14:02Zack Kirk · Talking Head
Top 5 AI Agents Selling for $20,000 Right Now
A 14-minute sales playbook disguised as a tutorial — five boring agents, two tiers, specific pricing, and the one thing everyone in AI gets wrong.
June 19thThe argument in one line.
AI gives ordinary people scalable reasoning but never true creativity, so the leverage lives in the human deciding what to build and structuring the context, not in the tool doing the work.
Read if. Skip if.
READ IF YOU ARE…
- A solo founder or small-business owner who wants to automate real internal workflows with Claude instead of buying yet another SaaS tool.
- A builder who has played with AI but keeps hitting a wall where outputs feel average and wants a repeatable context-engineering system.
- Someone who wants a concrete folder structure and CLAUDE.md pattern for running multi-session Claude projects without re-explaining everything each time.
- A creator or agency operator deciding which parts of a workflow to hand to AI and which parts still need human taste.
SKIP IF…
- You want a plug-and-play, hands-off 'fully automated' pipeline; the whole thesis is that a human must stay in the loop and do the thinking.
- You are looking for deep model internals or how to train your own LLM; this is explicitly a driver's course, not an engineering one.
- You need copy-paste code more than a mental model; most of the value is principles and workflow, taught on a whiteboard.
The full version, fast.
The course argues that AI is autocomplete-on-steroids: a guessing machine that is excellent at anything closer to math and weak at anything closer to art, so your job is to give it the exact context it needs, no more and no less. It reframes every app, plugin, and automation as a system (inputs to a process to an output) and teaches that you only add AI leverage after a system already works manually. The build half hands over a concrete setup: three Claude instances with distinct roles (main builder, research VA, and a fresh test-dummy), an Obsidian folder template with archive, journal-logs, and a step-by-step build-out folder, and a CLAUDE.md that interviews you and scaffolds the project. You then climb Level 1 to 3, from a one-shot YouTube-research automation to a live paid app, always designing the output by hand first and letting Claude execute.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:18
01 · Intro
Promises a three-part course: principles, execution template, and real running automations across ten-plus businesses.
01:18 – 22:00
02 · Principles of AI
AI as autocomplete on steroids and a guessing machine; the math-to-art slider; break macro tasks into subtasks; be in the loop; AI reasons but is not creative.
22:00 – 52:48
03 · Principles of Systems & Automations
Everything is a system (inputs, process, output, feedback); the three forms of leverage (labor, copies, money); build the output manually first, then add AI leverage.
52:48 – 1:31:41
04 · How to Build Systems With AI
Why AI leverage is weird; five real uses (understand, research, reason, reasoning-in-systems, VA commands); context clues vs instructions; folder structuring as the core skill.
1:31:41 – 1:53:48
05 · KJ's Automation Building Process
The template: Obsidian/IDE setup, three Claude agents (main builder, research VA, test dummy), the folder/context structure, and the juicy CLAUDE.md.
1:53:48 – 2:08:50
06 · Building Level 1 Automations
The simplest path: open the template folder in Claude, answer a few questions, and get a working one-shot YouTube-research automation.
2:08:50 – 2:22:50
07 · Building Level 2 Automations
Duplicate the template, open it in Obsidian, run a research agent alongside the builder, fill out prebuild context by hand, and produce a far better YouTube-research output.
2:22:50 – 2:33:36
08 · Building Level 3 Automations
Dialed-in automations, plugins, and apps; designing core value-prop logic and shipping to real users.
2:33:36 – 2:38:34
09 · Leveraging Your Automations
Turning a working automation into a plugin others can install, then a hosted app with UI.
2:38:34 – 2:44:40
10 · Real Example: AI Finance Plugin
Walks through the finance dashboard plugin, designed by hand first over months, with three skills and hand-read source files.
2:44:40 – 2:46:50
11 · Real Example: 1-1 Client Build
A custom-home-builder takeoff-and-estimate tool: blueprints plus formulas and prices into an artifact that saves hours per week.
2:46:50 – 2:52:08
12 · Real Example: SGNL App (Advanced)
His first live paid app; how he designs each page as its own folder/system, iterates HTML versions, and hands off to a developer via a written handoff.
2:52:08 – 2:53:54
13 · Final Thoughts
This is a skill that takes hours; he didn't touch Claude six months ago; make mistakes fast, build build build, and share the course.
Lines worth screenshotting.
- AI is autocomplete on steroids: it guesses the next token, feeds the whole batch back in, and guesses again, so how you start a prompt steers everything after it.
- The skill is giving AI the exact input it needs for the exact output you want, no more and no less, because too much context wastes tokens and dilutes the answer.
- Rate AI on a slider from math to art: the closer a task is to math the better it does, the closer to art the worse, and creative work is where it stays permanently average.
- Never rate a macro task like 'is AI good at thumbnails'; break it into subtasks and rate each one, because it's great at some steps and useless at others.
- Everything is a system: inputs run through a process inside an environment, with feedback used to improve it, and every process is just smaller systems chained together.
- The three forms of leverage are labor, copies, and money, and AI is a strange hybrid of labor and copies because it reasons like a worker but scales like a copy.
- Only automate a system after it already produces the output you want manually; there is no point scaling a system that doesn't work.
- A useful build heuristic: only make something if it would be worth the upfront cost even if no one else ever uses it, so ideas you'd actually use survive.
- Process worshippers brag that their whole pipeline is automated while the actual output is bad; output-focused builders ship boring processes that produce great results.
- The real leverage of AI is the creative application of average reasoning at scale, the same edge McDonald's got by making one simple system anyone can run.
- Context is just files and folders; structuring a 250-page PDF into a labeled 'testimonials' folder gave Claude two pages instead of 250 and fixed bad outputs.
- Run three Claude instances with separate jobs: a main builder, a cheap research/VA agent, and a fresh test-dummy that only sees what a real user would see.
- Keep the main builder's context clean; start a fresh chat every completed subsystem or around 50 to 75 percent of the context window and hand off via a written note.
- Journal-log what failed and what worked each day so a new Claude can read the last three days and resume without you re-prompting the whole project.
- You are the alpha: the reason a build works is that a human designed the output by hand first, and Claude just implemented it.
Design the output by hand, then let Claude build it
WHAT TO LEARN
AI scales reasoning but never supplies taste or vision, so the durable skill is deciding what to build and structuring the context, while Claude only executes.
02Principles of AI
- Treat AI as a guessing machine: it predicts the next token from what came before, so start prompts precisely and give the exact context needed, no more and no less.
- Before asking whether AI can do a task, break it into subtasks and rate each on a math-to-art slider; hand it the math-like steps and keep the art-like ones yourself.
03Principles of Systems & Automations
- Model every automation as a system of inputs, a process, an output, and feedback, and remember the process is just smaller systems chained together.
- Reach a working output manually first; only add AI leverage once the system already produces value, because scaling something that doesn't work just scales slop.
- Filter build ideas with one question: would this be worth the upfront cost even if no one else ever used it, which keeps you on cheap, fast, genuinely useful projects.
04How to Build Systems With AI
- Structure context as labeled files and folders; a testimonials folder that hands Claude two pages instead of a 250-page PDF is the difference between good and average output.
05KJ's Automation Building Process
- Split work across three Claude roles, a main builder, a cheap research VA, and a fresh test-dummy, so the builder's context stays clean and you verify like a real user.
- Keep daily journal-logs of what failed and worked and write handoff prompts near context limits, so a new chat resumes the project without you re-explaining it.
07Building Level 2 Automations
- Climb the levels deliberately: a one-shot automation, then a multi-Claude Obsidian build, then a dialed-in plugin or hosted app, reusing the same template each time.
13Final Thoughts
- Accept that this is a learned skill measured by outputs, not by how automated the process looks; build fast, make mistakes early, and let human taste be your edge.
Terms worth knowing.
- Context engineering
- The practice of deciding exactly what files, examples, and instructions to give an AI model, and how to structure them, so it produces the output you want without being fed irrelevant material.
- CLAUDE.md
- A markdown instructions file Claude reads at the start of a project. In this course it interviews the user about their project and scaffolds the folder structure automatically.
- Skill
- A reusable set of text instructions that tells Claude how to perform a specific task a particular way; skills now live inside plugins but are fundamentally just files.
- Plugin
- A packaged bundle of context and instructions someone can drop into their own Claude to run the creator's solution on their own account.
- Leverage
- Getting more output per unit of input. The course buckets it into three forms: labor (other people's effort), copies (reusable systems or software), and money (buying more of the first two).
- Level 1 / 2 / 3 automation
- The course's difficulty ladder: Level 1 is a quick one-shot automation, Level 2 is a more complex multi-Claude build in Obsidian, Level 3 is a dialed-in, vital automation, plugin, or live app.
- Handoff prompt
- A summary Claude writes for itself (or a teammate) when a chat runs low on context, capturing the state of the work so a fresh chat can continue without losing progress.
- Claudian
- An Obsidian plugin that runs Claude in a side panel inside the Obsidian vault, used here for the research/VA agent.
Things they pointed at.
1:39:10toolClaudian (Obsidian plugin to run Claude in a side panel)
18:00productSignal / SGNL app (his first paid app, used as a running example)
Lines you could clip.
