5,000+ Hours of Using Claude in 20 Minutes (Beginner to Pro) — Transcript
Full transcript
- 0:00So, in today's video, I'm going to give
- 0:01you the five levels to use Claude
- 0:03properly, condensing all of the
- 0:05knowledge that I've accrued running my
- 0:0730-person AI media company and hundreds
- 0:09of agents over the last 18 months,
- 0:11condensed into a single piece of content
- 0:13with the five highest leverage things
- 0:15that you need to know. This video is
- 0:17equally going to be applicable to you
- 0:19whether you're just using Claude to
- 0:21become more productive in your personal
- 0:22life, or whether you are building and
- 0:24scaling a big startup. I personally
- 0:26spend over $10,000 a month in Claude
- 0:29tokens alone. So, as a business, we've
- 0:31been forced to optimize and look at
- 0:33exactly how to use Claude across the
- 0:35business to get the best and most
- 0:37efficient results. So, subscribe to the
- 0:39channel if you want to stay at the
- 0:40cutting edge of AI, and I'm about to
- 0:42make you a Claude expert. All right, so
- 0:44before we get into the more advanced
- 0:46stuff, let me cover the three surfaces
- 0:48of Claude. Firstly, you have the chat.
- 0:49This is what you've probably interfaced
- 0:52with most if you've used Claude, where
- 0:53you can basically ask questions, review
- 0:55documents. There's honestly a lot of
- 0:57functionality now built into the chat,
- 0:59especially with the additional plugins
- 1:01and connectors, which we'll get into
- 1:02later in the video, that Claude has now
- 1:04added to make this even more useful. So,
- 1:07if you just use the chat, you can still
- 1:09extract a lot of value out of Claude,
- 1:11but in my opinion, the way to level up
- 1:13even further, instead of using the chat,
- 1:16is to use co-work. Because Claude
- 1:17co-work is what enables Claude to read
- 1:20local folders and documents on your
- 1:22computer. So, if you're just using the
- 1:23chat, it doesn't have access to any of
- 1:25your context, only the instructions that
- 1:27you feed to a project. Whereas if you
- 1:30use co-work, it can tap into your local
- 1:33memory system and context, and edit
- 1:35local files, which makes it even more
- 1:36powerful. And then the third surface
- 1:38within the Claude desktop application
- 1:41is, of course, Claude code. You can also
- 1:43use this through another harness like
- 1:45Visual Studio Code, or through the
- 1:46terminal, but for simplicity's sake, you
- 1:48can use it on the desktop just fine. And
- 1:51this is the next step up from Claude
- 1:53co-work. So, if you're vibe coding
- 1:55applications, if you're doing builds,
- 1:57it's going to be more efficient and
- 1:58overall a better user experience to use
- 2:01Claude Code. But out of the three, where
- 2:03I recommend people start building their
- 2:05habits on Claude is actually on Cowork
- 2:07because you get a similar user interface
- 2:09to the chat, but you gain the advantage
- 2:12of being able to still code technically
- 2:14and edit local files on your computer.
- 2:17So, I guess which one I use depends on
- 2:18the task. If I'm just on the go and I'm
- 2:20brainstorming content ideas, I'll still
- 2:22use the chat. It's fine for questions
- 2:24and, you know, light document reviews
- 2:25and stuff like that. But if I want
- 2:27Claude to access the memory system that
- 2:29I want it to see, I use Cowork and if
- 2:30I'm doing a bigger build, I use Claude
- 2:32Code. Now, before we get into level
- 2:34number two where I'll actually show you
- 2:36how to configure and manage your own
- 2:38memory system inside Cowork, let me
- 2:40touch on what might be the most
- 2:42underrated feature of Claude and that is
- 2:45the connectors capability in the
- 2:47application. Connectors essentially
- 2:50enable you to level up the chat because
- 2:52it can connect to other data sources and
- 2:55other applications via an MCP, which you
- 2:58can essentially think of as a bridge.
