structure plus my juicy cloud MD. So if
1:43:21you not seen the principles modules yet,
1:43:23what are you doing, bro? Go watch them.
1:43:25They're literally the secret to all
1:43:27this. But the way I've structured my
1:43:29context is with systems thinking in
1:43:32mind. And so at the top we have archive.
1:43:35This is where I will just throw in old
1:43:37project versions or old context or let's
1:43:40say I just have a big like research PDF
1:43:43that I don't want cloud to read every
1:43:44time but I want to have it in archive
1:43:46and so I can reference it. I'll throw it
1:43:48in archive. It's just a folder of old
1:43:50stuff. Next I have attachments. And so
1:43:53Obsidian can reference pictures
1:43:55throughout the notes. And so I store the
1:43:57pictures in attachments. And so anytime
1:43:59I bring a picture into Obsidian, it
1:44:01knows to just put it in attachments and
1:44:03then reference it in the note that I'm
1:44:05talking about. And then journal logs.
1:44:08This is so so beneficial. Um, and if
1:44:10you're not lazy with this, this will
1:44:12change the way you work with Claude. And
1:44:14I'll show you guys, you know, examples
1:44:15here in a little bit. But as I'm
1:44:17building a complex thing with Claude and
1:44:20automation, I will take notes about what
1:44:23I'm learning, the tests I'm running,
1:44:25what failed, what did good, what I'm
1:44:27building that day, what are the problems
1:44:29are. And so that way whenever I spawn in
1:44:31a brand new Claude instance, I can
1:44:33literally tell him, "Dude, go review the
1:44:35last three days of journals. We're still
1:44:37working on the same feature. Look at the
1:44:39open problems and then let's just kind
1:44:41of get to work." So, it allows me to
1:44:43prompt him with all some like the memory
1:44:46of the last couple days without having
1:44:48to like reprompt him every single time.
1:44:51It gives him a little bit of like
1:44:52short-term memory uh to get that context
1:44:55in faster. Then the buildout folder is
1:44:58where we're actually working on the
1:45:00steps of the system. So, inputs and then
1:45:03process steps and then the output. And
1:45:06so this CloudMD once it interviews you
1:45:08on your project, it will create these
1:45:11folders for you. Um, these are just kind
1:45:13of placeholders to show you what is
1:45:15inside the buildout. But if I'm going to
1:45:17build like an email system, I might have
1:45:19an ideas folder. So how do we come up
1:45:21with email ideas? Then I might have like
1:45:23a rough draft folder which includes how
1:45:26Claude creates the rough drafts and us
1:45:28iterating on it. Then I might have like
1:45:29a finished draft folder, etc. or like a
1:45:32uploading folder. It's just the steps in
1:45:34the process.
1:45:36And then the source folder is for when
1:45:38you're creating plugins or applications.
1:45:41And so if you're building an automation,
1:45:42you're probably going to work with
1:45:43Claude in the project folder. So it is
1:45:46the source. But if you're building a
1:45:47plugin that other people are going to
1:45:49download or you're building an
1:45:51application which is going to be hosted,
1:45:52you know, on a server on a website, um,
1:45:55I will put the like finished source in
1:45:58that source folder. So like the finished
1:46:00plug-in version will be in that source
1:46:02folder. Um, that way I always know like
1:46:04what I'm building and then what is
1:46:06actually shipped, like what is the
1:46:07current live version. Um, you'll learn
1:46:10that as you go. And something I want to
1:46:12explain is that this is a template. So,
1:46:15every one of my projects doesn't look
1:46:17just like this. I've got small nuances.
1:46:20That's why it's essential that you get
1:46:22the principles and learn the skill of
1:46:24working with AI. This is going to take
1:46:25time. It is a skill, but hopefully I'm
1:46:28saving you hundreds of hours by
1:46:29explaining to you uh the good stuff of
1:46:31what actually works. So then we have
1:46:34important info. This is just a document
1:46:36where I save I don't know passwords or
1:46:38things that I just need to easily find.
1:46:40And then lastly, we have the build
1:46:42process, which is just a good little
1:46:44reminder of how to build efficient and
1:46:47good systems. Um it's kind of like a
1:46:49very short synopsis of part of the
1:46:51principal modules. So, it's really good
1:46:53to remind yourself where you're at and
1:46:55how do you build good stuff. Finally,
1:46:58the cloud.md.
1:47:00So, how I've structured this cloud.md
1:47:02is really to help you get the best start
1:47:05to your project as possible. And same
1:47:07for me because this is what I use. And
1:47:09so, all it does is it basically says,
1:47:11hey, the user is going to build one of
1:47:13these three things. An automation, a
1:47:15plug-in, or an app. So, first thing you
1:47:18want to do is interview the user. So get
1:47:21the best inputs as possible out of the
1:47:23user. Figure out what their process
1:47:26might be like what they're thinking
1:47:27about. Offer to help them do research.
1:47:30Um and then second is to poke holes in
1:47:32it. And so offer to review the plan that
1:47:35you just created. Third is to write the
1:47:37spec. So it will go through and write
1:47:39like which tools we're going to use, how
1:47:40we're going to build this. Four, it'll
1:47:42build out the folders. So like I said
1:47:44over here, it will change the names of
1:47:46these to match whatever you agree on in
1:47:48your plan. Then it'll confirm with you.
1:47:51Then it will rewrite this entire
1:47:52cloud.md. So it'll rewrite the cloud MD
1:47:56with context of your project based on
1:47:58the answers that you just gave it. Super
1:48:00super cool. I know it sounds
1:48:01complicated, but like I said, this is a
1:48:03skill. Just get in there, build with
1:48:05Claude. I'm going to give you guys three
1:48:06different levels of difficulty to get
1:48:08you up to speed so you can do this.
1:48:10Speaking of the three different levels,
1:48:13let's go over how I work through these
1:48:15automations. Now, something I want you
1:48:17to understand is that it doesn't really
1:48:19depend on your skill level. It's more
1:48:21about the complexity of what you're
1:48:23building. And so, I will use kind of
1:48:25like level one workflow if I'm building
1:48:27something that is very simple and
1:48:29doesn't require a ton of like, you know,
1:48:32effort and thought. And so, I'll show
1:48:33you guys what that looks like. Level two
1:48:37is going to be, you know, an automation
1:48:40or a plugin or even an application that
1:48:43you're going to build over the span of
1:48:44weeks. So something that's going to take
1:48:45a little bit longer. It's going to
1:48:48require more iteration. This is what I
1:48:50think of when you're going to actually
1:48:52give things to your team or give these
1:48:54to your audience or sell these. So I had
1:48:57to spend weeks on the finance plugin
1:49:00because other people were going to use
1:49:02it. You guys are going to use it. I had
1:49:04to make it simple. I had to test the
1:49:05edge cases. I had to make it as valuable
1:49:08as possible with different features. And
1:49:09so there's different, you know, features
1:49:11inside of the plugin that took me weeks
1:49:13to build. And that's when you would want
1:49:14to use kind of a level two setup and
1:49:16workflow.
1:49:18Lastly, level three. This is something
1:49:20that you're going to work on for months
1:49:21or years. So, I am like uh like five
1:49:25months in to building my first paying
1:49:27web application called Signal, which I
1:49:29absolutely love. And so, I've learned a
1:49:32lot through this. And building something
1:49:35that users are going to use and like
1:49:37you're going to store their data, it
1:49:39requires a whole different level of
1:49:41work. It is not just building a simple
1:49:44claude skill or an artifact. It requires
1:49:46serious iteration, serious thought, and
1:49:49that's when you want to use the maximum
1:49:50kind of level three workflow that I will
1:49:52show you guys. Um, and what's funny is
1:49:55like level three systems are really just
1:49:57made up of level two subsystems. And so
1:50:01this application has like you know 25
1:50:04different features that are each their
1:50:06own system that we had to combine
1:50:08together and uh do that in a way that's
1:50:10safe and userfriendly. So yeah it's
1:50:13really tough to do. So level one like I
1:50:16said is kind of just for stuff you're
1:50:17exploring or it's really really simple.
1:50:20Um I don't know what that was doing
1:50:21there. So all you have to do is kind of
1:50:22just like use the cloudmd and rip it
1:50:24which I'll show you guys examples. Level
1:50:26two, how you do this is um you're
1:50:29actually going to use my template.
1:50:31You're going to have Claw interview you.
1:50:33You're going to kind of build a V1 and
1:50:34ship it. And then the only difference
1:50:36between level two and level three is
1:50:38you're not going to do that much
1:50:40iteration. Like you'll test it out, you
1:50:41iterate it, but just because you're the
1:50:44one using it or your team's using it, it
1:50:46doesn't have to be perfect, per se. Um
1:50:49then lastly, level three is when you
1:50:51want to go all out. heavy thinking, lots
1:50:54of pruning, keeping up with your
1:50:55journals, exploring different options,
1:50:58restarting everything over again. And uh
1:51:00I'll walk you through what a a you know
1:51:02process looks like with this. But
1:51:04really, this is how you build amazing
1:51:07stuff, amazing systems with Claude. The
1:51:10first is, you know, you want to get as
1:51:11deep as possible on the inputs, the
1:51:13process, the outputs. So getting as much
1:51:16thought as you can inside of the
1:51:18pre-build context. Like I said, we'll go
1:51:20over it. Phase two is building the V1.
1:51:22So just getting an output even if it's
1:51:25not great. Like I said, watch the
1:51:27principles if you haven't. We go over
1:51:28these phases in depth. And then phase
1:51:30three is the big phase. It is iterating.
1:51:33It is making it work. It's pruning the
1:51:36context, rethinking the systems,
1:51:38rebuilding all the little subsystems
1:51:40inside of everything. Starting over,
1:51:42struggling, thinking, rebuilding. Um, I
1:51:45know you guys are probably like, KJ,
1:51:46this is a like copout step. Uh, this is
1:51:49like great instructions here, but I
1:51:51can't show you this. Like you have to
1:51:52struggle through this. You've got to
1:51:55build these automations there. Like
1:51:56there's no way I could tell you how to
1:51:59think clearly and struggle on a notepad
1:52:02with how to structure things. You just
1:52:04have to do it. You have to do it. Um so
1:52:06anyways, yeah. Then phase four,
1:52:08leveraging your automations. We're going
1:52:10to cover that in an entire module. Yeah,
1:52:12those are the three levels of workflows.
1:52:13And like I said, we'll go over each one
1:52:15here in just a second. I'll give you
1:52:16guys examples, but some important
1:52:18reminders. Keep it simple. Um, iteration
1:52:22is the key here. Improving your process,
1:52:24improving your automations over time.