04:22
“The ability to give AI the exact inputs it needs to get the exact outputs you want. No more, no less.”
12:41
“The closer the slider is to math, the better AI will be. The closer it gets to art, the worse AI is going to be.”
1:11:40
“The real leverage with AI is the unaverage application of average reasoning.”
56:40
“Don't be cool. Be useful. Produce value.”
1:25:00
“Everybody will be able to code, but very few will make successful apps.”
1:40:00
“You are the edge. You are the alpha. You've got to think about what you want. Claude just builds it.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
00:00Welcome to the most in-depth AI automations course on YouTube. By the end of this video, you're going to be extremely confident in implementing AI into literally any business model.
00:11And I can only say that because I've spent the last five months using the exact templates I'm giving you today to automate systems in 10 plus US businesses. So, yeah, this will not be a light one, but it is legit. In part one, we're going to cover the principles of AI, systems, and then building systems with AI so that you finally understand how this technology works on a fundamental level and you can actually build valuable stuff with it.
00:35And then in part two, I'm gonna give you my exact automations project template. I'm gonna give you my workflow setup, my agent setup, and my building process, all of which took like five months to create. So that you can start building automations with this stuff today.
00:48And there's gonna be three different levels to the automation. So beginner, intermediate, and advanced.
00:53That way you can learn from start to finish how to build these complex automations that create insane value. And then finally, in part three, I'm going to break down multiple real running automations that I've built across my businesses and 10 plus other businesses.
01:06You can see how this stuff creates value right now as we speak. So I cannot wait. I have never been in this excited for a video that I've made.
01:14So lock in, grab some coffee, and let's get right into it. Welcome my friends to part one of this course, the principles module.
01:22Now if you really just wanna skip ahead and go get the tactics and all of my processes and frameworks, go ahead. It's in part two, execution, but I highly recommend you watch this.
01:33This is my secret. This is really the sauce. This is what I do not think you will find anywhere else on YouTube at least at this moment.
01:40I hope this will be the most valuable video you watch this month, maybe even this year. That would be amazing. I pray for that.
01:46But let's jump right into it. What is AI? How do we understand it so that we can actually use it?
01:52The cool part about AI is it's like hiring a brand new employee. So to get the most out of an employee, you have to understand them, their strength and weaknesses and how they work. But here's the difference.
02:05When you understand how, let's say, Claude works, you don't just unlock the leverage of one employee. You unlock the leverage of thousands, infinite Claude's.
02:14So if you don't understand Claude, none of them are that useful. But as soon as you do understand AI, in this case Claude, you unlock thousands of usefulness points from him and pennies on the dollar.
02:26Like, I don't I don't think we realize us, normal. Right?
02:30Average people, not kings, not powerful, famous, whatever. We've never had this much leverage available to us ever in the history of humanity, but you have to seek to understand.
02:43One of my favorite verses is this. The beginning of wisdom is this. Get wisdom.
02:48Though it costs you all you have, get understanding. And so I don't just wanna make shiny kind of cool looking stuff that doesn't work for anyone. I wanna understand this technology.
02:59I wanna get it so that I can leverage it and create amazing things in this world. And, obviously, these verses are talking even deeper about the wisdom of the Lord, the fear of the Lord. So you ever if you have if you haven't read the Bible yet, highly recommend that you do.
03:14And that is the true wisdom is fear of the Lord. But it applies in this scenario as well. Don't just follow the shiny videos.
03:20Understand the technology. Don't be lazy, and you'll be crazy how much everything changes.
03:26But that being said, do not worry. I do not care about the deep, deep technicals. I don't want to build my own AI model.
03:34I wanna be the driver. I want to be the Max Verstappen, the f one driver, the fastest man in the world who can use these machines. I don't care about building them.
03:42I care about using them. Um, so I wanna learn just enough to drive like crazy. Nothing more.
03:48I don't care about building engine. I want to drive that sucker. But in order to drive, you do need to understand the machine that you're driving.
03:56So what does it mean to actually drive AI? What is being an f one elite driver when it comes to AI? And I think it's as simple as this.
04:05The ability to give AI the exact inputs it needs to get the exact outputs you want.
04:12No more, no less. We'll talk about this throughout this entire course, but this is the goal. So get used to me referring to this a lot.
04:19Alright. What is AI? How does it even work?
04:23Uh, disclaimer, I'm ignorant on many things, and this is a very simplified version. But like I said, it helps me drive, which is why I'm sharing it with you guys. And the way I see AI is like autocomplete on steroids.
04:36It is not a magic genie, although it does feel like it sometimes. It is a guessing machine. It is getting very good at guessing.
04:43It's getting better and better. But essentially what happens is you give it words which have meaning in real life.
04:50It guesses about what you mean, and then it guesses about what you want, and then it tries to give you that output. It's guess.
04:58And this can be in the form of words, code, commands, etcetera. But if you say, want a dope website, it's going to guess what you mean by dope, what you mean by website, and it's going to try to give you a dope website.
05:11And so the better these models get, um, the better the outputs are going to be. But your job is this, the context.
05:20And so if you tell AI, like, I did it, it has no clue what you're talking about. If you say, I cooked it, it starts to understand a little bit more.
05:28If you say, I finally cooked it, or I finally cooked pasta, or the final full context, I finally cooked pasta for my late grandmother's secret pasta recipe.
05:38I got it right. My family loved it, and they all teared up around the dinner table. That has meaning.
05:44That has full context. It's technically the same as I did it, but now we all know what you're talking about and the full meaning behind it.
05:53And that's the skill of AI, helping it guess better. Now how it actually guesses is actually really cool.
06:01And if you understand this, you'll it'll make so much more sense with AI. And there's a video called large language models explained briefly. It is highly technical.
06:11I don't really care that much about this, but if you want to, go watch it. I've watched it a few times. But what happens is you give it a sentence.
06:17Right? Paris is a city in blank. Based on the model that they've trained it on, like trillions and trillions of data, it gives it probabilities for what the right answer it thinks is.
06:29And so it's going to say France. And so this is how AI works. It says Paris is a city in France.
06:35It takes that guess. Then it feeds all of that back into AI, and it guesses again. Paris is a city in France and.
06:43And then it takes all that and feeds it back into AI. And so the way it spits out every word is by reputting in the entire batch of words.
06:54That's why our goal is so important. We do we wanna give AI just enough context, but not too much.
07:02That's how we waste tokens and we dilute our outputs. So, yeah, AI just does this. It guesses each word, and then it plugs it all back in and guesses the next one.
07:10And it's getting really good at it, is why it feels magical. But here's a super dumbed down example of what we're doing. This is how I understand it.
07:19Let's say your friend says, I'm going home to walk by blank. Well, what do you reason he's going to walk?
07:26Based on your training data, your life, you're gonna think probably dog. Right? You're gonna say, you know what?
07:31He's going home to walk his dog. Now let's say that he says, I'm going home to blank.
07:37You don't know what he's going home to do. There's not enough context, and so you're like, you know what? I have no clue.
07:43But he's going home. I know that much. Now what if he gives you too much context?
07:48I'm going to the pet store. I got a cat leash, lizard food, and some dog stuff. I love my pets.
07:52I try to take care of them. I'm going home to walk my blank. And you're like, dude, I don't know.
07:58Is he walking his lizard, his cat, his dog? Who knows? But this guy's kinda weird.
08:02Right? This is how most people treat AI. They feed it their entire, like, life story to try to make a website, and it's just way too much context.
08:12Um, okay. Yeah. That wasn't so bad, was it?
08:15Right? The deep AI stuff is over even though I know that was not deep at all. This is what I care about.
08:21I wanna know how does what I just kind of showed you change what we do. I'm not in the business of making a bunch of technical crap. I'm in the business of providing value.
08:30How do us as business owners, as entrepreneurs, use this stuff? What does this short engine lesson change how we drive AI?
08:39That's what I wanna do. I wanna change your behavior so you get better outputs. Well, now you should understand that this is why our goal is so important.
08:47Give the exact input, the sweet spot for what you wanna get out. And this is a skill. It takes time.
08:52That's what I wanna teach you. Number two, knowing where you want to go, knowing the output you want, like, it's not a binary thing.
09:02Right? It's the clearer you get. So the clearer you are in exactly what you want and then how you start is extremely important.
09:10Just like how autocomplete works, the first words you type direct it down what it's going to give you the results.
09:17So if I say what is the name, it's gonna give you names because you start off by asking what is the name. Basically, how you start is so important because it's autocomplete on steroids.
09:29And so if you do not know where you're going and you don't get a good start, you're gonna get bad results for the AI.
09:36But if you have exact clarity on what you want, if you have examples of the website you want, if you're clear on everything you want included, the better chance you're gonna get to go there. But if you just say, hey, AI.
09:48Give me a dope website. It's gonna give you the same slop it gives everybody else. So the better output clarity, the better start you give it, the better starting prompt, the better it's gonna do.
09:57And we're gonna go over how to do all of that in tacticals, but these are just the principles. Number three, probably the most important takeaway here. AI is built to give you the most probable answer.
10:09It is built to give you the average answer, which means it is amazing at things that have the right answer.