- 3:00So, instead of having to go into Higgs
- 3:01Field, for example, to produce video or
- 3:04photo-based content, you can simply add
- 3:06the Higgs Field connector and stay
- 3:08within Claude, but get access to Higgs
- 3:10Field's models like Seed Dance, like
- 3:12Gemini's Nano Banana. So, you can
- 3:14actually use Claude as a harness through
- 3:15connectors, which basically means the
- 3:17place that aggregates things, the user
- 3:19experience layer. So, you don't need to
- 3:21be hopping from AI tool to AI tool. Most
- 3:23people don't know this, but through
- 3:24connectors, you can actually use GPT
- 3:26within Claude. That is for certain tasks
- 3:28where you think GPT would be the better
- 3:31model. You can also add Google Drive, so
- 3:32it can view and edit things on your
- 3:34drive, which is super handy, especially
- 3:36for our company drives, which we manage
- 3:37on Google Drive. You can add Notion, so
- 3:39it can edit your to-do list and you can
- 3:41use Claude as a personal assistant. And
- 3:43the one I really like is FMP and also
- 3:45Quiver, which enables me to tap into,
- 3:47which I discuss on my finance channel,
- 3:49all sorts of data around stocks, crypto,
- 3:52my portfolio, so I can get live data
- 3:54based on stock earnings calls and market
- 3:57movements. The issue is if you just use
- 3:58the Claude chat and you don't use any
- 4:00connectors, you are doing yourself a
- 4:02disservice because you're just going off
- 4:03what Claude knows. Whereas, there are
- 4:05data companies like FMP that are
- 4:07literally just built around niche
- 4:09financial data. So, if you use the
- 4:11connector, every time you chat to
- 4:13Claude, you're going to get much better
- 4:14responses. For example, I can use Claude
- 4:17via FMP right now to analyze Nvidia's
- 4:21earnings call. This will tell me
- 4:22everything I need to know about the last
- 4:24earnings report for one of my
- 4:26investments. And if you want to know how
- 4:27to add a connector, just go into Claude
- 4:29settings, scroll down, click on
- 4:31connectors, and then you can search from
- 4:32a pre-made library of connectors. You've
- 4:35got Gmail, Slack, Google Calendar,
- 4:37pretty much anything that you can do on
- 4:39the internet can now be done by Claude.
- 4:41Honestly, one of the most underrated
- 4:42tips because everyone's talking about
- 4:44Claude code or the Claude chat and not
- 4:46addressing the fact that Claude actually
- 4:48needs access to information. This is the
- 4:50best way to get the most accurate
- 4:51information out of Claude and it's now
- 4:53inbuilt into the app. All right, now you
- 4:54know how to get more power out of
- 4:56Claude, but unless you give it the right
- 4:58context about yourself and about your
- 5:01business, then you could have all power
- 5:02in the world, but it'll just be useless.
- 5:04I think we've completely moved away from
- 5:06a prompt engineering environment.
- 5:07Remember when you had to like sit down
- 5:09and write these massive prompts and
- 5:11focus on giving it context and all this
- 5:13information because that's what you
- 5:14previously had to do to write a good
- 5:16prompt. Well, that environment is
- 5:18completely shifted now. Well, now I
- 5:20blurt out a lot of my prompts. I either
- 5:22do it through voice, through
- 5:23WhisperFlow, or I just write badly
- 5:25put-together prompts because it doesn't
- 5:26matter anymore. Because as long as
- 5:27Claude has the context about my
- 5:29business, it'll be able to apply what
- 5:31I'm trying to say to my real context.
- 5:33So, instead of putting time into
- 5:35learning how to prompt, put time into
- 5:37building a memory system that you truly
- 5:39own, which more importantly can be then
- 5:42ported to any LLM. If you want to use
- 5:43GPT, if you want to use Gemini in the
- 5:45future, whatever you want to use, you
- 5:47will own your own memory and Claude
- 5:49won't own your memory. Because unless
- 5:50you own your memory, Claude is a black
- 5:52box. Obviously, if you've ever spoken to
- 5:53the chat, it might remember things about
- 5:55you, but you don't have any sovereignty
- 5:57over what it's remembering or why it's
- 5:59remembering a certain thing. You might
- 6:00notice sometimes Claude, you know, is
- 6:02really weird, it'll bring up like a
- 6:03random fact. And then when you're
- 6:04talking about buying a new car, it'll
- 6:06bring up, "Oh, but you know, remember
- 6:07the new camera you were supposed to
- 6:08buy?" It'll just like inject all of this
- 6:11random stuff. So, to get around that and
- 6:13to have full sovereignty and ownership
- 6:15over your own memory, what you want to
- 6:17do is create a memory system. So, there
- 6:18are two ways to set up a memory system.