1:52:26And the good news is AI gets smarter and
1:52:29smarter. Meaning like some of the the
1:52:31parts of this process might not be
1:52:32necessary in the future because of how
1:52:34smart AI has got. But there's one thing
1:52:36I can guarantee you. You have to be the
1:52:39one who has creative, clear, and
1:52:42innovative thinking. You have to be the
1:52:44one who is guiding Claude, pushing, you
1:52:47know, the agency, the human agency of
1:52:49like, I want to do this. How do we think
1:52:51outside the box to get this outcome? And
1:52:53over time, you should spend less time in
1:52:56front of Claude and more time in pen and
1:52:58paper. The better Claude gets, the more
1:53:00he can execute your vision. So, the more
1:53:03you should spend time cultivating your
1:53:05vision. And that is a hard thing to do.
1:53:07But that's why I gave you guys the
1:53:08principles modules because that's the
1:53:10only way I know how to think about it.
1:53:11And once again, the final reminder is
1:53:14what to build matters so much more than
1:53:16how to build it. So just like the you
1:53:18know this is really cool stuff about
1:53:19building but what matters so much more
1:53:21is like what you're actually working on.
1:53:23You could use the level three
1:53:25automations to build something that like
1:53:27you know I don't know just track
1:53:30something random that doesn't help
1:53:31anybody. I would highly encourage you
1:53:33guys to just try to build things that
1:53:34are as useful and valuable as possible.
1:53:37Solve problems that really uh are
1:53:39plaguing people and like you you know
1:53:41you want to solve. So that out the way.
1:53:43Let's go build something. Enough
1:53:44yapping. Let's go build. Let's get in
1:53:46Claude. Let's get into the template and
1:53:48let's go to it. All right, welcome to
1:53:49level one automations. Let's start
1:53:52building some stuff. So, here's the
1:53:54process for building these simple
1:53:56beginner automations. Also, if you've
1:53:59never even touched Claude, if you've
1:54:01never made anything with Claude, I
1:54:03highly recommend starting at level zero.
1:54:05I have a brand new Claude Co-work 101
1:54:07course, completely free. goes over all
1:54:09these things about settings, use cases,
1:54:12skills, plugins, uh your first
1:54:14automation, usage, lots of bonuses as
1:54:16well. Go check that out if you're brand
1:54:18new. I'm serious. Um all this is going
1:54:20to make way more sense if you actually
1:54:22have a little experience inside of
1:54:24Claude. So check that out. It's in the
1:54:26school. Uh probably link in description,
1:54:27but if not, just search up useful AI
1:54:29school. Now, level one automations. The
1:54:32entire goal of this level is just to
1:54:35just to build something kind of valuable
1:54:37with Claude to test out this new systems
1:54:40thinking you've learned to build
1:54:42something cool. So, let's do it. Here's
1:54:44the process. We're going to create a new
1:54:46folder just like that. Um, we're going
1:54:48to drag in my level one claim MD. So,
1:54:51not the entire template yet. We're just
1:54:53going to drag in a claud file. Super
1:54:55simple. You're going to rename it to
1:54:57just claw.md so that Claude actually
1:55:00recognizes what it is. Then you're going
1:55:02to open Cloud Co-work, select that new
1:55:04folder. So like select the folder you
1:55:06just renamed. Um, and then they say get
1:55:09started. Then you're going to answer the
1:55:11questions and build with Claude. So
1:55:12let's go do it right now. All right. So
1:55:14step one, we're going to create a brand
1:55:15new folder wherever you want to. So I'm
1:55:18going to call this um YouTube research
1:55:22level one. So now I have my new folder.
1:55:25I'm going to go to the link in the
1:55:26description of the school. Um, it's
1:55:28inside a classroom. Then go to start
1:55:30here and then go to YouTube resources
1:55:32and then click on the level oneclaw.md.
1:55:36I'm going to download this. It says no
1:55:37preview available because it is just a
1:55:39text file. There's no pictures, but you
1:55:41download it by clicking the top right or
1:55:43clicking the big yellow download button.
1:55:45Then I'm going to drag that to my
1:55:48desktop. I'm going to um actually drag
1:55:50this into our folder. I'm going to open
1:55:53the folder and I'm going to rename this
1:55:55to just cloud.md so that Claude knows
1:55:58exactly um what this is. Now the cloud
1:56:01app will actually register this as a
1:56:03cloud MD all caps. Um so now we're going
1:56:05to do we're going to open up Claude and
1:56:07get to work. All right. So now that I
1:56:09have Claude open, all I'm going to do is
1:56:12select that folder to work in. So I'm
1:56:14going to go to choose a different
1:56:15folder. Make sure you do folder and not
1:56:17project. I'm going to select a folder.
1:56:20Go to desktop. I'm going to select that
1:56:22folder we just made, which includes the
1:56:23cloud.mmd, which I gave you guys. I'm
1:56:27going to change this to act without
1:56:28asking because I feel like cloud is
1:56:30pretty safe these days. I'm just going
1:56:32to say um let's begin. And now what it's
1:56:35going to do is it's going to read that
1:56:37cloud MD that I showed you guys and it's
1:56:40going to interview you about your first
1:56:41automation or your beginner automation
1:56:44and then it's going to build it after
1:56:45it's done. Um it's going to create the
1:56:47folder. It's gonna do everything for
1:56:48you. All you need is just a folder, the
1:56:51cloud MD, and then cloud. Like I said,
1:56:53this is for beginner automations, but
1:56:55this is the best way to start because
1:56:57you can always take a beginner
1:56:59automation and upgrade it to like a
1:57:01level two or level three if you find
1:57:03that is really valuable to you and you
1:57:05want to make it more robust and better.
1:57:07All right, so Claude now has the first
1:57:08question for us. He says, "What are you
1:57:10building here?" Um, I'm just going to
1:57:12build an automation. So, basically, I'm
1:57:14just going to work with it in Cloud.
1:57:15It's not going to be a plugin or it's
1:57:17not going to be an application. And it's
1:57:18going to be just a uh a folder. So I'm
1:57:20going to say automation a plugin or an
1:57:22application. Sorry. All right. The next
1:57:24question is what comes out of it. So it
1:57:26wants to know what is the output? What's
1:57:28the point? And what does the perfect
1:57:30version look like? So I'm going to tell
1:57:31it to it real quick. Um I use something
1:57:33called Super Whisper. Uh you can get the
1:57:35free version. I have links about this in
1:57:38my school. I'm not affiliated or
1:57:39anything because it's free, but you can
1:57:40check it out. Okay, Claude. So, what I
1:57:43want to get out of this research um
1:57:45automation is I want to give you a video
1:57:48idea that I'm working on and then I want
1:57:50you to give me all the competing videos
1:57:54um based on the keyword demand that you
1:57:56can find. And so, we can probably use
1:57:59like the Vid IQ uh plugin for this um or
1:58:02connector, I mean. And then I want you
1:58:04basically to give me all the keywords
1:58:05around that idea that are doing well.
1:58:07Then give me some tags for the video.
1:58:10And then also find some videos and
1:58:12thumbnails that are a little bit broader
1:58:15than the actual video that I'm doing. So
1:58:17find me direct competitors and then find
1:58:18me um like broader niche competitors.
1:58:22And then based on all that data, kind of
1:58:24give me some YouTube titles and some
1:58:26thoughts about the thumbnails for my
1:58:28video.
1:58:30And all of this should be in like a nice
1:58:32HTML file as the output.
1:58:36Okay. So that is kind of some really
1:58:38rough just giving hit some stuff. But
1:58:40the better you give it on this context,
1:58:42the better you answer these questions,
1:58:44the better your output is going to be.
1:58:46I'm telling you guys. Um, but this is a
1:58:48good way to start just to make something
1:58:50get an output. All right, so Claude has
1:58:52taken that context and he has actually
1:58:55searched for some tools of what he could
1:58:57use. Um, there's a connector called Vid
1:58:59IQ which I've already using. So I'm just
1:59:01going to connect that. It's going to uh
1:59:03find the keywords.
1:59:05um and say what is your channel and
1:59:06niche the input when you bring me a
1:59:08video idea what form is it um okay so
1:59:12let's do this so I don't have any
1:59:14examples of research done before but the
1:59:17idea that I'm working on right now is a
1:59:19full AI automations course using claude
1:59:23and so the value of this video is going
1:59:24to teach someone how to build valuable
1:59:27systems and get crazy outputs um using
1:59:31cloud AI and using my templates it's
1:59:33going to be a long course a couple hours
1:59:35long. It's going to be awesome. So, that
1:59:36is the input. I want you to find kind of
1:59:38competing videos directly to that that
1:59:41value proposition and then also find
1:59:43videos that are just kind of in the
1:59:44general AI or claw niche.
1:59:48All right, cool. So, now we gave him a
1:59:50little bit more context and we'll see
1:59:51what he comes up with. All right, so
1:59:52he's asking a few more questions. He
1:59:54want me to run the real research on your
1:59:56idea? I'm going to say yes. Um,
1:59:59actually, uh, we'll build the system
2:00:01first, then run it. How deep's the
2:00:02research go? I'm going to go 15 each
2:00:05just to see if we can really juice this
2:00:07thing. All right, so now Claude is
2:00:08actually kind of taking this plan and he
2:00:11should be building the system. And so we
2:00:13should start to see some folders and
2:00:14things pop up. Like I said, this is
2:00:16level one. So you really don't have to
2:00:17look at this at all. You can just kind
2:00:19of run the system um and kind of get it
2:00:22get an output. And he's using this
2:00:24plugin that we found. But this is a
2:00:26little note here. In level two and level
2:00:28three, you would want to do way more
2:00:31research about what tools are available.
2:00:33You might want to test different tools
2:00:34like is Vid IQ good or should we use an
2:00:37API? Like these are all decisions that
2:00:39you have to make as the system designer
2:00:43because Claude's just going to head in
2:00:44the direction you say. It's just like we
2:00:46talked about in the principles the like
2:00:48how you start with Claude is so
2:00:50important because that's where he's
2:00:51going. Like it's it's autocomplete on
2:00:53steroids. And so you have to be the one
2:00:55to pump the brakes to slow down and say,
2:00:57"Hey, let's actually test this. Should
2:00:59we use Vid IQ or should we use the YTJS
2:01:03API or super data API?" Like you have to
2:01:06ask questions, go deep. Like I said, for
2:01:08level one, doesn't matter. But this is a
2:01:10point here. This is where like things
2:01:11differ from just rolling with Claude and
2:01:14actually freaking thinking about the
2:01:16system and building something um super
2:01:19dialed. So, as we can see here, Claude
2:01:21is actually building out the system in
2:01:23the forms of folders. These things don't
2:01:25matter much because, like I said, you're
2:01:26not really going to use uh too much
2:01:29context and use Obsidian. You're just
2:01:30literally going to build a system in
2:01:32here in a Cloud Chat, but you can check
2:01:34it out if you want. You can see kind of
2:01:35all the steps that it's doing. You can
2:01:37see the actual outputs, the HTML report.