10:19If you say what is four times four, there's only one right answer. And so AI is amazing at this. But on the flip side, it is awful at things where there is no one right answer.
10:30And so if you were to ask a lot like, hey. Does this look good? If you ask me this, I'd be like, heck the freak.
10:35No, bro. I would not hang this on my wall if you gave me a million dollars. But this painting sold for a 137 mil, by the way.
10:43So to some people, this is a dope painting. But my definition of a of a dope painting is not this, which is why it's really hard to do kind of subjective stuff with Claude is because we all have different meanings of the words that we're giving Claude.
10:58And based on how he was trained, he has different definitions of what a dope website looks like, and it might mismatch from your definition of a dope website. And so what does this translate to?
11:09Well, AI is incredible at pattern recognition, at research, at summarization, at searching, converting or repurposing existing kind of subjective stuff, sorting big volume, big data, repeating tasks accurately, rough drafts of creative tasks.
11:26It's great at that. And basic coding. Not the problem solving or not the, like, creative side of coding, but kinda just like the the objective side.
11:35Really, anything that you can easily explain to it what right looks like and what wrong looks like with very little room for discrepancy, it is good at. But it is awful at things like creative writing.
11:46Really anything super creative. Anything meant to be experienced by humans.
11:54Anything that needs to be fresh or new, it's bad at because it's literally giving you the average of everything in the past. So it's hard for it to be new. So music, art, content, writing, etcetera.
12:08Although, it can trick you into thinking that it's good because it's putting out the amalgamation of past creative work in the past. So how do you actually, like, think about this in the right way when it comes to automations and and plug ins?
12:23Well, it's this. When you ask yourself, will AI be good at blank? You want to imagine a slider.
12:30So the closer the slider is to math, the better AI will be. The closer it gets to art, the worse AI is going to be.
12:39And here's the secret sauce. You don't wanna rate the macro tasks. So if you were to ask me, KJ, is AI good at YouTube thumbnails?
12:48I can't answer that as far as the macro because there's so many different sliders for each subtask.
12:56And so here's the real here guys, this if you get this, man, it's balling. Break down the big task into smaller task and then rate each step. And, also, I'm gonna show you guys how AI will do this for you, which is crazy in the in part two.
13:12But if you were to ask me, KJ, is is AI good at creating YouTube thumbnails? Here's what I would tell you. For for finding outlier thumbnails and videos around topic, yes.
13:21For analyzing and comparing which elements are best for thumbnails? Yes. For making a bunch of mock ups?
13:26Yes. For taking the actual pictures? No.
13:30For composing elements? No. For finishing touches?
13:33No. These are all things that really influence the taste of the thumbnails. And so it's good for half of the thumbnail process, but not good for the other half.
13:43And that's where you can really get crazy results with AI by understanding what it's good at and what you should do. So like I said, I'm gonna show you guys, like, literally in the setup I'm giving you for this entire, like, course.
13:56I have Claude rate what it thinks it's good at and what it thinks you should focus on. So I'm literally gonna help you guys do this with the Claude MDM I give you later. But this is so important.
14:06Feel free to watch that back if you need to get this. This is the leverage with Clot is by getting that. Now number four.
14:14Because of number three, because the fact that AI is built on averages, it is not inherently creative.
14:22Caveat, it can do some creative domain things, just not great.
14:29So if everyone's using what AI considers to be a dope website, then it's average. It's no longer creative.
14:37And so AI is creative if if you were the only person using AI, but then everyone uses it, and it's no longer cool. It's like trends.
14:46It's kinda like, uh, in school when all the cool kids would, like, do some sort of thing, and then the, like, not cool kids like me would catch on and start wearing, you know, the silly bands or wearing, like, a certain certain type of socks or something.
14:59And then all the cool kids quit doing it because the uncool kids like me started doing it. That is happening exponentially fast with AI because it's just copying people.
15:08It's not really creative. Now this this is so important.
15:13Please lock in for this. By learning how to prompt AI, you can get better outcomes that you can get more alpha.
15:23So there's extreme alpha, meaning like edge or value, from learning how to context AI well. So if you just say build me a dope website, you will get bang on mediocre average.
15:36But if you get better and better at contexting, better at driving the machine, then you can get better results that are further on the curve to being, like, better.
15:47Right? Just imagine there's one side of this, and this is like a super amazing, you know, valuable website.
15:53By doing better context, you can get better results up to a certain point. Do I ever think AI will be better than, like, the world's best website designer?
16:04No. No. Never.
16:06But you can get pretty good. You can get better than most people with AI. Um, and my favorite quote is this.
16:11I came up with it. I don't know. How can AI truly think outside the box when it is the dang box?
16:18Um, everything that we do as humans is an experience. Like, this is why people would drink beer.
16:23Right? It's not because it's, like, freaking hydrating. It's because it makes you feel a type of way.
16:28We go to movies because we want to feel like the experience. And so since AI really is just math, it's really bad at creating things that make us feel stuff.
16:39That is the human element, the experience. And so an old tweet of mine is AI has has about as much taste as a white rapper. Now it's just a joke, guys.
16:49If you're a white rapper, don't get offended. But, um, it's a really good analogy because what do most white rappers do? They try to say as many words as possible, pronounce every word right, and it just lacks a lot of taste versus other rappers that have a more culture than us, um, white people.
17:08They're able to make songs that you'd like, sometimes you don't even know what they're saying, but they they feel epic. And so that's the problem with AI is it lacks taste.
17:19But you can help it with context, get a little bit better taste. Just know that it maxes out.
17:26And so one thing that I feel like I run into a lot is I hit a wall with AI where it's like it gets way better, and then all of a sudden, I just cannot get it to go past this certain level of, like, good. That's because I feel like I'm hitting the wall where I just need to implement human taste. So cool.
17:42This is what this course is gonna teach you to do, by the way. It's gonna teach you to go from here being, like, bang on average with AI to getting, like, really good results that honestly, 99% of people can't get even with, um, AI. So here is the reminder.
17:57The true value of AI does not lie in what it can do necessarily, but it lies in what it can take off your plate so that you can do the stuff that really matters. The value is in the opportunity cost.
18:11So I can spend all day every day, really, like, fifty plus hours on this Miro board you're watching. Not automated at all. I did all of it myself because this is what matters.
18:21This is the taste. This is why you're watching this video right now instead of some other AI slop is because, man, this k d dude is freaking weird. He's funny sometimes because he's so stupid, and his mural boards are pretty entertaining even though they're kind of lame.
18:36This is the power of AI. It does all this other stuff in my life so that I can make these silly mural boards. And guess what?
18:42It's kind of working, guys. Like, this video actually has over 300 k views now, so love you guys. You guys are amazing.
18:49Alright. Number five. You have to be in the loop.
18:53So AI is getting really dang good at guessing, but it's still guessing. So what happens is if you're not in the loop, when you give AI an input, it spits out an output, and then that output becomes AI's next input, and then that becomes another output, which becomes the next input.
19:11AI is going to loop itself into slop oblivion. This is why I do not like the, like, slash goal or slash loops or whatever. AI will get better over time, but right now you have to be in the loop to to fix the trajectory.
19:27So you might start off in the right direction, but AI can make a bunch of slop and then get way off way off track. And the more complex your systems are, the more this is gonna happen.
19:38And so do you have to use this for like, you know, one shotting some silly little thing? No. If you're building real systems or applications that people are gonna pay money for, you gotta be in the loop.
19:48You have to. And these companies that are saying that they're not in there and the these AI agents are creating everything are are just honestly lying. So there has to be a human in the loop.
19:59And when you do that, AI becomes insanely magical. Number six, AI lacks what matters most.
20:06It does not have God given agency. Despite the name agent, it does not have agency.
20:12It does not have God given creativity. Now dispute me if you want. Right?
20:17You believe whatever you want. I believe that God created us in his image. God is the ultimate creator.
20:23When I look at the trees, when I look at creation, I am in awe of how good our God is. The fact that we're breathing right now is a gift, and he has given us that gift to create inside of us.
20:35That's why people are able to create rockets to Mars and create amazing paintings and and videos and, uh, Oceans 11 movies, know. God given creativity. And AI does not have that.
20:48It just copies past examples of that. It is a watered down version of the beautiful state of of humanity.
20:58And so it will never be able to replicate you. Keep in mind that. Now if you don't put in the work and you don't do anything useful with your life, then like, yeah, AI is gonna replace you.
21:07But AI can never replace these mural boards, and I'll prove that in the future because AI is not me. AI is a blob of all previous humans' data, and when you mix all colors together, what do you get? Kinda like just dark brown or black.
21:23Um, but you were completely opposite. Right? You are uniquely one of one.
21:26And so, uh, yeah, don't ever forget that. Do not be discouraged. AI cannot replace you.
21:32Um, so, yeah, I hope that was encouraging. Alright. Now what?
21:35We just finished the first principal module. Things are about to get a little bit deeper after this. I'm not gonna lie.
21:40This was kind of just a little teaser. Now we're going to head into the principles of actually automations, which I cannot wait to show you guys. This is like literally this is my secret sauce.
21:50Um, but all of this is to once again turn you into the point zero zero one percent driver. So let's go over to principles of automation. I will see you guys over there.