- 6:20The first way is to set it up locally,
- 6:22which I'll show you, and the second way
- 6:24is to set it up on the cloud, which I'll
- 6:26also run you through. Now, both have
- 6:27pros and cons. Locally is easier to
- 6:29manage and is probably what I would
- 6:31recommend for building your first
- 6:33context and memory system. But then, if
- 6:35you're going to create applications in
- 6:36the future, and if you're really
- 6:38hopping, you know, if you're more
- 6:39advanced between multiple LLMs, you may
- 6:41want to consider putting it on the
- 6:42cloud. As long as you have good
- 6:43protection, like two-factor authenti-
- 6:45cation. The other benefit is, if I don't
- 6:47know, I spill water over my laptop and
- 6:48my laptop dies, I have a backup of all
- 6:50my memory and I don't lose it. Maybe if
- 6:52it's like a basic context profile on
- 6:54yourself, it doesn't matter so much. But
- 6:55if you were to build out a full memory
- 6:58system for your business and something
- 6:59happened to your computer and the file
- 7:01got corrupted, you'd be distraught. So,
- 7:03that is where maybe considering a cloud
- 7:05option makes sense. Or of course, a lot
- 7:06of people that just have their memory
- 7:08systems locally and then every now and
- 7:09then they'll do a backup of their
- 7:10computer and they'll upload it to a
- 7:12drive or something, you know, so they
- 7:13have a cloud backup. Firstly, going to
- 7:15create a local memory system, this is
- 7:16what I'd recommend. Just create a folder
- 7:18on your computer. This is my AI brain.
- 7:20It essentially has context, so this is a
- 7:22file with all the context about myself,
- 7:25my business, what I do. To create this
- 7:27for the first time, all you need to do
- 7:28is do a voice prompt insertion with
- 7:30Claude. Ask it to ask you questions
- 7:32about yourself and about your business
- 7:34and it will write an entire MD, which is
- 7:36a markdown file. Don't do PDFs because
- 7:38Claude finds it harder to read. And
- 7:40also, make sure when you create an MD
- 7:42file that you scrutinize and you go line
- 7:45by line through each line and you put a
- 7:47bit of manual work in up front to make
- 7:49sure that what it's saying is correct
- 7:51because what's going to happen in the
- 7:52future is every business decision, every
- 7:54prompt is going to be based on those
- 7:56facts. So, you know, if you say that you
- 7:58have a $100,000 net worth and it
- 8:01accidentally transcribes, you know, a
- 8:02million-dollar net worth, then all
- 8:04future spending decisions are going to
- 8:05be based on, you know, 10x your actual
- 8:07net worth. That's obviously a big issue.
- 8:09So, you want to make sure that each line
- 8:11is actually, you know, true to you. The
- 8:13next thing you want to do is create an
- 8:14instructions.md
- 8:16file. This file tells Claude how to
- 8:18behave. So, you can put in preferences
- 8:20like, you know, always make your
- 8:22responses super analytical, abbreviated,
- 8:24don't ramble, don't confirm my bias,
- 8:27think strategically, and very
- 8:29importantly, in the instructions file,
- 8:31you want to put a line saying, "Notify
- 8:34me every single time you make a change
- 8:36to my memory.md file and make sure that
- 8:39every substantial new piece of
- 8:41information is used to update materially
- 8:45the memory.md file." So, what is the
- 8:47memory.md file? This is essentially
- 8:49Claude's brain. So, it's going to have
- 8:51all the context on you and what you tell
- 8:52it. So, every single time you give it
- 8:53new information like, "Okay, I just
- 8:55landed a new big client for my
- 8:57business." or, "You know, I'm pivoting
- 8:58more into SaaS products and I'm going to
- 9:00have SaaS attached to the info product."