2:01:40And so, yeah, let's see um if he can get
2:01:42some decent outputs with a very, very
2:01:44limited context that I actually gave,
2:01:46bro. All right. All right. So, what he's
2:01:48doing now is he's actually rewriting the
2:01:50CloudMD. So, the CloudMD I gave you was
2:01:52prompted to, you know, collect context
2:01:55from you, ask questions, figure out the
2:01:57build, but now that it's built, he's
2:01:59actually rewriting the CloudMD to have
2:02:01context about what you've already done
2:02:03and what what it's built. It's pretty
2:02:05crazy, but it's really cool. Like I
2:02:07said, you don't have to know how this
2:02:08stuff works in level one. Um, but just
2:02:10know like that's what it's doing right
2:02:12now is it's actually rewriting its own
2:02:14claude. So that if you open a brand new
2:02:16chat with this folder, that claude
2:02:19instance knows what's going on. It knows
2:02:20what the system is. It knows how to run
2:02:22the system, which is just really, really
2:02:24cool. And just like that, he is done and
2:02:26he actually has us an output. Um, so
2:02:29here we have pre-built context, the
2:02:31template, the cloud MD. Um, so I want to
2:02:33actually see this template report and
2:02:35see what we get. Let's pop that open and
2:02:37see. Um, so keyword demand, direct
2:02:40competitors, broader niche, tags,
2:02:42there's titles, thumbnails. Dope. So
2:02:44that's a pretty good looking template, I
2:02:45would say, to start with. So let's
2:02:47actually run this. So let me tell Claude
2:02:50that looks great. I want you to actually
2:02:52run um this search and get me actually
2:02:56finished output on the course idea that
2:02:58I talked to you about.
2:03:01Cool.
2:03:02And while he's doing that, we'll talk
2:03:04more about how to progress your builds
2:03:07into each level and how to kind of work
2:03:09your way up to like a finished live web
2:03:12app that people can pay for. But this is
2:03:14why you want to start with Claude first
2:03:16instead of immediately building an app
2:03:18is it allows you to go through and
2:03:20actually iterate really quickly because
2:03:22I can be like, "Oh, I didn't like that.
2:03:24Get, you know, move the tags down or
2:03:26change the titles like this or do this."
2:03:28um before you actually make it into a
2:03:30plugin because once you make it into a
2:03:31plugin and then into an application it's
2:03:34really hard to make these changes um but
2:03:37while doing it at this early stage just
2:03:38with claude you can dial it in and then
2:03:41wrap it into a plugin and just like that
2:03:43claude has an output for us so let's
2:03:46open up this uh research in Google
2:03:48Chrome and see what we got let's
2:03:50actually let's take a peek all right
2:03:52just like that uh full quad AI course
2:03:55okay the read uh keyword demand And
2:03:58okay, cool. So, we have all our keywords
2:04:00here. That's actually saves me so much
2:04:01time. Y'all have no idea how long that
2:04:03would take me to just type these in
2:04:04manually, come up with these based on
2:04:06every video. This already saves a lot of
2:04:09time. Here we have some direct com
2:04:11competition. And so, not bad. I don't
2:04:14like the formatting of the thumbnails or
2:04:16whatever, but it did find me videos um
2:04:19that are compet competing with the video
2:04:21that I'm going to make that you're
2:04:23watching right now. How how crazy is
2:04:24that? And so, I've got about 20 of
2:04:26these. pretty cool. Then I have some
2:04:29broader niche competitors, so things
2:04:31that are just in the AI space. Um, once
2:04:35again, not perfect, but I do have some
2:04:37thumbnails that I can look at for
2:04:38inspiration and uh packaging inspiration
2:04:40from these videos. So, honestly, you
2:04:42know, this is valuable. It's got tags
2:04:45for the video already. Amazing. Super
2:04:46valuable. We have some suggested titles
2:04:49here, and they're ranked based on kind
2:04:50of Vid IQ. I don't love these, but some
2:04:53of these are okay. Not bad at all. and
2:04:56then some thumbnail ideas. So, here are
2:04:58some concepts on that. And so, literally
2:05:01the automation is working. It is going
2:05:03from the input to an output and it is
2:05:06valuable. Now, you can already start to
2:05:09notice the problem with level one
2:05:11automations. And this is why people
2:05:12probably like yourselves use Claude and
2:05:14you're like, you know what, this is
2:05:16cool, but it's really not life-changing
2:05:19yet. And that is because co-work hides a
2:05:23lot from you. Like it's hard to go
2:05:25through and read these. It's hard to
2:05:27change them. Like you can't edit the
2:05:29text. Um you can if you like click on
2:05:31the code, I think. Uh yeah, you can't
2:05:34even do that. You have to tell Claw what
2:05:35to change. You can't really see the
2:05:37structure. If you want to open this up
2:05:39in a folder, it's honestly not that much
2:05:41better because now you have to go
2:05:42through and sort through all these
2:05:44folders by clicking into them and then
2:05:46trying to edit with text edit. This is
2:05:48the problem. So level one is great for
2:05:51systems that you don't have to dial in,
2:05:53but it becomes a huge issue when you're
2:05:55trying to really iterate on something
2:05:57and that's where the value comes because
2:05:59you need that human taste. And so what I
2:06:02want to take you guys through is we're
2:06:03actually going to rebuild this same
2:06:05automation but using the level two
2:06:07process. That way we can compare the
2:06:09outputs and you can see just how much a
2:06:11little bit of tweaking and preparation
2:06:13can really change an entire system. And
2:06:16you're probably wondering, KJ, um, is
2:06:17this an automation? Like, don't you have
2:06:19to build these things and make and
2:06:20whatever? Not anymore, guys. AI does it
2:06:23all. So, if I want to run this again.
2:06:25So, let's say I have an input, I have a
2:06:27new video, and I want this research. All
2:06:30I have to do is click on new task. I
2:06:32just select that folder, and all I do is
2:06:34say, you know, all right, Claude, I want
2:06:37you to do research for a new idea. This
2:06:39video, we're going to go over how to
2:06:41decrease the usage use in Claude. So,
2:06:44how to maximize usage as possible. I do
2:06:47that. It's using that folder which has
2:06:49already written the cloud MD. It's
2:06:51already told itself what to do when I
2:06:53ask this. And then I just run this off.
2:06:55Um, so I'm not going to obviously, but
2:06:57that's it. That's the whole automation.
2:06:59It knows what to do. All you got to do
2:07:01is select that folder when you want to
2:07:02run the automation and you're good. Um,
2:07:04and in level two, we'll talk more about
2:07:06this, but when you dial this in, that's
2:07:09when you create a plug-in. In that
2:07:10instance, you don't even have to use a
2:07:12folder. All you got to do is go new task
2:07:14and then you just unselect this and you
2:07:17just choose the plugin called like you
2:07:19know KJ thumbnail research and I can run
2:07:21it. And the cool part about that is you
2:07:23can then give that plugin away as a free
2:07:26value gift like to an to your audience
2:07:28or you can give it to your team so that
2:07:30your team members can run that and do
2:07:31research for you. So super cool. Let's
2:07:34move on to level two. Hopefully you guys
2:07:36can see this is dope but you can already
2:07:38see the limitations that level one has.
2:07:41Another important note on these level
2:07:43one automations is that if you were to
2:07:46go through and try to edit this system,
2:07:48you're now using a ton of usage because
2:07:51everything that you run in this chat,
2:07:53it's having to reread all of it. We we
2:07:56learned about this in the principles
2:07:57section. And so if you're going to edit
2:07:59this, it's really hard to edit it and it
2:08:01takes a lot of usage versus using Claude
2:08:05and the app. So having a separate
2:08:07instance of Obsidian, I mean not not
2:08:09Claude, having another Obsidian using
2:08:11Claudian or using Claude code in
2:08:12Obsidian, we can open up this folder and
2:08:15go get tactical. We can read the MD
2:08:17files. We can have another instance of
2:08:19Claude change them and reword them and
2:08:21then test it again without having to use
2:08:23a ton of extra usage. Anyways, I know
2:08:25that's complicated, but you will figure
2:08:27this out over time. I promise. But right
2:08:29now, you should just build something. So
2:08:31if you have not built anything yet with
2:08:33Claude like this, I want you to stop
2:08:35right now. Don't go to level two. Go
2:08:37download the cloud MD, open a new
2:08:39folder, run a new project, and get an
2:08:41output. Get something that is somewhat
2:08:44valuable to you, even though it's not
2:08:46perfect. And yeah, then you can move on
2:08:48to level two. So, I'll see you guys over
2:08:50there. All right, ladies and gents.
2:08:51Welcome to building level two
2:08:53automations. Now, the goal of level two
2:08:56is to build something a little more
2:08:57complex, to have your hands kind of
2:09:00working with Claude a little bit more,
2:09:01getting used to working with Claude and
2:09:03Obsidian, having kind of two different
2:09:05Claude instances for different um
2:09:07functions. This is the point of level
2:09:09two. So, if you haven't seen level one,
2:09:12go watch that one. Do it first. It's
2:09:13pretty simple. Now, let's get into level
2:09:16two. So, here's the process of building
2:09:18level two automations. First, you're
2:09:20going to download my full template. to
2:09:22the entire folder template inside of the
2:09:24school which a link will be in the
2:09:25description of course. You're going to
2:09:27unzip that folder on your desktop. So
2:09:29just rightclick extract here. Then
2:09:31you're going to duplicate that folder.
2:09:33I'm going say that again. Duplicate the
2:09:35folder. The folder is the template. You
2:09:37don't want to go into it unless you're
2:09:39trying to modify the template. If you're
2:09:41trying to make a new project, duplicate
2:09:43the template. It is a template. That way
2:09:46you have all my Obsidian presets.