21:59Let's get it. Alright. Welcome to the second principle module, principles of systems and automations.
22:05Now apps, plugins, skills, automations, all these shiny things you hear about, they're all just systems.
22:13And when you learn how to think in terms of systems and leverage almost anything becomes possible. Elon is a great example. That dude lives and breathes systems and leverage and, uh, yeah, he's a trillionaire.
22:24So let's let's get a little closer to that. Why don't we? And we become real entrepreneurs when we stop thinking in terms of tactics and identities like, oh, I'm a YouTuber.
22:34I'm a copywriter. I'm a growth operator. And instead, we start thinking in problems in systems and leverage.
22:40And that's what I wanna share with you today. You wanna make money? We do that by creating value.
22:44We create value by solving problems. You create more value by solving problems easier, faster, better, cheaper. How do we do that?
22:52By building leverage systems, plug ins, automation, software.
22:57Man, we know how to do this. Everything changes. And hope you see that on my channel.
23:00My videos have been blowing up all because I stopped thinking like a tactic hunter, and I started thinking, how do I solve problems with the leverage of AI. I'm gonna give you guys my secrets today. So let's get into it.
23:13Principle number one, systems. A system is just this. It's inputs that get processed into an output.
23:22And this is all happening inside of an environment, and we have feedback from running the system to improve the system. And so it's really that, um, simple.
23:31The best way to look at it, in my opinion, is examples. So you've had let's say we're we're gonna cook some lasagnas. The output we want is an incredible tasting lasagna.
23:41That is the desired output that we're trying to get to. That is what we're building the system for. We do that by taking ingredients.
23:49We process them by, like, mixing them together, cooking them, doing whatever. This happens inside of a kitchen and, uh, after we follow the process, we get an output.
23:59How do we get a better output? Well, we do feedback. Right?
24:02We taste the lasagna and we say, needs more salt or it's a little undercooked. And we can improve the output by changing the inputs, so getting better ingredients or changing the process.
24:14Right? How we process them. And this all happens inside of the kitchen we're working in.
24:19And so I used to watch a show with my wife called, like, I think it was Next Level Chef. And if they, like, took an l, they had to cook in this disgusting basement.
24:29And if they won, they got to cook in the state of the art kitchen. And so that shows how the environment can really play a factor on your output.
24:39What the heck does it have to do with anything, KJ? Why are you talking about lasagna? Well, this is how I think about life.
24:45This video you're watching right now, if you're watching it on YouTube or school, this is how I made this video is I have a system for creating these things. I have a desired output that I want to put out.
24:56I have inputs that I put into the system. So what is the thing that I'm giving?
25:01What is the problem I'm solving? Where is the existing market demand? These are inputs.
25:07And I take those inputs and I come up with a title, thumbnail, intro, body segments. I record this using a certain tool, you know, editing, tags, upload, and then my feedback are what you guys think.
25:20Right? The views, the likes, the comments, the, uh, CTR, average view duration, CPM.
25:25And based on those metrics, I make changes. Right? Add this to the title, less saturation, talk slower.
25:31Y'all got y'all know I talk like a crackhead. I'm sorry. I'm working on it.
25:35But, um, yeah. This is a system.
25:38And same with my finance plugin, which has, you know, generated me tons of amazing leads and helps a lot of people. This is happening in the in the environment of Claude, so it happens in Claude. So how Claude works determines how I make the plugin.
25:51And the output is a fully customizable private finance dashboard plugin.
25:57And so it adds these these kinds of things of value. Right? Saves people money.
26:02That's awesome. And so I built a system to do that. Right?
26:05We input the user transactions. It processes them using the plug in. So this is the system I built to be the process.
26:12And, uh, yeah, the iterations are me improving the system based on your guys' comments. Pretty cool. Now let's get a little bit deeper.
26:20I told you guys I'm a give you guys the real stuff, but it's going to be kinda deep. So my first ever application that I'm still building called Signal, and we have real paying users on this. Shout out to the person who bought the yearly plan yesterday.
26:32You're amazing. Um, but these users have a problem that requires human depth and perspective.
26:39Therefore, they want a YouTube video. So they don't just wanna ask AI. They wanna see what a human has created around this problem.
26:47A good example is how to speak. So I'm learning how to speak better for you guys to make these videos better, and that is something that is hard for AI to explain because, well, it can't speak. And so I wanted to solve this problem for myself of not knowing how to find the best videos.
27:05And so I give AI this input. Right? So context on my problem.
27:10The output I want is the best videos to solve this problem and the process is signal. The main you know, the application is it it does context to query reasoning.
27:21It API calls, um, the correct parameters. Anyway, I won't get into the details.
27:25But, basically, I made this system to give me an output, and now people pay me to use this system. Um, and, obviously, this is just one feature inside of signal, but this is the main core value prop.
27:38And you guys are gonna learn how to do this today. So, um, this course that you actually, I decided to get this for free, so no one even no one bought this. The course that you're watching is my system for creating systems, which is kinda meta, but that's why this course is, I think, so valuable.
27:55And it is this. You are the input. This course is the process.
28:00So by you watching this course and doing taking the action steps, this is all happening on Earth, by the way, laws of physics, you are able to then go make valuable systems that you can sell, make tons of money, change your life or business, whatever.
28:12And the feedback is what you guys tell me in the comments, in the emails, in the school. I read everything. So thank you guys for your feedback.
28:19Um, so couple useful notes on systems while we're here. The process section in a system is usually just a bunch of smaller systems chained together.
28:30And so think about Tesla, for example. The input are raw materials.
28:34The outputs are finished Tesla cars. But inside of that macro system, there's a lot of small systems for how they make the wheel wells and just all this other stuff. Another example is I have a thumbnail system inside of my YouTube system.
28:48So how I make thumbnails is a system inside of the bigger system of how I make YouTube videos.
28:55Number two, learning systems thinking is mandatory in the AI era. This is the only way computers thinks in and out.
29:04And so the better you understand systems, the better you will be with AI. Three, as powerful as systems thinking is, it will not magically give you creativity, passion, or grit required to make a system actually work?
29:21If so, every single software engineer would be a billion dollar founder, but they aren't. And so that's kind of the thing with AI is now that we can all be, like, you know, mediocre junior developers, writing the code isn't the value really.
29:37It's knowing what to code. That's always been what determines who makes a billion dollars. And so this is important, but you still have to have the judgment.
29:46Number four, desired output, desired output, desired output. We're gonna go in detail on this in a second, but do not cope with a cool process. Do not cope with a cool assembly line.
29:59What do people care about? They care about the car. The output is what matters.
30:03Okay? And sadly, this is 99% of the AI space right now. Um, the secret is, like, don't don't be cool.
30:10Be useful. Produce value. That's my goal at least for my life.
30:14So principle number two, leverage. Getting more output per input. This is, uh, entrepreneurship one zero one.
30:22If you don't know this one, man, you are missing out and this is going to be hopefully life changing for some of you guys. So by adding different forms of leverage into your systems, you're able to get more for less.
30:35This is why everyone's freaking out about AI. It's not because the clankers are cool, it's because of the leverage that it gives us.
30:43So instead of you having to spend ten days building a website, you can send off two prompts. So this much input, insane output.
30:54That's why AI is so important. But in order to get the most of AI, you have to understand what leverage is. And this is how we make big money.
31:01Right? We talked about this in the last module. So for probably the first time ever, because sadly our school system really fails us as entrepreneurs.
31:08Right? I want to cover different forms of leverage and what they look like incorporated into systems. Then we'll talk about AI leverage.
31:17And then we're going to incorporate them in part two of this, uh, entire course execution. So, yeah, it's gonna get juicy. These are the three forms of leverage that I kind of bucket leverage into, which helps me a lot.
31:30The first is labor. So this is using someone else's energy towards your goals. This is the oldest form of leverage.
31:37Right? Back in the day when the Egyptians built the Pyramids, they had tons of, you know, slaves building them. Oh, lord.
31:42What is this? I quit this. Um, that is labor leverage.
31:47Other human beings working for your goals, your agency, working towards your outputs. And the way that I like to think about the leverage gained is energy times skill times other forms of leverage.
32:01And so if you just have an employee who's just, like, moving atoms around, right, like moving these around your room, not that much skill. But if you have an employee who's incredibly skilled at, say, like, you know, I don't know, editing, something like shout shout out to Caleb who's editing this video.
32:17If he edits a video, he might take him an hour. A bad editor would edit the same video, and they would get less views. So same input, wildly different outputs for Caleb because he has skill.
32:29And this is why people hire skilled employees. And then if those skilled employees have other forms of leverage, well, then you're getting a lot of leverage out of hiring this person. Right?
32:39So if you hire someone who's incredibly good with AI, like so also shameless plug, we will build AI systems for you if you don't want to do this yourself. Shout out to my agency seven eighty.
32:49Book a call with us. We'll build custom software for you. If you hire us, you get the leverage because we are using other forms of leverage.
32:57So what are those other forms? Well, the second form is copies, which sounds lame, but this is the the secret.
33:04This is the sauce. So the biggest leverage we get as as humans is copies of previous systems or solutions.