- 9:02You would tell that to Claude and if you
- 9:03have the instructions set up correctly,
- 9:04your memory.md file gets automatically
- 9:06updated. And every now and then, what I
- 9:09recommend is you manually go through the
- 9:11memory.md file and you call it down just
- 9:14to make sure that it's manageable cuz
- 9:15the problem with Claude or any LLM for
- 9:17this matter is if the memory context is
- 9:19too bloated, sometimes Claude actually
- 9:21gets confused and the response time gets
- 9:23slower and it burns more tokens because
- 9:25it's reading through this five-page
- 9:27memory.md file. So, what you want it to
- 9:29have is just the core memory. Then, what
- 9:31you can do is in the instructions.md
- 9:33file, you can have a note that there is
- 9:35another folder attached called, for
- 9:37example, AI Edge YouTube, which is one
- 9:38of my business, or whatever your
- 9:40business is called, that has additional
- 9:42business context, only read that folder
- 9:44when I'm asking a question applicable to
- 9:47the information in that folder. Because
- 9:49within that folder, that's where you can
- 9:51have all of your business context and
- 9:53data. So, this is quite a light example
- 9:55for this video, but you could literally
- 9:56have an MD file or HTML mind map of your
- 9:59entire business, your funnel strategy.
- 10:01Like, this is where you can just dump
- 10:02and save all of the more complex
- 10:04information about yourself and your
- 10:06personal life, or
- 10:07>> [snorts]
- 10:07>> you know, your business if you're
- 10:08running a business. And if the
- 10:09instructions are set up correctly, then
- 10:11it won't be reading through that entire
- 10:13folder every single time that you need a
- 10:15response to a question, or once again,
- 10:16it'll just bloat the system. So,
- 10:18essentially, creating a memory system
- 10:21comes down to context, instructions, and
- 10:23memory. If you do that, and you plug it
- 10:25into Cohere, by selecting the folder
- 10:27that you store all of this information,
- 10:29every single time you do a prompt on
- 10:31Claude, it can edit that folder, and it
- 10:33can review that folder so your responses
- 10:36are tailored to you. This is the single
- 10:37biggest improvement that you can make to
- 10:39using any AI because it actually has the
- 10:41context on you. Now, the second option,
- 10:43as I said, is the cloud option. I won't
- 10:44go into too much depth, but you can
- 10:46build your own custom system on
- 10:48Superbase using Postgres. The way to do
- 10:51this is basically just ask Claude to
- 10:53build it for you, it'll be able to build
- 10:54it. Or secondly, you could use a
- 10:56purpose-built tool like Supermemory. Or
- 10:58if you're using an application to vibe
- 11:00code like Base 44, they also have their
- 11:01own proprietary memory systems. I think
- 11:03Supermemory is great for people that
- 11:05want a unified experience across all of
- 11:07their LLMs and apps. So, it's worth
- 11:09considering. Superbase, you have some
- 11:11memory sovereignty. If you ask Claude,
- 11:13once again, what is best for your needs,
- 11:15depending on the complexity of your
- 11:16business, it'll be able to tell you.
- 11:18But, I would recommend, if you have not
- 11:19built a memory system, to start with a
- 11:20local folder so you understand it, and
- 11:22then upgrade to cloud later. All right,
- 11:24now it's time to move on to level three,
- 11:26which is the first level where Claude is
- 11:29literally going to be doing the work for
- 11:30you. And I just want to remind you,
- 11:32because now things are getting more
- 11:33advanced, that the entire cheat sheet
- 11:35from today's video will be available for
- 11:37free down below in my free school. This
- 11:39is a brand new community that I just
- 11:41launched to help you level up with AI.
- 11:42It has all of my free resources, and I
- 11:45think we're building an absolutely
- 11:45amazing community there. So, if you want
- 11:47the cheat sheet for today's video, which
- 11:48includes the setup guide for you to set
- 11:50up skills, set up a memory system, and
- 11:52everything I've talked about that you
- 11:53can actually just put into your own
- 11:54Claude to make the information from this
- 11:57video applicable, then check out the
- 11:59free school down below and come be a
- 12:00part of the AI Edge community. All
- 12:02right, so level three is extremely
- 12:04important. As I said, this is how you
- 12:05genuinely start automating your life and
- 12:07business, and that is Claude skills.