2:09:48Anytime you want to start a new project,
2:09:49you just duplicate it. Then we're going
2:09:51to open that folder as a vault in
2:09:53Obsidian. Then you're going to be able
2:09:55to actually see that new project that
2:09:57the template that you used. Then we're
2:09:59going to open up that folder also inside
2:10:01of Claude and we're going to start
2:10:03building the automation level two. Um
2:10:06but this time you can see what's going
2:10:08on because you have Obsidian and you can
2:10:10actually work with a separate instance
2:10:12of Claude to improve the context to
2:10:14build things to clean things up as
2:10:16Claude is working on the main
2:10:18application. And then we're going to
2:10:19kind of iterate this with one good
2:10:21iteration and see what we can come up
2:10:23with for level two. So let's go build
2:10:24out that YouTube research automation
2:10:26again, but this time on a whole another
2:10:28level. So like I said, first step is
2:10:30downloading the template. So you want to
2:10:31come to the classroom in my useful AI
2:10:33school, click on start here, click on
2:10:35YouTube resources, and then click on KJ
2:10:38automation template folder. So it is a
2:10:41folder. So there's no picture, but um
2:10:42you can download it by clicking the big
2:10:44yellow download button. Please just like
2:10:47re-watch this and don't ask me question.
2:10:49I get so many questions about this. Just
2:10:50click download or click in the top right
2:10:52on download and boom, there you go. Now
2:10:55you have the zip file which is just a
2:10:57folder that is compressed. Drag this to
2:10:59your desktop. I want you to rightclick
2:11:01and extract here. And so boom, just like
2:11:05that, we have the folder outside of the
2:11:08zip. You can now just delete this zip
2:11:09file. You don't need it. And you can
2:11:11download again if you ever need it. So
2:11:13this is the project template. You can
2:11:15put this inside of your second brains in
2:11:17your project folder. You can put this
2:11:19wherever you have the things that you're
2:11:21building with Claude. Just stick this in
2:11:23there. And anytime you want to do a new
2:11:25project, it's very simple. You just
2:11:26rightclick and duplicate. That's it.
2:11:29Watch that again. If you guys need to
2:11:31rightclick, duplicate anytime you want
2:11:33to start a new project. Now, I'm going
2:11:35to rename this to the actual project
2:11:37we're going to build today. So, YouTube
2:11:39research
2:11:41level two. Awesome. Now, what I'm going
2:11:44to do is I'm going to open this up in
2:11:46Obsidian. So, I'm going to grab my
2:11:47Obsidian and I'm going to actually um go
2:11:50to manage vaults if you already have a
2:11:52vault open. I'm going to open folder as
2:11:54a vault.
2:11:56I'm going to choose that folder that I
2:11:58just created. So, YouTube research level
2:12:00two. I'm going to hit trust author and
2:12:02enable plugins.
2:12:04And I'm going to just do automatically
2:12:06check for plug-in updates, which I think
2:12:08is just uh cool to add. And now we can
2:12:12see this project which is really
2:12:13awesome. So now what's going on if
2:12:15you're new to Obsidian is I am actually
2:12:18inside this folder in Obsidian. So here
2:12:21I can see all the folders, the
2:12:22subfolders. So like this is like what
2:12:24you see in the file viewer. Same exact
2:12:26thing as what you're viewing over here.
2:12:27It's just an easier way to work with
2:12:29Claude and to see your files. And so
2:12:32what I'm going to do is um check out the
2:12:34build process, important info, all this
2:12:36good stuff. Now, before we actually
2:12:38start working with Claude, how we're
2:12:40going to make this level two better is I
2:12:43want you to spend some time really
2:12:45filling out the pre-build context. What
2:12:48I like to do is I'll actually fill in as
2:12:50much as I can in here, and then I'll go
2:12:53with a note, like a pen and paper, and
2:12:55I'll write out what I think the system
2:12:57should look like. I'll get really clear
2:12:59about what I want because like we talked
2:13:01about in the principal section, if you
2:13:03don't know what you want, it's very hard
2:13:05for Claude to read your mind and give it
2:13:07to you. But the clearer you can get on
2:13:09what exactly you want to see, the better
2:13:11Claude can make it. And that's why in
2:13:13level two, we're going to spend more
2:13:14time than just a few questions um
2:13:17answering it here. So now, how do we
2:13:19actually start building with level two?
2:13:21Well, same reason, same way that we
2:13:22start with level one, we just go and
2:13:24open up Claude. But before that, like I
2:13:26said, if you're wanting to use Claudian
2:13:28here to do like the research and VA
2:13:30tasks, um you can do it by just by
2:13:32popping open Claudian or you can click
2:13:34over here on the terminal and you can
2:13:36spawn in a terminal instance and then
2:13:38type in Claude. Um but make sure that
2:13:40you have Claude Code downloaded the
2:13:43actual CLI. So like I said in the
2:13:45original overview module of the
2:13:46execution, go download Claude Code in
2:13:49your actual terminal app. So, this will
2:13:51be able to give you both of these
2:13:53options to work in Obsidian. So, let's
2:13:55go start up this project. So, what we're
2:13:57going to do is we're going to go to
2:13:58Cloud Co-work or Cloud Code. Whichever
2:14:00one you like, they work the same. Um,
2:14:02I'm going to basically turn off all my
2:14:04connectors to start with, and I'm going
2:14:06to choose this brand new folder we made
2:14:09to work in. So, I'm going to go to
2:14:11desktop, choose research level two, and
2:14:13I'm going to say let's let's start, I
2:14:16guess, is what I'll say. Now, this is
2:14:19the exact same way that you start
2:14:23um with level one. The same exact thing,
2:14:25same process. The only difference is I
2:14:27want you to spend some time researching
2:14:31on your own how you should build this
2:14:33thing. So, a couple examples are what
2:14:36tool are we going to use to actually
2:14:38find the YouTube videos. And so,
2:14:40something that I might do is I might
2:14:42come inside of of Obsidian here. I might
2:14:45open up this Claudian and I will say
2:14:49um I'll just do this. Hey Claude, I'm
2:14:52going to work on this build here in a
2:14:54second. So I don't want you to start the
2:14:55build quite yet, but what I do want you
2:14:57to do is figure out what are the
2:14:59different ways that we could actually
2:15:01pull YouTube videos and find YouTube
2:15:03videos. Is this like through an API? Is
2:15:05there connectors? Like how can I
2:15:07basically search YouTube by using Claude
2:15:11to do it for me inside of an automation?
2:15:14So, I might spawn this off and do some
2:15:15research on that. And it's actually not
2:15:17connected. Great. So, I have to start a
2:15:18new conversation. Uh, try it again. But
2:15:22you get the point. I would actually So,
2:15:24there we go. We're actually building
2:15:25now. What I would do is I would send
2:15:26that um prompt off of like, hey, I want
2:15:29to actually go research this before I
2:15:31start. That way, I can fill out this
2:15:33pre-built context better. And then I
2:15:35would spawn in another instance and I'd
2:15:37be like, "Okay, Claude, I want you to um
2:15:40let's work. I want you to figure out
2:15:42like how can I find keyword demand? Like
2:15:45what tools can I use to figure out where
2:15:46demand is on YouTube?" I would spawn
2:15:48that off. And so I would do research to
2:15:51fill in this pre-built context. And that
2:15:53is the [clears throat] difference. So
2:15:54what I'm going to do is I'm going to go
2:15:56through I'm going to fill out this
2:15:57project. I'm going to just work on it
2:15:59for like, you know, a couple hours. and
2:16:02then we're going to see the difference
2:16:03in a level two automation and a level
2:16:05one automation and I'll kind of show you
2:16:07what I did and what the project looks
2:16:09like. So yeah, let's go do it. All
2:16:11right, so I now have kind of the level
2:16:13two of building this YouTube packaging
2:16:15research automation. The first thing I
2:16:17want to do is I want to show you the
2:16:18output. So if you watch level one, you
2:16:21saw the level of output that we had from
2:16:23the level one system. Here's level two.
2:16:26So here we have a little bit uh cooler
2:16:29looking thing off the rip. We have its
2:16:31like big title, which is honestly a way
2:16:33better title than the last ones. We've
2:16:35got kind of some thumbnail things here.
2:16:37And then the best kind of golden
2:16:39competition, who had the best video
2:16:41that's similar to mine. We've got the
2:16:43keywords most relevant to me. So, it
2:16:44kind of separated the keywords into most
2:16:47relevant to the video and ones that
2:16:48aren't relevant. Next, we have the
2:16:50competitor section. Wow, these look way
2:16:52better as far as just like, you know,
2:16:54the formatting looks better. Also, I
2:16:56have the ability to see which ones are
2:16:58outperforming the other videos on each
2:17:00channel. So, like this video is a huge
2:17:03outlier for this channel, which is
2:17:05really cool to see as well. Um, also
2:17:07these videos look like they're a lot
2:17:10more on point than the original level
2:17:12one automations videos. So, these are a
2:17:14lot more accurate. Um, pretty cool. And
2:17:17then we look at the broader inspiration.
2:17:20So, there's actually a lot more here.
2:17:22They're in this smaller format, which is
2:17:23nice. and they're separated by what type
2:17:26of broader it is. So this is just the
2:17:28claude AI. So broad as far as claude is
2:17:30concerned. Um then we have AI automation
2:17:34is just the keyword there for all of
2:17:35these. And AI automation agency. So all
2:17:38of these are fall in that bucket.
2:17:41Overall this is way better inspiration.
2:17:43So there's a lot better inputs for me to
2:17:46come up with thumbnails from this clot
2:17:48output. So overall we're already off to
2:17:50a way better start than level one. Like
2:17:53I said, I'll show you guys how I did
2:17:54this here in a little bit. Um, and it
2:17:56looks really cool. And I I can actually
2:17:57click on any of these videos, I think.
2:17:58Um, yeah, I can I can click on them,
2:18:00they'll open on YouTube. Pretty dope.
2:18:02Next section is title elements and
2:18:04ideas. So, I really like how it's done
2:18:07this by talking about the different um
2:18:10elements of titles I can use. Then it
2:18:13gives me some titles based on my voice.
2:18:15So, it actually knows how I like titles,
2:18:18which is huge. This is like a big This
2:18:20these titles are way better. Like go
2:18:22back and watch level one, look at the
2:18:23titles, and look at these. These are a
2:18:25lot more like a KJ video. I might even
2:18:27use literally some of this, which is
2:18:29awesome. Um, experiments worth testing.