33:13And so entertainment. Right? That is a solution to boredom for humans.
33:19Information is copies of different instructions for humans of how to think and act. Code is instructions for computers of how to act.
33:28AI is like code two point o. So it's code that can reason. So it's kinda like a like a a just a weird form of, like, labor leverage too because it can solve problems that it wasn't specifically coded for, which is crazy.
33:43But you accessing these models is just you accessing a copy of Claude. So someone built this model, and now we all benefit as long as we pay the Claude overlords $20 a month. Right?
33:56And last form is tools and machines. So these multiply physical effort.
34:01So because someone came up with the invention of a tractor, we can all benefit from having tractors long as we pay the money, of course. Once again, that's the whole leverage.
34:10Right? Now the cool part is digital copies leverage. So the upfront input cost of creating these copies is massive.
34:21So writing a book, me recording this video is gonna take, you know, fifty hours to make and then more to record it. A lot of input. But the cost of distribution is a fraction.
34:33So this video you're watching right now could be seen by a billion people without me doing any extra work. So if YouTube was like, man, KJ's cooking, dude. Like, let's send this to everybody on Earth.
34:45I do no extra work, and a billion people get this information. Versus back in the day, if I had to teach you one on one, I'd have to give this two hour long talk to every single person.
34:58It is so unleveraged, but that's the power of copies. And so the person that invented the tractor unlocked a ton of leverage.
35:05He can sell the tractor. Me making this video, uh, Kendrick recording Good Kid Mad City. Right?
35:11He'd record the album once, but now they can copy it and everybody can benefit from listening to that epic album. So, yeah, writing the book, writing the software, training the AI model, this is all digital copies. So the effort, um, an input goes like this.
35:28So to go from zero, so to go from nothing to having a thing that works requires all the effort. That's literally it. And then distributing it requires so little effort in comparison.
35:40Um, this is what this is like the zero to one effect. And if you wanna see this kind of shown in a a more typical way, this is it.
35:47Right? This explains exponential growth. It's because to get something that works, to build a system that gets the right output, that is the hard part.
35:59Distributing it, scaling it up, that is when it gets a lot easier. It's still not easy, but it is a lot easier in comparison.
36:08And else this is also why people so so many people fail with AI is because I will show you guys how to use AI to build the system, but AI is kind of most useful in copying what you've already built and getting more of the thing that you've already designed or built, whatever.
36:26The third is money. So money just allows you to buy future instances of one and two.
36:33So with money, you can hire people. With money, you can buy more copies, you know, whatever. Um, or you can lend to people for them to buy more one and two.
36:43And so if you use capital, if you use money to buy a stock, well, you are benefiting from someone else buying these things.
36:52So when you buy into the SpaceX IPO, you are giving Elon more money to buy these things and to make more of these things.
37:00And so you benefit from it if he succeeds. This is how the world works. I hope you guys are like, is this clicking?
37:07This is what no one's taught you. Seriously, this is it.
37:11This is getting past all the hype and looking at, like, how do billionaires become billionaires? This is it, guys. Now just knowing this alone is not gonna make you a billionaire, clearly, or else I have more money.
37:23But this is the start. This is how you play the game. I'm explaining the rules to the game so that we can play it.
37:29Now here's a goofy example that really helped me get leverage. This is Chick fil A. If you're not from America, man, I'm sorry for you because America is the greatest country on earth.
37:41But Chick fil A is one of the best food that we have in in America. And what's funny is my friend Luca is from London.
37:47He's actually staying with us right now. We've been taking him to Chick fil A almost every day because how much he loves his chicken bugger. Back on track, they use leverage in every single form.
38:00So this system to create this chicken sandwich is a they're copying a system. The recipe, the inputs, the type of chicken, how long they cook it, what ingredients.
38:12That is a system that they've copied across the country. They how they build the stores is a system. The application.
38:18Right? Code, software, copies. The labor, how they train the labor using information copies.
38:27It's all leverage. And because they have all these forms of leverage, guess what?
38:32They make $22,000,000,000 with a b selling fried chicken. But it's because they spent seventy years nailing one simple output better than anyone else, a good chicken sandwich at scale.
38:47So, yeah, they have scaled up like crazy, and now they're doing 22 billions of dollars a year with chicken. So every anytime you have Chick fil A, remember this example and think about all the leverage. I actually used to have a friend who worked on the Chick fil A software, but he worked on the software that they had in in the store for their, like, assembly line of chicken.
39:06Like, that's how freaking leverage Chick fil A is. They're crushing it. So it's freaking crazy.
39:12How does this all tie back to AI? Well, there has never in the history of humanity been this much leverage available to the average bloke like me and you.
39:22So because of AI, we can now achieve massive leverage like people who always had all the power before.
39:30We can do this with AI. Alright. But how?
39:34How do we actually leverage AI in a system? So let's first talk about how do we apply stuff.
39:42Now quick warning here. Uh, this is this is a serious one.
39:46I'm about to contradict a lot of the stuff you see online, and I'm I'm not saying I'm right. I I could be wrong, but this is how I view the world. This is what is this is what has served me.
39:55And, uh, I'm gonna piss off probably some people because I'm gonna kinda call out some some YouTubers and teachers, but I do not wanna do this and like I'm better than them. It's more of like I feel like you guys are being lied to, and I would I wanna show you guys the truth as far as I know it. Um, I could be wrong, but just just a disclaimer here.
40:14Um, now I'm talking about this through the lens of value creation. So if you care more about just building stuff for fun, right, like you're just having a good time, disregard what I'm saying.
40:26Okay? This is only through the lens of solving problems. If you're just, like, having a good time playing around with AI for fun, keep doing it, man.
40:33I'm not here to shame you on that. That is how you learn. But we're gonna talk about value creation.
40:38Uh, one of my favorite quotes from Elon is they asked him what a a young person should try to do, and he's like, try to be as useful as possible, and that's what I wanna do.
40:47So step zero in creating a system is deciding what system to build in the first place. And this is quite honestly the hardest step. If this step was easy, every single software engineer would be a billionaire, but they aren't because they can't figure out what to build.
41:03And so Elon says the most common error of a smart engineer is to optimize a thing that should not exist. Same with YouTubers. You can replace this with any single title.
41:15Yeah. Here's the evidence. Right?
41:17Around 90% of all startup to tech startups ultimately fail. Crazy. Here's the problem.
41:25Once again, guys, I am a sucky entrepreneur. I don't have a million dollars, so I can't even speak on this really myself, but I'm trying to learn. But I have one heuristic that has served me really, really well this year, which has allowed me to do these amazing things.
41:39Um, so 500 k views in three months in a brand new niche, first live app with real paying users and it actually works, and a community with on on 20,000 people now since I made this Miro. And I know that these aren't great indicators of, like, actually value creative problem solved, but I can't really show you guys the the DMs and emails I get.
41:58But, hopefully, you're one of the people who I have created value for, so this is proof in itself. Anyways, this is my heuristic.
42:06This is my threshold for evaluating whether or not I should build a system, and that is this. Is it valuable enough to me that even if no one else buys it or uses it, it was still worth the input to create it?
42:20And so this is it. If the value it adds to my life is greater than the upfront cost of creating things, so not just money, but, like, time, then it's good.
42:32And so, basically, is the zero to one cost worth it even if it stays at one? So if it doesn't go from one to a thousand, is it still worth it?
42:40And that helps me focus. Um, this helps me verify the idea from a few different angles. So first off, it helps me do ideas that have very low upfront time and capital.
42:54So it's cheap and fast to go from zero to one because guess what, guys? I don't have a million dollars, so I don't have time and money to invest in ideas that may or may not work. And then second, only ideas that I would actually use survive.
43:08And what's cool is the bigger the problem is in my life, the more capital and time I can allocate to solve it. Because the more value it adds to me, well, the higher cost it can be and it still be worth it as far as my heuristic goes.
43:24Next. Even if I have an idea that I think is dope, because it's built for me, if I don't use it after a week, then I know that it's trash and I just don't have to build it any further.
43:34And lastly, I only build it if I think the skill and insight gained would be worth it as well. So, obviously, building these courses doesn't really do much for me besides, you know, cementing cementing my, like, knowledge, I guess.
43:49But I learned the skill even greater by doing these courses, which is why the only courses that I make are things that I want to learn the skill in. And what's funny about this heuristic is when you only make stuff that you would actually use, other people use it too.
44:03So my videos get views and my app gets paying users because I only made it for me, uh, which is kind of kind of funny. So that's the first step in deciding what system to build.
44:14Now step one, once you know what to build, we have to get extreme clarity on the output of the system. And so you have to get very detailed.
44:24What do you want to see? And then also, what is the deeper why that you even want that output?
44:30And so, um, we'll go about this in the the execution. Like, uh, you know, I'll show you guys how to get AI to interview you to answer these questions. But just remember, the market cares about output, not your process.
44:44What comes out of the machine? That is what matters. I'd rather have a boring process and an incredible output than I would than I would have an incredible process, boring output, which this module alone should be proof of this.
44:59So far we've not we've not touched Claude once. I've not built a single vibe coated slab website, but I care about changing the way you think.