- 12:09Because a memory system means Claude
- 12:11knows you, but skills learn how you
- 12:13actually complete specific tasks. I
- 12:15believe this is the highest leverage
- 12:17feature in the product alongside
- 12:19connectors, and almost nobody uses them
- 12:21properly. So, right now I'm going to
- 12:22quickly show you how to set up a Claude
- 12:24skill and how to use a Claude skill.
- 12:26What you need to know is that a skill is
- 12:27essentially a folder with instructions
- 12:29in it. So, whenever you complete a task,
- 12:31Claude knows exactly the steps and the
- 12:34preferences that you have in order to
- 12:36complete that task in the most efficient
- 12:38manner. So, you know, if I want to
- 12:40script a video like this, every single
- 12:42time I work on a video, I use a Claude
- 12:44skill, which learns what I like because
- 12:47I keep giving Claude feedback, and it
- 12:49updates the skill every single time. So,
- 12:51I can have an entire script spoken like
- 12:53me, formatted like I would, and
- 12:54presented in the way that I prefer. And
- 12:56every single time I do a script, I give
- 12:58it feedback on how it did, so it gets
- 12:59better and better and better. Or you can
- 13:00have a copywriting if you're trying to
- 13:02get sales leads or if you're trying to
- 13:03write posts for X, whatever you're
- 13:04trying to do, where you can train it to
- 13:05speak like you. Obviously, it's easier
- 13:07for me cuz I'm creating content, so we
- 13:09can just take the transcripts, and every
- 13:10time I film a video like this, we can
- 13:12feed it to Claude so it learns my real
- 13:14intonation, my real vocabulary choices.
- 13:17So, copywriting for a content creator is
- 13:18easier, but you can mimic this just by
- 13:20asking Claude to ask you some questions,
- 13:22and you can just speak like you normally
- 13:23would, and it can learn how you speak.
- 13:25And you can improve it over time. And
- 13:26the possibilities go way bigger than
- 13:28that. You could build a Claude skill to
- 13:30scan your competitors. You can build an
- 13:31ideation skill. You can build a skill
- 13:33which hunts properties for you, finds
- 13:35opportunities in the e-commerce drop
- 13:37shipping space. Anything you were doing,
- 13:39you can automate with a skill. And now
- 13:41I'm going to show you how. So, what you
- 13:42want to do is you want to go {forward
- 13:43slash} skill creator. This is basically
- 13:45going to set up the creator that's going
- 13:47to ask you questions in order to train a
- 13:49skill. This is best used when you can
- 13:51either verbally through voice
- 13:53transcription explain how you like to do
- 13:55something. Or if you want to feed it the
- 13:56data which it'll ask you for, like
- 13:58documents, PDFs, files, MD, whatever, to
- 14:00actually train the skill. But something
- 14:01that is very cool that Claude recently
- 14:04launched which made skill creation even
- 14:06easier is this feature right here. If
- 14:09you click on the plus button next to the
- 14:11chat, you can actually now record your
- 14:14screen. So, it'll look at what you're
- 14:15doing in real time. And if you
- 14:17transcribe what you're doing, it'll also
- 14:18be listening to your microphone. And you
- 14:20can run it through an entire process.