2:18:32So, Claude actually gave me some out of
2:18:34the box titles that aren't like mine
2:18:36that could be worth looking into or
2:18:38stealing inspiration from. And then, uh,
2:18:41which ones Vid IQ like the best? We have
2:18:43some tags, pretty similar to the other
2:18:45one. And, um, yeah, that's pretty much
2:18:47it. But overall, way better output. Like
2:18:50way better. Like I said, it looks
2:18:52similar, but it's not about the looks
2:18:54here. Like, why am I building this
2:18:55automation? It is to come up with a the
2:18:58best packaging for the video you're
2:19:00watching right now. So, what I can
2:19:01guarantee you is I will be using what
2:19:03you see right here to make the title and
2:19:05thumbnail that you just clicked on,
2:19:07which is crazy. So, this is very
2:19:09valuable to me and it's a lot more
2:19:10valuable than the last one because these
2:19:12are a lot more accurate to the video I'm
2:19:14trying to make and it gives me better
2:19:16insights and I can make better judgments
2:19:18on my own stuff. So, pretty pretty cool.
2:19:21Now, how do I do it? How do we build
2:19:23this level two? Well, this is what the
2:19:25obsidian looks like. I know it looks a
2:19:27little crazy, but I promise don't don't
2:19:29worry about this. I built this in one
2:19:30day. So, here's the pre-build context. I
2:19:33took time to really answer those
2:19:35questions. So, instead of just firing
2:19:37off like a couple short prompts to
2:19:39Claude like we did in level one, I
2:19:41actually took the time to sit down and
2:19:42figure out what am I building? Why am I
2:19:45building this? What do I want to see
2:19:47come out of the system? What does a
2:19:49perfect output look like? Like why am I
2:19:51doing this? Um what is a process that
2:19:54might work? Also, what I did is I tested
2:19:57out different things. So these are like
2:20:00um I tested like where is a good place
2:20:02to search these videos. Do I use the Vid
2:20:04IQ MCP super data or YouTubeJS?
2:20:08I tested different tile variations um
2:20:10based off of my own skill and I did all
2:20:13this and putting it into these different
2:20:15folders. Um, so this looks complicated,
2:20:18but the only difference here, seriously,
2:20:20the only difference is that I took way
2:20:22more time to fill out the pre-built
2:20:24context. I know that sounds crazy, but
2:20:27that's literally it. The closer you can
2:20:29get to where you want to go, like, or
2:20:31the clearer you can get the output for
2:20:33Claude of what you want, he can help you
2:20:36do the rest. And this log I just kept
2:20:38manually. So I literally said like okay
2:20:41what input will I give Claude and what
2:20:44is the output I want out of this system
2:20:45like what do I want to see and so by
2:20:47clarifying that I came up with questions
2:20:50like should I use Vid IQ or should I use
2:20:52super data or is there something else I
2:20:54can use and so guess what I spawned in
2:20:57Claudian and I asked it hey do some
2:20:59research on what kind of tool I could
2:21:01use to find these videos so it's all
2:21:03about spending time to really make
2:21:06things clear building the system with
2:21:08Claude as opposed to level one where we
2:21:10just let him make assumptions and run
2:21:12with it. Now we actually got a lot a lot
2:21:15clearer. And so yeah, that is really
2:21:17cool. And now if if I wanted to make
2:21:18this into a plugin now, I could because
2:21:21I think it's really dialed. The input is
2:21:23literally just, hey Claude, here's an
2:21:25idea for a video. Do the research. The
2:21:27output is this um amazing document which
2:21:30literally does all the work that used to
2:21:31take me 4 hours at least a video, maybe
2:21:34even eight hours of video and now does
2:21:36it in minutes. and it does it better
2:21:38than I did, which is absolutely
2:21:40incredible. So, yeah, this is ready to
2:21:42then make into a plug-in or automation.
2:21:44We'll talk about this in um the
2:21:46leveraging your automations um module,
2:21:49but just know like this is when you get
2:21:51to the point where you're getting really
2:21:52good outputs with just a you know files
2:21:54and claude, you're ready to actually
2:21:56package this into a plugin that you
2:21:59could give to other people and you can
2:22:00test that plugin first and then give it
2:22:02to them and they can do the exact same
2:22:03thing for themselves. So yeah, let us
2:22:06move on to level three where we really
2:22:08get deep on the crazy automations and
2:22:10building full applications. All right,
2:22:12welcome to the final level. Level three,
2:22:14automations. Now, the goal of this is
2:22:16really the end goal of everything we're
2:22:18building to build something valuable
2:22:20with multiple features, subsystems that
2:22:22you can eventually sell or give away
2:22:25like I do to people like y'all so that
2:22:27you can use them and create value in
2:22:29your life. Like this is really kind of
2:22:31entrepreneurship one-on-one, but it's
2:22:33online. So, it's online building
2:22:35building um valuable solutions using the
2:22:38tools that we have. This is like max
2:22:40level. So, let's get into it. The
2:22:42process uh for level three for setting
2:22:44up is the exact same as level two,
2:22:46right? You want to download the folder
2:22:48template. You want to unzip it. Um you
2:22:51want to then copy and rename it. So, if
2:22:52you haven't watched level two, go watch
2:22:54that. I won't redo it again. Then,
2:22:56you're just going to open whatever your
2:22:58renamed project is in Obsidian and then
2:23:00open that in Claude and you're good to
2:23:02go. So, what is the difference between
2:23:04starting at level two and level three?
2:23:06Well, level three is you really want to
2:23:10fill out the pre-build context in depth.
2:23:13Like, do some serious thinking. When I
2:23:15build level three automations, I mean,
2:23:17I'm like literally on a whiteboard like
2:23:20mapping out stuff, testing out different
2:23:22processes, um, because I want to start
2:23:25in the best way possible. And normally I
2:23:28don't start by building a level three
2:23:29automation. Honestly, it starts out as a
2:23:32level two and then once I decide that
2:23:34man, this is really valuable. I should
2:23:36like build this out into a plugin or
2:23:40like a like a full app like I will then
2:23:42transition to level three. So, a good
2:23:44example we'll look at here in a second
2:23:46is my finance plugin. I built that for
2:23:48myself. It was not a plugin. It started
2:23:50out as just a simple automation. Then I
2:23:52realized, wow, I could build this into
2:23:54an artifact. Like, let me actually step
2:23:56this up. It became a level two project.
2:23:59Then it became a level three project and
2:24:00I was like, "Man, this thing is
2:24:02in-depth. I worked on it for over a
2:24:04month. It has different features. I'm
2:24:05ready to package it up into a plugin and
2:24:07give it away to you guys." And so, um,
2:24:10yeah, you don't really start as level
2:24:11three. Start it off as level two. Fill
2:24:13out that pre-build context as much as
2:24:15you can, but you know, don't don't get
2:24:18too narrow thinking like leave some room
2:24:20for Claude to kind of poke holes and
2:24:21stuff. And then you start it off by once
2:24:24again opening up that project in claude
2:24:26as a folder and then just running say
2:24:28let's get started do whatever. Now
2:24:30here's the difference between level
2:24:32three and level two. It's really on the
2:24:34back end. It's learning how to work in
2:24:37cloud like in obsidian with claude and
2:24:40uh changing the context like pruning the
2:24:42context iterating iterating iterating.
2:24:46Um there's a reason why there's so many
2:24:49software engineers but very few of them
2:24:52are billion-dollar founders. It is not
2:24:54the technical ability is it is the
2:24:56judgment and the clear thinking and the
2:24:59ruthless iteration to to simplify and to
2:25:02make things better and to focus that is
2:25:04something that I cannot teach you. I can
2:25:07make you aware of how important it is
2:25:09which is what I hope this course has
2:25:10done but I can't teach it to you. So I
2:25:12want to make you aware of it. I want to
2:25:14give you examples of how like I've
2:25:16started doing this in my life and the
2:25:17things that I've built and then I want
2:25:19you to do it yourself and to learn this
2:25:21skill. So, that being said, let us um
2:25:24dive into what a level three automation
2:25:26actually looks like um behind the
2:25:28scenes. All right, what you guys are
2:25:29seeing right now is my signal
2:25:32application. So, this is a real
2:25:33application that is live and we have
2:25:35paying users on this thing. So, shout
2:25:36out to you guys who are actually, you
2:25:38know, subscribed and doing this um using
2:25:40our app. But it is dense. I mean, we
2:25:42have everything here. Billing, um, we
2:25:44have a big like settings. Like, there's
2:25:46so many features on this app. And I'll
2:25:48show you kind of what the core feature
2:25:50is. And that is to help you go from just
2:25:52a basic um, problem that you have that
2:25:56you want to find videos on to having a
2:25:58ton of perfect [snorts] videos for you.
2:26:01So, I'm just going to do like, what do
2:26:02you want to learn? Oh, hold on. That's
2:26:04the developer tools. Let's try this now.
2:26:08I want to learn how to do YouTube
2:26:10thumbnails really well for my YouTube
2:26:12channel. And so I want to learn how to
2:26:14like, you know, use Photoshop, what
2:26:16other tools I could use to just make
2:26:17better thumbnails to get clicks and get
2:26:19more views and help more people. So I'll
2:26:22just throw that off. Um, and we'll spam
2:26:24that here. And so I'm just going to say
2:26:26some experience. How deep you want to
2:26:28go? Let's just go deep. And then it's
2:26:30going to reading your goal. So now this
2:26:32application that I built is going to
2:26:35give me this brand new project and give
2:26:37me videos to help me learn. It's like
2:26:39YouTube but on steroids for people who
2:26:42learn. Um yeah, pretty cool. So I'm
2:26:44going to show you guys like how I built
2:26:45this entire application, how I did the
2:26:48um UI for it, the actual like player UI,
2:26:51all that good stuff. It took a long
2:26:52time, took months, but this is what a
2:26:54like level three project really looks
2:26:56like. So yeah, the input was I had a
2:26:59problem I want to learn more about. And
2:27:00then the system, the output presents me
2:27:03these videos that I can then save to my
2:27:06project. So I can click on them and save
2:27:07them. Um, and then it'll actually
2:27:09download the transcripts and kind of
2:27:11give me summaries and stuff for them.
2:27:13Then if I want to watch these videos, I
2:27:15can do it right here inside of the
2:27:17Signal app. And once again, I don't you
2:27:19guys do not have to use or download
2:27:20this. Um, this is not supposed to be
2:27:22like a sales pitch. Honestly, I'm more
2:27:24focused on my agency stuff right now,
2:27:26which I'll show you guys that in a
2:27:28second. So, if you guys want us to build
2:27:29you custom software, me and my
2:27:31incredible team of engineers. If you're
2:27:33a business owner, hit me up. I'll build
2:27:35you everything we're talking about um
2:27:36for a really good price. So, you know,
2:27:38book a call here if you actually want
2:27:39that. But back to Signal. Um I want to
2:27:43show you guys how I made all this stuff.