45:06I care about this video might be boring, but the outcome that you get from watching it should hopefully be incredible. And that's what I wanna do.
45:15Um, there's some more stuff about this if you want. But, basically, like, avoid these videos about Notion, Obsidian, like AI operating system setups because they're only building it to get views.
45:26They don't actually use these things. And if they do, the outputs that come out of them are typically not great because they're just trying to build a cool process, not the outcomes.
45:37Um, and fundamentally, a teacher can only really teach what they have done themselves.
45:43This is the problem I have at business college. Most of my teachers have never owned a business. Not to say that they can't teach good stuff, but I I wanna listen to people who have done what I what I wanna do.
45:55Right? So step two, you achieve the desired outcome by any means possible.
46:02I know this is going to piss off some people. What I mean by this is if you want to be a good YouTuber, do not use AI to make your YouTube videos.
46:12If you cannot get a video that gets a lot of views without AI, the chance of you do getting a video with views with AI is extremely low.
46:22There's two types of people who are using AI, and the first type is a process worshiper. So they say, hey.
46:28Look. My YouTube channel is a 100% automated by AI. It's crazy.
46:33It's insane. Editors are dead. YouTube has changed forever.
46:37And the video that they actually make in their video gets like no views. It's all about the process, how cool the machines are. But those who are output focused, the process sucks.
46:50It's boring. Guys, if you if you watch me make these mural boards, man, it is boring. It takes fifty plus hours, but it gets an amazing value.
46:59And once again, I can't show you the actual value because views are not value, guys. You you could slap a grandmother and get a million views and not change anybody's life, but it is hard for me to quantitate life change.
47:10So, hopefully, you personally have experienced value through my creations, and that enough is proof alone. And once again, I am not the end all be all guys. I suck at this.
47:18I'm just trying to explain how I view the world. Do not be process blind. This is what I mean by this.
47:24Most people look at the the, like, assembly line. They're like, wow. AI did the whole video.
47:30That is sick. But the actual outcome is not cool versus a really boring process, but a great outcome, and people are like, dude, gross.
47:40Get with the times, man. You don't have automated thumbnails. This is what process bindless looks like.
47:46Please don't fall for it. Another good example. I absolutely love Nate, but I feel like his videos can be misleading.
47:53I don't think he's doing it on purpose. But he is saying, look how great AI editing is, guys.
47:59I'm getting 200 k video views on this video. But here's the thing. He's getting those views because it's about it's in an AI niche.
48:08If you were to go offer this editing, like if you were to use this editing style and offer it to people, you would not get paid anything for it. So it is the output is only cool because AI had made it, but it's not good editing.
48:22And so if you were to use this outside of the AI niche, it's going to flop for you. So anyways, I really pray that this makes sense. We want to care about the outcome, not the process.
48:35What is actually happening? Long story short, there is no point in automating, in using AI inside of a system that doesn't get desired output.
48:46There's no point in scaling a system that doesn't work. And so the highest leverage thing that you can do is by putting in maximum effort, so doing things manually until you actually get the outcome you want and then using AI.
49:01So I've designed over, like, 2,000 thumbnails in my day. I used to be my my full time job. So the only reason I know how to use AI in my thumbnails is because I know how to make good thumbnails without it.
49:12And so that is the kind of lie I want to free you guys from. You have to get the output first. Um, you can use AI to help you learn, but you have to actually do the work.
49:22And so phase zero is basically this system doesn't work. Phase one is, hey. This system worked, but, man, it takes a lot of work from me.
49:31Um, and then phase two, which we'll talk about in a second, is, uh, or phase three is actually decreasing your input. So that's step three.
49:39Once the system works, that's when you add leverage. So you make the system work, get the output you want, write the email, write the marketing thing, um, make the video, build the application, whatever.
49:53Get the output, and then you add an AI leverage. Um, and so phase one, system doesn't work.
50:00Phase two, hey. Look. It works, but we have to do a lot of effort.
50:04This is where you add in all of the leverage. So now the system works, and I don't have to do much.
50:11Here's a simple example. I used to sell a high ticket program, and I used to sell the program myself on the calls.
50:18And so I learned how to get people to close into my program, so pay me money for my program. And then I turned what I learned into a script that I hired salespeople to run for me.
50:32So I used to have to take every call and I would make money, but then I would take a fraction of the calls and I would still make the same money because I added leverage of labor and I average I added leverage of copies, meaning like my system for taking sales calls for my offer.
50:48Really, really cool. And so, yeah, guys. Um, I know that was dense.
50:53I hope that you're still watching this video, but hopefully this is helpful. I really pray that it is. Um, but now we can actually go do this.
51:00We're gonna spend the rest of this course. So in part two execution, we're going to actually do this.
51:06Um, the next course is going to be, like, adding more, like, AI leverage, like the next principal module, but the entire part two is going to be doing. Don't worry. We're gonna be doing this.
51:16So, yeah, phase four is adding more leverage. We won't talk about this much, but, basically, you just bring money in and you, uh, scale. So in recent news, SpaceX just IPO ed, and they're hitting phase four like crazy.
51:30So now that they figured out how to launch rockets that actually get to space, they are saying, hey. Give us a lot of money, and we'll launch a bunch of rockets because we'll buy more people, and we'll buy more copies, and, uh, yeah, we'll make more rockets.
51:43Yeah, that is phase four scaling, which is really cool. This is why as a software developer, as a founder, once you hit that one, the VC gang piles in.
51:54So once you get a working system, everybody wants to invest in your stuff because they wanna see that they wanna see it scale. Um, Yeah. So pretty cool.
52:03So how do we leverage AI leverage? I know this was a lot, but hopefully I if if this was helpful, please, you know, leave a comment or send me an email.
52:11So I'll make more of this if it is helpful. If this is just boring, I'll go back to doing shiny stuff, but just let me know. Now what?
52:19Well, we're going to get into building systems with AI. So like we just talked about, the real leverage in adding AI into a system is phase three where you're adding in leverage after it already works.
52:31But the cool part about AI is that it can reason, and so it can help you get to phase two. It can help you build the system, but it's a different way of thinking about it.
52:41And so that's what we're gonna talk about in the next module. So I will see you guys over there. Uh, let's get to it.
52:47Alright. Welcome to the third principle module, how to build systems with AI. This is what I've been waiting to get to.
52:54We had to cover that other stuff first, but man, this is going to be epic because AI leverage is weird and it's hard to explain which is why you've probably never heard it explained before.
53:08And I am not some philosopher genius. Right? But I'd be talking to the clankers.
53:13Trust me, bro. Eight to ten hours a day for months, I'd talking to the bots. And all I wanna share is my perspective that I have found truly useful, meaning has created insane value in my life and business.
53:26So, yeah, let's get into it. Why why is AI leverage so weird? Well, number one, it can reason.
53:33And so it's it kinda feels like labor leverage a little bit. Like, it can think a little bit, but it's still just math.
53:41And so some stuff, it royally drops the ball on, and you're like, what in the world? How did you not get that right? So it's like labor leverage, but it's technically copies leverage because someone built this model and you're, like, copying this, like, fake employee.
53:58And so, basically, if you treat it like a human, you're overselling it. It doesn't have perspective, and it it lacks the human touch.
54:06But if you treat it like just a computer, you're underselling it because it can reason. And so a good example of why this thing is not because all you guys probably like, well, it's gonna become, you know, sentient and all this.
54:19I don't believe that crap. Why? Because look at this.
54:21Look at this meme. Would AI have ever decided to put this meme right after what I just said? No.
54:29But I'm a human. I'm from Alabama. I, uh, my granddad dips.
54:33I have perspective. I have taste. Therefore, I wanted to hit the hit you with this meme to prove that, man, stuff is weird.
54:40And that's probably why you're watching this video. If you're still here, it's because I'm not going through a stupid AI slop slide deck. I'm giving you some, like, some raw crap.
54:51But KJ won't need to transcend. It will be creative. It'll have perspective.
54:54We're all screwed. I don't think so because every model is going to make stuff like that model.
55:01So if we're all using Claude and we all ask it to do a task, it doesn't matter how much context we give it, it's gonna do the task like Opus 4.8. Right?
55:09And as humans, we like new stuff.
55:14Imagine your favorite song ever. What if you had to listen to it every hour on the hour for the rest of your life? You would probably hate that song.
55:24And that is because we like new stuff. We like unexpected, and AI is expected, which is why it's very hard for it to create incredibly tasteful things.
55:35Now can it do some stuff pretty well? Absolutely. But when it comes to really good experience things that humans are gonna use like this video, it's very bad at.
55:45Um, and remember, value is in comparison. AI slop will always be AI slop. So although this looks really good or it used to look good, it became what?
55:56Sloppish. But now this is the new AI slop even though it looks incredible comparatively. Right?
56:02And soon this will be awful and there'll be something new. It's just like why there's there's trends in fashion. Right?
56:08It's why we look at people from the sixties and we're like, oh, dude. What were they wearing? It's because it's comparison.
56:14Things change. And so, um, this is why models are always going to be kind of average because we all have access to the same model.
56:21Once again, values in comparison. If you had the iPhone one in 2007, you were a freaking legend.
56:28Right? Let me tell you. I was in middle school and people that had this were, like, instantly goaded.