- 14:22So, what I did is I ran it through my
- 14:24entire competitor scanning process where
- 14:26I sit down, I look at competitor
- 14:28channels, I look at what performs well,
- 14:30and I train the skill to automatically
- 14:31look at what makes it bad video a good
- 14:33video. I talked it through the entire
- 14:34process. And now, whenever I want to
- 14:36scan YouTube, it searches the search
- 14:38term that I'm interested in, for
- 14:39example, Claude. It looks at all my
- 14:41competitors. It analyzes the data on a
- 14:44reoccurring basis. And then it comes
- 14:45back to me with proven vetted content
- 14:48ideas. And I trained it just through
- 14:50this screen recording. Or for example,
- 14:51if you're in a job where you have to do
- 14:52data entry, you can do the data entry
- 14:54like let's say you work at a finance
- 14:56firm or an accounting firm. You can do
- 14:57the data entry and you can explain why
- 14:59you're doing what you're doing. Because
- 15:01maybe Claude isn't the best if you just
- 15:02give it a spreadsheet because it might
- 15:04not fit the criteria that your company
- 15:06is asking for or it might not fit the
- 15:08exact work flow you're trying to
- 15:09achieve. But if you've walked it through
- 15:11and you've spoken while showing it what
- 15:13you're doing, it'll pick up on those
- 15:14little nuances. So, this is kind of
- 15:16crazy, but something I would recommend
- 15:17that you actually do is whenever you're
- 15:19working on your computer, try and record
- 15:21as much as possible. Like every time
- 15:23you're doing a task, try and record it
- 15:25and explain to Claude what you're doing.
- 15:26Because now it doesn't only just hear,
- 15:28it also sees. And this is a massive
- 15:29opportunity for you to be able to
- 15:31literally automate your work. And as
- 15:33your work evolves, you can keep doing
- 15:34more recordings to update the skill over
- 15:36time. Now, anytime I want to do certain
- 15:38things like analyze competitors, like
- 15:40writing, like video scripting, now I can
- 15:42literally just use my skills and I can
- 15:44have it done exactly how I want it done.
- 15:46This saves so much time. Yes, there's a
- 15:47bit of upfront investment, but it's an
- 15:49absolute game-changer. If you want to
- 15:50get the most out of Claude, instead of
- 15:52what most people are doing, which is
- 15:53just opening new chats with no context.
- 15:55This point in the video is going to get
- 15:56even better with level four and level
- 15:57five, by the way. You've just stacked a
- 15:59memory system on top of skills to
- 16:02automate work, but now let's get into
- 16:03the really fun stuff. All right, so
- 16:05everything so far still required a bit
- 16:07of manual work. Even skills, technically
- 16:09you need to sit down and prompt a skill.
- 16:11What I'm about to show you next is the
- 16:13best way to use Claude in 2026, and that
- 16:16is loops and scheduling. I'll focus on
- 16:19loops first. A loop is essentially
- 16:21Claude doing a task on repeat without
- 16:24you triggering it based on a preset of
- 16:27variables and instructions. There are
- 16:28two commands that I love using in
- 16:30Claude. The first one is forward slash
- 16:32goal. This tells Claude exactly what you
- 16:34are trying to achieve. For example, I
- 16:36want to build a personal habit tracking
- 16:38dashboard with the following
- 16:40constraints. And then I'm going to use
- 16:42forward slash loop. Based on my entries
- 16:45every day and my behavior, automatically
- 16:47adjust the dashboard to better optimize
- 16:49my productivity. That's one example of a
- 16:51loop. An even better example is for
- 16:52content. The example I showed you
- 16:54before, you can actually use a Claude
- 16:56skill with a goal and a loop. Let me
- 16:58show you. Goal, come up with the top
- 17:00five video ideas based on what is
- 17:02currently performing well on YouTube.
- 17:04Loop, every 24 hours, you need to
- 17:06specify the duration, look through the
- 17:08top ranked videos with the following
- 17:10search terms, and update the list on a
- 17:12spreadsheet ordered by your highest
- 17:15conviction ideas. Use my video research
- 17:17skill to follow the exact same mental
- 17:19framework that I trained you on. So now
- 17:21it's using a skill which is trained on a
- 17:23loop, so telling Claude how often, what
- 17:26variable, what interval it needs to
- 17:28repeat that task, and with the goal in
- 17:30mind, so it knows what the end goal is.
- 17:32The way the new Mythos class models
- 17:34work, and GPT also works with this as
- 17:37well, is they are so smart now that they
- 17:39are able to self-correct their own work.