2:27:44Like, you can literally do different
2:27:45notes. I can add notes in here. I can do
2:27:47different sections. I can uncheck these.
2:27:49I can skip directly to this section. I
2:27:52have the ability right to change.
2:27:55I have the ability to see a summary. I
2:27:57can control the player from the notes. I
2:27:59can see more from this channel. Like
2:28:01this is a full live big paid
2:28:04application. That's what I'm saying.
2:28:05This is what we're building in level
2:28:06three. It is it is a lot. Um but
2:28:08hopefully you guys can see how cool this
2:28:10is and you're excited. So let's go take
2:28:11a look at how I did this. So as you can
2:28:13see here, this is the actual files, the
2:28:16context, the project of Signal V3. Um so
2:28:20on the left you can see like the archive
2:28:22the journal stuff like that and then we
2:28:24see the process of going from the the
2:28:26first step so context to query and then
2:28:29filtering batch one like all this you
2:28:31don't have to know about but this is how
2:28:33deep it can get and so if you see my
2:28:35journal of like building this one
2:28:36feature for you can see my brainstorming
2:28:38which was all of my like doing all the
2:28:41research beforehand for this one
2:28:42feature. The feature that I'm building
2:28:44here is literally talking about just how
2:28:46to go from an input someone has a
2:28:49problem to literally finding ranking
2:28:52creating tons of like um results for
2:28:56people can like filter through. So I
2:28:58won't go deep in all this cuz honestly
2:28:59it just it won't help you guys. But
2:29:01here's what it looks like. It's me every
2:29:03day like putting out my thoughts for
2:29:05version two, iterating, testing over and
2:29:08over again. And so from going from
2:29:10context to query, I have t tons of
2:29:13variations of just this skill. And I
2:29:14won't show you because this is literally
2:29:15like the money maker. This is what I
2:29:17make money from. Um, but it's figuring
2:29:19out how do I go from that one step. So
2:29:22we talked about in principles like this
2:29:23is where the principles get extremely
2:29:25useful because otherwise you're going to
2:29:26be very confused watching this. But if
2:29:29the input is user has a problem and the
2:29:32output is we give them every single
2:29:34video that's going to perfectly solve
2:29:36their problem, the process is what we're
2:29:38building. How do we get there? How do we
2:29:40get from input to output? How do we make
2:29:42that output better and better and
2:29:44better? Um, and so this is how I'm doing
2:29:46it. The first step is literally taking
2:29:48their their context and creating search
2:29:51queries to then push into APIs. So
2:29:54figuring out what API do I use, what
2:29:56queries do we actually create, how do I
2:29:58give the skill to Claude or whatever LLM
2:30:00I'm using the application to actually
2:30:02create those queries. Like I said, I
2:30:04can't show you because this is my my
2:30:06money maker, but this is me testing it
2:30:08out over and over and over again. And
2:30:11then once I actually have the results
2:30:13back, like how many videos do I collect?
2:30:15How do I filter them? And that's where
2:30:16the second step comes in of like the
2:30:18filtering batch. And then how I actually
2:30:21test this before going to application is
2:30:24I literally have it um give me like uh a
2:30:28result. So like here's the actual
2:30:29report, the filtered report. And so I'm
2:30:32I'm doing the main value of the
2:30:34application without it being in the
2:30:36application. And so here I can have it
2:30:38do an HTML where I say, "Hey, this is
2:30:41the input. I'm a copyriter. I want to
2:30:42get clients through cold outreach, but I
2:30:44don't know how." Boom. Now I'm able to
2:30:46see clearly how the algorithm that I've
2:30:49programmed and how Claude is ranking
2:30:51these videos and I can see which ones
2:30:53that it's cutting off and not showing.
2:30:55Uh I can see all the ones that is
2:30:56bringing in. So this is behind the
2:30:58scenes of what's going on in the
2:31:00application. And here I can iterate it.
2:31:02I can get really like like in the weeds.
2:31:05This is what I'm trying [laughter]
2:31:07the only value I'm trying to give to you
2:31:09guys by this is just showing you like
2:31:10there's levels to all of this. And if
2:31:12you're going to make something that's
2:31:13really really valuable, you have to
2:31:15spend a lot of time on even the simplest
2:31:18the simplest sounding feature. And so
2:31:20that's level three. It's going deep.
2:31:22It's taking days to iterate and redo and
2:31:25prune and try again and all this good
2:31:28stuff. And then I had Claude give me a
2:31:30customer view so that I can see how it
2:31:32actually shows these videos in the
2:31:34project to a customer. So if we go back
2:31:36to this project, this is using the exact
2:31:38same skill right here. So what it's
2:31:41showing here is what it's going to show
2:31:42me in the HTML. But instead of having to
2:31:44like push this to a live application, I
2:31:47can test just the algorithms right here
2:31:50on my computer and then send it off to
2:31:52um my software engineers who actually
2:31:54implement all this. So super cool. I
2:31:56also tested the refresh feature. So, I'm
2:31:59like testing like how much does it
2:32:00refresh? Which ones does it refresh? Um,
2:32:03if they click refresh, like what what
2:32:05happens at the top, and how does it
2:32:07actually choose which videos to show.
2:32:10It's really cool. And so, all this is
2:32:12what it looks like to really go deep in
2:32:14level three. But, like I said, you will
2:32:17get here over time. I just want you to
2:32:19understand how to work with Claude in
2:32:22the app and in Obsidian. how to see what
2:32:24you're working on with Claude so you can
2:32:26actually read it, understand it, and
2:32:28then make changes. Um, and you can see
2:32:31all this in here. Brainstorming, old
2:32:33skill, questions to answer, test notes,
2:32:35problems to solve. You have to dig deep.
2:32:38You got to understand. And then also,
2:32:40don't be lazy. Take time to keep a
2:32:42journal of what you're learning each day
2:32:45so that you're thinking clear. And then
2:32:48Claude, whenever you spawn him in, you
2:32:50can just say, "Hey man, read day two. I
2:32:52want you to check out day two and then
2:32:54let's start working on this problems at
2:32:55the end of day two. And so yeah,
2:32:57hopefully this was valuable. I know it
2:32:58seems like a little bit confusing, but
2:33:00that's the whole part of level three,
2:33:01man. Like this is building
2:33:03million-dollar software. Like it's
2:33:05really hard to explain in a course, but
2:33:06hopefully I've given you at least some
2:33:08nuggets or I've changed the way that you
2:33:10think about this stuff so that um you
2:33:12can build these better. Like I said in
2:33:14previous ones and in level like one and
2:33:16two, you work up to level three. These
2:33:18are things that you want to spend months
2:33:20and years on, but just start with level
2:33:22one. Move on to level two. When you're
2:33:24ready, you will naturally progress into
2:33:25level three, and you'll use all these
2:33:27features from the template. Yeah, I hope
2:33:29you guys enjoyed. So, now I'm going to
2:33:31actually go take a look at how you
2:33:32leverage these things that you're
2:33:34building. So, I'll see you guys in the
2:33:35next module. All right. So, how do you
2:33:37actually leverage these automations to
2:33:39uh bring them to the market and to
2:33:41create value to generate leads or to
2:33:43make sales or, you know, do whatever you
2:33:45want? This is the most efficient
2:33:47progression that I found for bringing
2:33:49these things into the market. So the
2:33:51first step is starting with just an
2:33:53automation. So this is you with a folder
2:33:56in cloud. We just talked about this in
2:33:58the the previous modules about the
2:33:59different levels. But even when I'm
2:34:01designing an application, I start with
2:34:04just an automation, just project files,
2:34:06just me and Claude working locally on my
2:34:09machine. That is because when you start
2:34:11to worry about the user interface and
2:34:14other people like kind of using it and
2:34:16like how to actually wire up the the
2:34:18coding on a live site and implementing
2:34:21it where it doesn't fick with your
2:34:22existing features on application things
2:34:24get messy. But when you build it locally
2:34:28and you get the exact outcome you want,
2:34:31the exact output, talking about systems,
2:34:33you know, when you get the the desired
2:34:35output, that's when you can go ahead and
2:34:37move on to productizing it. And so I
2:34:40highly recommend that you iterate
2:34:43heavily locally just with HTML files,
2:34:46just with getting the output you want
2:34:48with Obsidian, with Claude. Um, and then
2:34:51once your value proposition is dialed,
2:34:53like the core thing you do, the core
2:34:55thing your automation does, that's when
2:34:57you can package it. You can make it into
2:34:58an application, you can make it into a
2:35:00plug-in. And that brings us to step two,
2:35:03the plugin. So, when I'm doing things
2:35:05for signal, just like you saw with this
2:35:07like kind of signal, the new algorithm,
2:35:09what I'll do is once it's dialed
2:35:11locally, I will package it up into a
2:35:14plugin. That way I can send it off to my
2:35:17team or my friends or beta testers and
2:35:19they can see what are the results
2:35:22without having to have a finished
2:35:23application yet. Um, and so basically
2:35:27the cool part about this is you have
2:35:28people can can test it without it
2:35:30needing to have any kind of UI. And so
2:35:32the only thing I care about in this one
2:35:34feature of Signal is someone has a
2:35:37problem. How good is our application at
2:35:40giving them the exact videos they need
2:35:42to solve that problem? That's what I
2:35:44want to figure out. And that's a complex
2:35:46problem if I'm really going to solve it.
2:35:47And so by making it into a plug-in after
2:35:49it's done lo like locally working, I'm
2:35:52able to give it to more people who can
2:35:54test it and not just me, people with
2:35:56different problems, different
2:35:58perspectives and they can tell me how
2:36:00good the actual algorithm and like you
2:36:02know the decisions were from the system.
2:36:06And so the cool thing is real people you
2:36:08know test it. Now to leave this phase is
2:36:10real people actually use it. So, I know,
2:36:14you know, we all have friends and family
2:36:15who are gung-ho on like checking out our
2:36:18cool stuff. But before you make
2:36:19something into an application, um I
2:36:22highly recommend that people actually
2:36:24use it a lot. So, people actually use
2:36:26your stuff, not just test it, but they
2:36:28use it in their daily life. That's a
2:36:30good sign that you know what, I should
2:36:32build this into an application and uh I
2:36:34should actually charge for this and
2:36:36actually spend the time putting this,
2:36:38you know, deploying it, getting it live,
2:36:40doing the safety and stuff. Now, I'm
2:36:42extremely blessed. I have a team of
2:36:45engineers. One of them is an incredibly
2:36:46talented ex Google engineer. Shout out
2:36:47to you, Kaden. He finally quit his job
2:36:49at Google to do this kind of stuff
2:36:51full-time, which is amazing. And working
2:36:53with other companies, but um he does all
2:36:56the backend to keep these things safe.