56:32But if you have this phone, twenty three six, right, you're gonna be shamed. It's the same phone. And so the comparison, what's happening in society is more important than the actual, like, thing.
56:43So that's why AI is tough. AI will never replace actual human creativity. Right?
56:49Debate me on this. I believe it from the Bible. Right?
56:52So God created mankind in his own image. The image of God, he created them. Male and female, he created them.
56:58We are created to create. Look at God's beautiful creation. Um, we get to create in his image, and that's why I don't think AI will ever fully replace just that raw spark of life.
57:10I don't know. Um, the good news is it will replace everything else so that we can do more creative stuff. This is the good news.
57:18People are like, oh, AI is gonna take jobs. Bro, so did the calculator. Did you know they used to have entire buildings of people who would be, like, one calculator.
57:27Right? They'd all crunch these numbers. Thank god.
57:30No one has to do that anymore because of the calculator. AI will do the same thing. It will get rid of some really boring stuff so we can do more amazing free will stuff.
57:39Alright, KJ. What the frick do we use AI for then if it's not good at this stuff? Well, I'll explain to you.
57:45Um, the first is understanding, explaining things to you. And so as I was making this module in my office, um, my editor Caleb and best friend was editing something.
57:56He was building this insane model, but he was asking, um, Claude questions on the technical tools.
58:04So Claude is helping him learn, but he is still the creator. Um, as I was typing this, that's what he was doing.
58:11Same with you building systems. Um, two, researching. AI can help you find the best tools, the best copies of the latest information tools that you can use to build your systems.
58:24And so an example is I was working with one of my one zero one clients, and we're going to build this this product for him so that it'll save him $250 a month. And I asked AI to go look at each one of these features and tell me which tools should we use for this.
58:41Dude, normally, this would have taken me, like, all day. It did this in five minutes. This is the leverage.
58:47It's so it's so good. Alright. Cool.
58:51Number three is reasoning. So AI can think, and it's really good in poking holes in your plan.
58:58It it helps you explore every options. The only caveat here is do this after you've already thought of stuff.
59:05This is it. Right? Um, so it's like labor leverage of skill.
59:09So it's like having a thinking partner, it is so good for. Um, and you'll see in the build template later, but imagine imagine hiring Claude as a thinking partner.
59:18He's still an employee. You're still the visionary. You're still the CEO, but he can think with you and help you come up with the better outcomes, which is epic.
59:27And four, this is where things get pretty fun. This is using reasoning in your systems.
59:34The first three were all about building the system. This is about it being inside the system. And so using basic reasoning at scale in your systems for your users to then use.
59:47So this is kinda like copies leverage, um, plus artificial skill leverage. It is awesome.
59:53Good example is the single app. And so input, user has a problem that they want to see videos.
1:00:00On how to solve. They explain that to Claude or to Signal, which uses Claude. That's the input.
1:00:06Then I have a skill, which I've programmed Claude to do, and he thinks about what queries he should run based on the instructions. And so I won't give you guys all that because that is literally, like, the core of my app. That is the money.
1:00:20And this is the output. So normally, if I was, to hire my mom and say, every time someone has a problem, take that problem and come up with 10 search terms.
1:00:30She could do it. Right? She could totally do that.
1:00:33But she could service one user probably every 20. AI gives us basic reasoning at insane scale. I know this is a little bit tough to get, but this is the the crazy leverage.
1:00:48So one of my quotes, and we'll go over this in a second, but the real leverage of AI is the on average application or no. The the creative application of average reasoning.
1:00:59And this is kind of a good example of that. Use case number five, basic reasoning and commands.
1:01:05This is like a a virtual I'm such an idiot guy. Virtual VA.
1:01:11A virtual virtual assistant. Oh my gosh. So, basically, doing literally anything that is closer to objective, so anything math related so you can spend more time doing the subjective.
1:01:22So once again, writing code. Right?
1:01:25Not the vision or the deep problem solving, but just the executing, translating your words into code on the screen. Summarizing, so reading stuff, summarizing it, uh, very average kind of creative stuff, so, like, writing an email, a YouTube title.
1:01:39It's good at that small task. Pattern and research analysis, running commuter commands, all this good stuff.
1:01:46So an example is this is not loading, but this is one of the skills that I actually we'll we'll probably build this plugin in this course at some point. AI does all the research for my videos that I so I I can come up with thumbnails and titles. It comes up with the keywords.
1:02:00It does all the research for, uh, the titles. It ranks the titles. It has literally given me the titles that have given me, like, my last two viral videos.
1:02:09Epic. That's using all these things mentioned above to save me a lot of time. This is a project for one of our clients, and it is a pricing estimate for he builds custom houses.
1:02:21He inserts the the, uh, blueprints. We then go through and add in these custom, like, variables that he wanted in here, and then it gives him the takeoff and estimate. This has to take them, like, a week manually.
1:02:34But with AI, it can be done in a matter of literally minutes, which is crazy. But honestly, anything that you can think of that you would hire an assistant or a junior developer to do, AI can do.
1:02:45It is, um, absolutely awesome. So the real leverage with AI is the unaverage application of average reasoning.
1:02:54But if we're honest, this is how it's always been. Right?
1:02:57Think about McDonald's. They can hire high schoolers because the process of making a burger is not that hard.
1:03:04Someone made this system so simple and so easy, but they made it they created it in a creative way that gets a lot of value. I know this is trippy because me and you have never had this amount of leverage before, but this is something that you really need to think about.
1:03:21And, uh, the first time I ever used Claude, these were my thoughts. You can pause it and read if you want, but, uh, I thought that was pretty cool and a lot of things have have come out to be true. But a couple of things I wanna talk about is people will feel so good about doing something that they will cope for doing the right things.
1:03:38So AI makes it so easy to make a website, to make some emails, to do whatever. It gets outputs that you otherwise would have never gotten, But the problem is they're not the right outputs.
1:03:49They don't actually work. You don't actually get clients. You don't actually get views, and that's the issue.
1:03:55Um, everybody will be able to code, but very few will make successful apps.
1:04:00People will spend 95 percent of their time asking how to build something and 5% asking what they should build. This is bad. It should be the opposite, but I guess that's how it's always been.
1:04:10So, anyways, there's some more, like, takes in there if you wanna check those out. Yeah. What it really did is AI unlocked scalable reasoning, but it did not unlock true creativity.
1:04:22And a lot of times when people approach me with for help with AI stuff, they're not getting results because they think it is fully creative. It is not, but it can reason.
1:04:32And so if you give it the right guidelines, which you'll learn how to do, it can do things for you, but you have got to think and create for yourself. And so the alpha is in the creative application of the scalable reasoning.
1:04:47How can you creatively use a very, you know, 200 IQ person with not a whole lot of street smarts? How can you use that at scale, um, to make value?
1:04:57But to be honest, this is how it's always been. Right? Steve Jobs, this was him when when he first made Apple.
1:05:03Whenever they fired him, they almost went bankrupt. They brought him back, and he made everything go crazy. The reason I'm showing you guys this is because they had the same engineers.
1:05:12So the same engineering talent. Right? Real humans, not just AI, but, like, real creative humans were not able to create value because they lacked the vision in the agency.
1:05:23When Jobs is gone, things fell apart. When he came back, things went exponential because it's all about how or it's all about what you choose to use your your reasoning on.
1:05:34That's what I'm trying to say. How you actually use their reasoning. So, anyways, here is my optimistic prediction.
1:05:40AI will take over everything black and white, um, so we can work on stuff in full color. I know that sounds extremely cliche, but AI will take over doing all my research for these video thumbnails so that I can do what?
1:05:52Make a cool thumbnail. Like, that's the cool part about AI. And the problem that you run into with AI is you're trying to get it to do everything, and it can't.
1:06:02If you could tell AI, make me a million dollars. Make no mistakes. You'd have a million dollars, but it cannot.
1:06:08You have to have that perspective and that agency. So, KJ, how do we do this well?
1:06:14Like, how do we use AI like a non Goober? Well, I'll remind you, this is the goal.
1:06:18Right? Figure out what to give AI so that you can get out what you want.
1:06:23So you have to figure out what you want, then figure out what to give it, figure out what instructions to tell it, and you'll give you get what you want. So to accomplish this, we need do three things well. The first is context structuring.
1:06:35So how do we structure what we give Claude? The second is workspace setup. How do we make it easy for you to work with Claude?
1:06:44And three, how do we understand giving Claude different roles? So the brain of the the model, we can give him different kind of agent setups.
1:06:54So he can have different roles that can allow us to get bigger, juicier outputs. Number one, context structuring. What is context?
1:07:04Well, it's just files and folders. So it's files, the actual stuff you give it, and then it's folders, how you structure and group those files.
1:07:14That is context. I like to break context into two different types. This helps me a lot.
1:07:20So the first is context clues. We covered this in the first module. Basically, you're explaining the situation in more detail to Claude so that he knows what you mean.
1:07:30So if I say build me a dope website, he doesn't really know what dope means. But if I say build me a website for this thing, it needs to look like this.
1:07:39Here's the parameters. Here's examples. Now he knows what dope means to you.