- 17:42So even if you're vibe coding an
- 17:43application, running loops to do audits
- 17:46is a massive game-changer because it can
- 17:48actually vibe code for you on a loop
- 17:50automatically. And if you want to
- 17:51schedule a task, you can go into the
- 17:53scheduling function of Claude, and you
- 17:55know, have a set time when a task is
- 17:57scheduled. For example, every single
- 17:58morning I have a research report which
- 18:00is sent to my inbox based on what
- 18:03happened in the market overnight,
- 18:05connected to the FMP connector from
- 18:06before, that actually has my context on
- 18:08my own portfolio. So every morning I see
- 18:10exactly what I need to know about my own
- 18:12portfolio. That is done on a scheduled
- 18:14cadence. You don't need to run a loop
- 18:17alongside a schedule. So you could just
- 18:18schedule a task on its own, but for
- 18:20tasks with shorter intervals, I prefer
- 18:22to run loops. And for tasks that are
- 18:24more like daily or weekly, I prefer to
- 18:26run schedules. But forward slash goal, I
- 18:28use nonetheless to actually tell Claude
- 18:30what I am trying to achieve. And now
- 18:32let's move on to the final level, the
- 18:34master level, and something that
- 18:35honestly in my business took a long time
- 18:37to crack, but I feel like we're finally
- 18:38getting there and it's so exciting. This
- 18:40is the creation of agentic workflows.
- 18:43Now technically in level four, a loop is
- 18:45[music] an agentic workflow because it
- 18:47is going to spawn agents on your behalf.
- 18:49But to level up, you can create a system
- 18:51which orchestrates, especially for
- 18:53complex tasks, automatic behavior and
- 18:56automatic model selection. For example,
- 18:58you can have Claude Fable, which is like
- 19:00still the smartest model on Claude, but
- 19:02it could change. They're always updating
- 19:03new models. But right now I use Claude
- 19:05Fable as the orchestrator, and then I
- 19:07have Opus 5, and then I have Claude
- 19:09Haiku and Sonnet based on how complex
- 19:11the subtask is to complete work so I can
- 19:14actually reduce token cost. So
- 19:16essentially have different models doing
- 19:17different jobs. And what you want to
- 19:18tell Claude Fable is to spawn sub-agents
- 19:21to complete a certain task. You combine
- 19:23that with the scheduling feature that I
- 19:25talked about before or the looping
- 19:26feature and you can actually have tasks
- 19:28running automatically. I did a video on
- 19:30this recently but something that's
- 19:31really been helping me in my work is
- 19:33graph engineering. So basically breaking
- 19:35down a process, for example, a video
- 19:37production, you know, all the way from
- 19:39ideation to publishing the video, I will
- 19:41write down each step of the process and
- 19:44I will visualize in a mind map graph who
- 19:47and what model needs to be involved at
- 19:49every step of the process. I'll also put
- 19:51in where a human needs to be involved.
- 19:52Like I'm the bottleneck a couple times
- 19:54in the process but I can visualize which
- 19:56agents are used at different points in
- 19:57the process. And if you create this for
- 19:59every single task and create a skill for
- 20:01it, Claude will be able to orchestrate
- 20:03your agentic workflow on your behalf.
- 20:05This is the ultimate way to do things
- 20:07and I've done a full video on this. So
- 20:08if you want to check out my graph
- 20:10engineering video, I'll link it below.
- 20:11And if you want the full setup guide
- 20:12from today's video, it's available down
- 20:14below for free. Come and be a part of
- 20:16the community and you'll get access to
- 20:17it. You can put it straight into Claude
- 20:19and you will automatically be able to
- 20:20apply all of the knowledge I gave you
- 20:22today to your real workflow and set up
- 20:24your own memory system from scratch.
- 20:26Hopefully today I've helped you level up
- 20:28your knowledge of Claude. These are the
- 20:30things that have really moved the needle
- 20:31for me but of course these models are
- 20:33only getting better and better so make
- 20:34sure you subscribe to the channel for
- 20:36more content like this and when I learn
- 20:37things I'll share it with you to make
- 20:38sure we're at the cutting edge. See you
- 20:40in the next video. Peace out.
About this transcript
This page contains the full transcript of 5,000+ Hours of Using Claude in 20 Minutes (Beginner to Pro) by AI Edge, generated from the public captions YouTube serves with the video. The transcript has 4,666 words across 663 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.