2:36:58So, I will warn you when you go to do an
2:37:00application, do not charge people for it
2:37:02until you have made this thing super
2:37:04super safe and you actually know what
2:37:06you're doing on the software like
2:37:08scalable side of things. And if you want
2:37:10to hire us to do it, we just started our
2:37:12brand new agency. We're like launching
2:37:13it officially this week, 780. We will
2:37:16build you amazing software like you've
2:37:18seen that we've already built for
2:37:19ourselves here um for your small
2:37:21business. We're really excited to do
2:37:22this. I'll be doing kind of the business
2:37:24thinking and I'll have engineers
2:37:26implementing it, making it safe, and
2:37:28building you custom software. So yeah,
2:37:30there's the sales pitch for our agency
2:37:31if you are interested. Link in the
2:37:32description. Yeah, that is the best way
2:37:35I found to leverage these things. Start
2:37:37with just a project like a folder and
2:37:39claude. Move to a plug-in. Get other
2:37:41people to test it. See if they actually
2:37:43like it and use it. And then if they do,
2:37:46start moving it to an application. So
2:37:48come up with UI, build the branding,
2:37:50wire everything up to happen on a live
2:37:52site, and then you can actually, you
2:37:54know, send it to people, charge it, and
2:37:56uh yeah, make some amazing value builds
2:37:59and make some money. So yeah, that is
2:38:00how to leverage your automations the
2:38:02best way that I found. All right, let's
2:38:04break down one of my favorite
2:38:04automations I've made in recent months,
2:38:07which is my finance plugin. So, I
2:38:10released a plugin which helps you do
2:38:11your finances from start to finish in
2:38:13Claude, helping you with everything. And
2:38:15uh it's pretty awesome because I did not
2:38:18think a finance video would get any
2:38:19views and it's actually doing incredibly
2:38:21well because it's not just a video. I'm
2:38:23giving away a solution, an actual
2:38:26automation I'm giving to people for free
2:38:28to solve a real problem. And you guys'
2:38:30comments and emails have been
2:38:31incredible. So, thank you so much for
2:38:32that. But I'll show you guys how I did
2:38:34this. Like, how did I actually manage to
2:38:37build this thing so that people can get
2:38:39this like software inside of their cloud
2:38:42and it work for every single person? It
2:38:44have all the amazing features that it
2:38:46does. How did I do this? Well, let's
2:38:48dive into the project and uh let's take
2:38:50a look. All right, so now we were inside
2:38:52my finance plug-in project. So you can
2:38:54see this is kind of like an an older
2:38:56version of the template that you guys
2:38:58got in this course, but this is actually
2:39:00I made a lot of strides learning how to
2:39:02make this template through building this
2:39:03plugin. So here we can see my CloudMD.
2:39:06It's very simple. It's just explaining
2:39:08like this is the plugin that I'm trying
2:39:10to create. This is what I'm doing. This
2:39:12is my workflow. Here's what the plugin
2:39:14is supposed to do. Um here's how we're
2:39:17going to work on it. And then whatever
2:39:18the heck this is. I didn't even write
2:39:20that. He just, you know, did that in
2:39:21here. And then the folder structure
2:39:23layout. Um, so pretty simple. Then some
2:39:26rules here that I just came up with as
2:39:28we were working on the thing together.
2:39:30But what I would do is these are the
2:39:32source folders. And so this is like all
2:39:35the skills in the plugin. So I told
2:39:37Claude, "Hey, I'm going to make a
2:39:38plugin, but I want to be able to see the
2:39:41skills that you're actually working on
2:39:43before you package it into the finished
2:39:46plugin." And so that's what the source
2:39:48folder is for in the template I gave you
2:39:50guys. is for seeing like what the actual
2:39:53version you're working on, all the files
2:39:55in here so that you can actually edit
2:39:57them um instead of them only being
2:39:59inside of the claude app which is hard
2:40:00to edit. So what I would do is I would
2:40:04it would give me the plugin. So I would
2:40:06like see the plugin here. I would
2:40:08download this into my claude and then I
2:40:10would run it. I would test it as if I am
2:40:12a user if I'm if as if I'm one of you
2:40:14guys. Um, and I would test different
2:40:16cases and edge cases. And then what I
2:40:18would do is I would just continue to
2:40:20make notes of like what was going wrong
2:40:24and what I needed to change. And then I
2:40:26would keep like you see all these tests
2:40:28here. Like it's so messy. But real
2:40:29projects just get messy, guys. Um, I
2:40:32would build the dashboards and HTMLs
2:40:34before I even like did like I would just
2:40:37test it with test data before I even did
2:40:39the artifact. And so that's what I would
2:40:41do. And so from start to finish really I
2:40:45just wrote out a big like pre-build
2:40:48context of the purpose of this
2:40:50dashboard, what the features were. Like
2:40:53I thought through all of it really
2:40:55before I touched Claude. And that's the
2:40:57secret. That's the kind of thing that uh
2:40:59everyone's scared to say and I'm kind of
2:41:01scared to tell you guys, but you are the
2:41:06edge. You are the alpha. Like you've got
2:41:08to think about what you want. Claude
2:41:11just builds it, which is really, really
2:41:13cool. And he's a good thinking partner.
2:41:15But the reason why I feel like people
2:41:16have have liked my stuff is because I
2:41:19did the thinking for the dashboard. I
2:41:21used it in my life. I figured out what
2:41:22features were useful to me and my wife
2:41:26with our finances and then I came back
2:41:28and would build a brand new version. I
2:41:30literally start from scratch and be
2:41:31like, "All right, Claude, here's the
2:41:32process I think is going to work. Here's
2:41:34what we should do." Um, I I wrote out
2:41:38like what are the elements that we want?
2:41:40So, the the template uh the setup flow
2:41:42and questions like how I literally wrote
2:41:45these out by hand of how should the flow
2:41:49go for someone actually setting up the
2:41:50system. Like this is all written by hand
2:41:52and then I had Claude actually go and
2:41:56implement it to make it the actual
2:41:58plug-in. Um here's a handoff prompt. So,
2:42:00I will show you guys example of that of
2:42:02I was halfway through working on
2:42:03something and I was like, "All right,
2:42:05Claude, we're running out of context.
2:42:07you know, write yourself a handoff
2:42:08prompt so that we can open up a new chat
2:42:10and keep working and then a bunch of
2:42:12sample data for testing it, screenshots.
2:42:14Um, yeah, and that's how I finally got
2:42:16to the finished plugin, which is has
2:42:18these different skills. So, three
2:42:20different skills. Um, I explained to
2:42:22Claude how the skills should work. I
2:42:24gave references for how the dashboard
2:42:26should look like inside the plugin. I
2:42:28made sure and I read all of these skills
2:42:31just by hand. Like I literally go
2:42:33through and read every line to make sure
2:42:35Claude isn't getting something wrong or
2:42:38adding something that shouldn't be in
2:42:39there. And uh yeah, so I mean it took
2:42:42like over months of working on it and
2:42:44testing it to actually build a plugin
2:42:46that was worth giving to you guys in the
2:42:48video, but this is what makes these
2:42:51things worth it, guys. Like it is not
2:42:53simple building scalable automations
2:42:56that actually add value. If it was, we'd
2:42:58all be millionaires. But I just want to
2:43:00encourage you to take the time to dig
2:43:03deep. Go through every single subsystem,
2:43:06every single step of input to output.
2:43:09Figure out what is cloud doing, how do I
2:43:11make this better, research tools,
2:43:13research how you can solve problem in
2:43:15different ways, um find better examples,
2:43:18do all that good stuff, and you'll be on
2:43:20your way to making some really, really
2:43:22cool plugins. And so yeah, basically the
2:43:24workflow for this was like I said work
2:43:27in this Obsidian with a claude co-work
2:43:31chat and then we would build a plugin
2:43:33and then I would open up a new co-work
2:43:34chat where I would download the plugin
2:43:36and then I would test it in that fresh
2:43:38chat, take away all I would make notes
2:43:41on a notepad of like the things that
2:43:42were wrong. I'd go back to the builder
2:43:44cloud and I would tell it, hey, these
2:43:46things were wrong. Re-edit the the
2:43:48plugin inside the source. I would read
2:43:50the source files to make sure it got the
2:43:52changes right. Then I would say, "Okay,
2:43:54ship this into a plugin. It would
2:43:56package it up into a plugin and I would
2:43:58repeat. I'd give that to the test cloud,
2:44:00see what went wrong, take notes, fix it,
2:44:03package a new plugin, repeat." And
2:44:05that's what I did over and over again to
2:44:07make this finance dashboard. So
2:44:08hopefully that makes sense and helps you
2:44:10guys. But yeah, let's move on to another
2:44:12automation overview. All right, so
2:44:14welcome to this plugin and automation
2:44:16that I set up for one of my one-on-one
2:44:18clients. And so he's a custom home
2:44:20builder and he wants to figure out how
2:44:22can I give blueprints to Claude and get
2:44:25an actual takeoff and estimate on the
2:44:27other side of things. And so I won't
2:44:29show you all this because this is his
2:44:31project. Um so I want to keep it, you
2:44:33know, proprietary to him. But basically
2:44:36how we built this is we started by just
2:44:38having a project in Obsidian with
2:44:40Claude. And so it started out as not
2:44:42even a plug-in. It was just an
2:44:43automation. And we said, "Hey Claude,
2:44:45here's some blueprints. here's the
2:44:48process we think could work. So, we know
2:44:50that we're going to have to give you a
2:44:52couple variables like the formulas and
2:44:54what the current prices are. And then we
2:44:56want you to kind of take the blueprints,
2:44:59apply the formulas, the prices, total
2:45:01everything up, and give us a takeoff and
2:45:03estimate. So, like I said, I'm sorry I
2:45:05can't show you guys, but this is like
2:45:06one of their builds. And uh so, yeah,
2:45:08that's what we did from there. Once that
2:45:11worked, we made it into a plug-in so
2:45:13that he could test this in his cloud and
2:45:15basically like run um takeoffs on his
2:45:19own to see like what was going wrong and
2:45:20what we needed to fix. And then once it
2:45:22looked good, we made this into an actual
2:45:24artifact so that now everything has some
2:45:27UI to it. He can easily change the
2:45:29variables here instead of having to
2:45:30answer and claude. He can start a new
2:45:32project. he can go through the
2:45:33measurements, um, formulas, all that
2:45:35stuff, and then get the output that he
2:45:37wants, which could save honestly a lot
2:45:39of hours per week because this stuff
2:45:41sucks manually. But, um, yeah, Cla's now
2:45:43helping out, which is really, really
2:45:45cool. Um, also, I want to teach on if
2:45:48you guys want, we actually just launched
2:45:49our agency for this. Uh, we are now
2:45:52offering to build custom software
2:45:54automations for small business owners.
2:45:57So, through doing the 101 ones, I
2:45:58learned that not everybody has a deep
2:46:01passion for AI and systems like I do.
2:46:04Like, I just I love it. I figured
2:46:05everybody loves it, but not everybody
2:46:07loves it as much as I do. And so, we
2:46:09decided that launching an agency would
2:46:11be a great um addition to the market to
2:46:14fill a hole of people who just want to
2:46:15pay us to build really complex and
2:46:19amazing custom software and like AI
2:46:22agent stuff for them. So, if you're a
2:46:24business owner, you're watching this,
2:46:25you're like, "Dude, freak yeah, like
2:46:27I'll just pay you money. You build this
2:46:28stuff to make my business better." Then
2:46:30book a call with us here and we can get
2:46:31to work on something epic for you, which
2:46:33is super cool. So, hopefully we have
2:46:34slots by the time you're seeing this.
2:46:36Yeah, that's what I want to encourage
2:46:37you guys if you do want to get some
2:46:39client work done from us. Um, this is
2:46:41just a plugin, but we obviously build
2:46:42things way more professional when it
2:46:44comes to like real applications. So,
2:46:46yeah, check that out. But, let's move on
2:46:47to the next project breakdown. All
2:46:49right, welcome to the breakdown of my
2:46:51actual real live scalable paid
2:46:54application. So, this is the first app
2:46:56I've ever built which has like real
2:46:58paying users on it who are getting value
2:47:00and subscribing every single month,
2:47:01which is just amazing. Honestly, a dream
2:47:04come true. And I want to show you guys
2:47:05kind of how I built this. So, we went
2:47:07over in level three how I built like the
2:47:10core value prop algorithms and like what
2:47:13happens behind the scenes. So, I figured
2:47:15in this module, we'd look at how I built
2:47:18the UI and how things actually operate
2:47:20and how we came up with um this UI and
2:47:24you know, basically how I built the app
2:47:26to make it look the way that it looks.
2:47:28So, this is if you're like just viewing
2:47:29it as an outsider and you haven't signed
2:47:31up yet. Let's take a look at what it
2:47:33looks like when you've actually signed
2:47:34up and can use it. So, this is like the
2:47:36homepage. Um on the homepage, you can
2:47:40add different topics. So, mine are very
2:47:42simple. It's Runescape and Claude. And
2:47:44you can click on different topics and
2:47:45add subtopics. So if I want to look at
2:47:46like Claude skills, I can type that in
2:47:49here. It'll then change the search and I
2:47:51can go through and see all these videos
2:47:53like a custom YouTube algorithm for me.
2:47:57I can sort them based on views or upload
2:47:59date. I can add a brand new topic. So
2:48:01I'll add something like um US World Cup.
2:48:05Do that. So shout out to the American
2:48:07gang. Hopefully we win going into next
2:48:10week. Yeah, I can search this and it
2:48:12will then just bring me all the best
2:48:14videos um for myself by doing that. So,
2:48:17super cool. I can add these videos to
2:48:19different projects. So, the entire goal
2:48:21of this app is to learn and so I have
2:48:24all these different projects for
2:48:26learning about YouTube thumbnails or
2:48:27actually there's one I was working on a
2:48:29lot which was um YouTube lighting and so
2:48:32figuring out what to light. So, I
2:48:34lighted this intro by learning from this
2:48:36stuff which is pretty cool. So these are
2:48:38the videos that are inside my project
2:48:39that I actually save to them. I can save
2:48:41them by just clicking on this button.
2:48:43And so everything you see here, I spent
2:48:45a lot of time designing exactly how it
2:48:47looks. And so that's what we'll kind of
2:48:49cover how I did that um right now in
2:48:52this project. So this is the Signal V2.
2:48:55This is how I designed all of the things
2:48:57you see here. And so the way that this
2:49:00project works is I create the HTMLs, the
2:49:04designs, and I hand them off to my
2:49:06developer Kaden who works with me with
2:49:07our agency and with this app. And uh he
2:49:10actually puts them into the live
2:49:12application. And so what I'm trying to
2:49:14do here is just to get the actual um
2:49:19looks right. Like that's the whole goal
2:49:21of this system is to get the UI. So I
2:49:23have a separate project and system for
2:49:25getting the things to work. And this is
2:49:27how I get it to actually look cool. And
2:49:30so what I do is um every single page has
2:49:33its own folder. So it has its own kind
2:49:35of sub project. And then in there I just
2:49:37work on the different versions of what
2:49:39I'm looking at, what the goals of the
2:49:40homepage are. And so I want Claude to
2:49:43know like whenever we're working on the
2:49:45homepage, I want it to know what the
2:49:47point of the homepage is, what the point
2:49:49of the player are, is what are the what
2:49:51are the features. And so this is just a
2:49:53different way to structure your
2:49:55projects. Also, I have the design
2:49:56system. So, like all the things that
2:49:59make um signals branding are in here so
2:50:02that Claude can reference. And so, if I
2:50:04want to design a new like page or I want
2:50:07to iterate a page, what I'll do is I
2:50:09will spawn a new Claude instance, he
2:50:11will knows, he knows in the cloud MD
2:50:13like what we're doing here, what the
2:50:14purpose of Signal is, what the folder
2:50:16structure is, um like our text stack.
2:50:19And then I will just say, "Hey, we're
2:50:21working on the homepage today. I want
2:50:22you to look at the homepage folder. look
2:50:24at the homepage goals, look at where we
2:50:26left off on V3, and then um let's start
2:50:29working on changes for V4. That's it.
2:50:32I'll literally start and then hand it
2:50:35off to Caden whenever the thing is done.
2:50:37And so then I'll, you know, pull up the
2:50:39HTML. I will go through and test
2:50:42different things. So I'll test like the
2:50:44brightness of the chips. I'll test um
2:50:47what like the one-click UI looks like.
2:50:49So whether it's a a chip or if it's if
2:50:51it's a ribbon or a magnet, I will just
2:50:54iterate on each little part of the
2:50:56feature like crazy. And when I find
2:50:59something that I like, what I'll do is I
2:51:01will get Claude to say, "All right, I
2:51:03think this version is done." What I want
2:51:05you to do is to create a handoff for
2:51:08Caden. And so based on the cloud MD, it
2:51:10knows who Kaden is and what the purpose
2:51:12of, you know, this workspace is. And
2:51:15then it writes a handoff. So, it will
2:51:17tell Kaden, hey, here's the V3. Here's
2:51:20what your cloud should do. And so, what
2:51:22Kaden does is he takes this um into his
2:51:25cloud and actually builds it on the live
2:51:27site, which is epic. Now, um I also do
2:51:30build applications like without other
2:51:32developers. And the way that you do this
2:51:34is you literally just hand this off to
2:51:36another clot that is connected to your
2:51:38GitHub that actually builds the app. If
2:51:40you guys want a full course on like
2:51:41building applications, that is a whole
2:51:43different beast. But let me know in the
2:51:45comments or let me know in the school if
2:51:46you want that. And uh I can convince
2:51:48Kaden, my my Google developer, to
2:51:51actually teach you guys the real sauce,
2:51:53like the backend stuff that no one talks
2:51:55about on YouTube. I can show you guys
2:51:57how to make these graphics and branding
2:51:59and features and UI and all that jazz.
2:52:02But yeah, hope you guys enjoyed and uh
2:52:04I'll see y'all in um I guess the next
2:52:06course. All right. So, the last thing I
2:52:08want to say to you guys is thank you so
2:52:09much for watching this. I hope this
2:52:10course has been insanely valuable. If it
2:52:12is, um, and you you want me to keep
2:52:14these things free, then please send this
2:52:16to your friends. I really just want to
2:52:18share as much knowledge of as God has
2:52:20given me as possible. I consider myself
2:52:21to be extremely blessed beyond belief.
2:52:24Um, God has blessed me more than I could
2:52:27ever even say. I mean, just the fact
2:52:30that he sent his son Jesus to die for me
2:52:32for my sins and how broken I am. I am
2:52:35just grateful every day to be alive and
2:52:37I'm so honored that he's allowed me to
2:52:39make this stuff free for you guys. All I
2:52:41ask is that you share this with somebody
2:52:42and help them get value from it as well.
2:52:45And a couple of last things of
2:52:46encouragement is is this um this is a
2:52:50skill and just like how you cannot make
2:52:54a course on like a basketball or make a
2:52:57course on carpentry that's going to
2:52:59immediately give someone thousands of
2:53:01hours of skill. The same thing applies
2:53:03here. If you're seeing some of this
2:53:05course, you're a little discouraged
2:53:06about how complicated this is and
2:53:07everything. I want you to know I didn't
2:53:10touch Claude six months ago. I had no
2:53:11clue how to use AI. I cooked my usage
2:53:14rate in about 2 minutes of using cloud.
2:53:16I did the dumbest stuff. Trust me, I
2:53:18made dumber mistakes than you can make.
2:53:21But that was the whole point. I made
2:53:22mistakes quickly. And so if you want to
2:53:24learn this stuff, if you want to be a a
2:53:26a master of AI, like we have to put in
2:53:29hours. So just build stuff. Have fun.
2:53:32Don't get caught in perfectionism.
2:53:34Build, build, build, and you will get
2:53:36better. You'll learn how these machines
2:53:38work. You'll learn how you work and how
2:53:40you can prompt them better. So, I hope
2:53:43this was encouraging. If you like this
2:53:44course, you know, then uh share it with
2:53:46somebody. Don't don't comment or don't
2:53:47like. Just just share it, you know.
2:53:49Share it with somebody who this could
2:53:50help. And I will see you guys in the
2:53:52next one.

About this transcript

This page contains the full transcript of FULL 3 HOUR COURSE: Automate Your Business With Claude AI (2026) by KJ Rainey, generated from the public captions YouTube serves with the video. The transcript has 34,465 words across 4,678 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 translate the transcript into 100+ languages.

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, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.