1:07:45He can get a better outcome. So And this example is what we talked about earlier. Right?
1:07:48Giving just the right context. The second is instructions.
1:07:53These are this this is a specific direction on how Cloud should do something the way you want it. This is also referred to as skills which are now inside of plugins.
1:08:03But guys, it's all just text instructions. It's all just files. That's all that it is.
1:08:09And so here was my old skill for my second channel that I used to post daily is I have a certain way I like Claude to do my YouTube titles. And so if I tell Claude, just base Claude, hey, give me a YouTube title.
1:08:22It's going to give me the most average cookie cutter title it can think of. But if I say give me a YouTube title and I give it some instructions for how it should create them, now I'm going to get a result that's closer to what I want.
1:08:35And so you're either giving him more context or you're giving him instructions, which are all just once again context, but I like to break it down and it makes things clearer. And so here's an example from my finance plugin.
1:08:47If you guys have used this from my last video, I gave Claude instructions for how it should create your finance dashboard when you run my skills.
1:08:58And so there are so many ways to create a finance dashboard, but I figured out the way that I think is very valuable. I gave Claude instructions, and I gave that to you so that Claude can use his reasoning and my instructions to build you something epic.
1:09:12So, yeah, that is the kind of two different ways I break context into. Now context and folder structuring.
1:09:21This is how you achieve the goal. So if we want to give Claude exactly what he needs, no more, no less, we have to learn how to structure context and folders.
1:09:33And so this is why folder structuring is so hot right now. There's a meme. It's because this is the key to unlocking efficiency and outputs with Claude.
1:09:42And so a good example of this is I used to have an email system that I ran for one of my marketing clients, but I still did marketing full time. And I used to give Claude this 250 page PDF every time I went to write an email.
1:09:56And I kept wondering why does he not get the testimonials right? He's not doing the testimonials good at all. And then I realized, woah.
1:10:04I can break this 250 pages into different folders, different segments, and I made a testimonial folder.
1:10:12And so whenever Claude needed a testimonial, he didn't have 250 pages. He had two pages, and he knew where to find them because the folder was labeled testimonials.
1:10:21That is a simple way to get better outputs by structuring. So imagine it like this. I want you to imagine that you hire an assistant, and you gave them a thousand page PDF with random words scrambled all across the pages.
1:10:34So not like there was any coherent, like, thoughts to this PDF. It's just a bunch of random segments of random words with random, like, meaning. That's level zero.
1:10:44Level one. Now I want you to imagine you separate the PDF into, um, different pages and files that are titled what they contain.
1:10:52So it's titled like, hey. Here's a research paper on this. Hey.
1:10:56Here's some past examples of what I did here. Right? It's still one file, but it's at least, you know, separated a bit.
1:11:04Now I want you to imagine that you separate those files into folders, and so you group these things in folders. So now if Claude or your assistant needs something, he doesn't have to look through a massive stack of papers.
1:11:18He goes and looks at the folder. Right? Now level three.
1:11:23I want you to imagine that you have organized the text into files. You've organized the files into folders. You've organized folders into different filing cabinets.
1:11:32They're all separated. They're all labeled in ways that are focused on the output you want, and then you give your assistant specific instructions on where everything is.
1:11:42This is how you save tokens. This is how you get good outcomes. This is what is happening with Claude behind the scenes.
1:11:49Now as these AI models get better and better, you have to do less and less of this, but the skill of knowing how to structure context is one that will serve you, I think, for the rest of time with AI. So you should get as good as possible and making it as easy as possible for AI to do what it wants, uh, what you want it to do.
1:12:10So this is how you become powerful with AI. And I'm gonna give you my templates for this. Don't worry.
1:12:14I'm gonna give you, like, my filing cabinets that I've spent months creating, but you have to know why I'm doing this and and why you should learn it. So do not let it fool you.
1:12:23This is a skill. I'm not gonna lie. This is really, really tough, and there are levels to it.
1:12:28I'm not anywhere close to the peak. But if you just understand what I just showed you, you will be in the top 1%, maybe 01%. Because most people are just giving their entire life to Claude in some crazy, like, carpet y wiki and saying, you know, fix my life.
1:12:43And it's just it's so Claude is suffering. He's so confused. So, yeah, you're about to skip straight to level three just by watching this course, but there are so many levels to this at doing this well.
1:12:54Right? So, uh, I don't even know what I was talking about this.
1:12:59Yeah. Uh, I'm gonna give you my dialed context structure setup in this course. You're gonna skip to level three.
1:13:04You're also gonna get my clod, which will basically set up this entire setup for you. It's gonna be epic.
1:13:10That'll be in the next part. But first, ClaudeMD. What is a ClaudeMD?
1:13:15What does it do? How do you make a good one? Let's let's talk about it.
1:13:19So ClaudeMD is just a markdown file with the name Claude in all caps. It is just a text file.
1:13:26What makes it special is that when you title a file Claude in all caps, anytime you open Claude in that folder, it will reference this file first before it does anything else. And so this is useful because now we can spawn in different Claude instances, and they sort of have this, like, memory and instructions for working in that project.
1:13:50So you don't have to prompt it every single time. So really what CloudMD is is it's it's a it's a starting prompt that you don't have to type out. Basically, it's a saved prompt that's gonna load every time you open that project.
1:14:04And so we we wanna put in here important context clues and instructions for working in that project. This will make sense in a second. Now the best analogy that I've came up with this is a movie called fifty first dates.
1:14:17Now in this movie, this girl gets in a car wreck, and every day she cannot remember what happened any days before. So every day she spawns in, she has no memory of what happened yesterday. She remembers up to, like, her eighteenth year of life, and then the rest every day is fresh.
1:14:34And so in the movie, the guy who ends up marrying her, what he does is he creates a little tape which fills her in every morning about who she is, what happened to her, their marriage, their life together, and then she's like, wow. This is who I am?
1:14:51She walks upstairs on the boat and she joins her family and she's like, wow. This is, you know, this is my life. And so before he made this tape, she didn't know who he was.
1:15:01She's like, who is this stranger? And it was awful. But he made for her a Claude MD so that when she spawns in, she knows what she's doing here, what her purpose is, and she can go about doing the rest of her life.
1:15:13So, yeah, that is the CloudMD. That is the tape. Um, so what do you put in a CloudMD?
1:15:20Here's the mistakes I see all the time. Dumping a load of junk in there. Okay?
1:15:25Treating it like a long term memory. CloudMD is not a long term memory. It is just for some brief macro context about what it needs to get the output.
1:15:37Right? So Cloud does not need to know how you like your coffee, how many kids you have, or every team member on your payroll, and all this other junk, okay, for every prompt. It just needs to know where does it find stuff when it needs something, what are the preferences of working with you, so, like, what tools do you use, um, your workspace preferences, how does it how is it supposed to name stuff, etcetera, and then concise macro context on the big goal your system is a part of and who it's for.
1:16:05So for example, um, the GitHub for my application that me and my developers work on together, the CloudMD for the GitHub explains what type of code that we use, our, like, structure, who does what, what the different features are.
1:16:19It's it's literally like one page long for an entire application because we wanted to keep it concise, um, because we we don't wanna tell Claude how to do its day.
1:16:28We just want to orient it about where it's at and what's going on, and then let him work with us every single day and what we need. So, um, I'm gonna give you guys my own Claude. Like, don't worry.
1:16:39You're gonna get mine. But I want you to know why I've made this one the way it is. And you should prune your Claude constantly.
1:16:45You should make this as short as humanly possible. Um, most people get Claude to write its own Claude.
1:16:52That is dope. Okay? You're gonna do that in my course, but you have got to prune it.
1:16:56He will put in there stuff that you do not want in every single instance of Claude that you're gonna do from now on. Okay? So just think about it.
1:17:04If you're not going to use it, if you're not gonna use a certain line or instruction for every single prompt, then it should not be in your Claude MD. Okay?
1:17:12Just put it in a separate file and refer to it whenever you need it. So here's an example of just a a CloudMD. This is a YouTube course builder, like example thing.
1:17:24So it's like key context of, like, what we use, where where everything is at, so what the folders are, and then the working rules. So here are some logs.
1:17:32Here are some separate chats, whatever. So, yeah, that's just basic CloudMV. So context folder instructioning.
1:17:39Here's a little, like, sneak preview of my proven structure. We'll cover this in-depth in part two, but let's go over some context structuring. So once again, we want to give the exact inputs to get the exact outputs.
1:17:51That is our goal. To accomplish this, we can do three things well. Once again, we just covered context structuring.
1:17:57Now we're going to cover workspace setup. Right?
1:18:02Uh, here we go. Let's start by reminding you of your job with AI. Setting initial trajectory.
1:18:09So clarifying where you want to go is one of the most important things with AI. We are lazy. We don't want to do this, but this will give you the good results.
1:18:17Then number two, realigning Claude to that trajectory. So making sure that he stays on track, pruning all the slop, keeping him in the right direction, that is your job.
1:18:29How? Well, we have to be able to work with Claude. How we do these two things right here is by improving the context clues, so that part of context, giving better instructions in the form of context, and improving the structure so that Claude can find these things easier and easier.
1:18:48And this is extremely hard to do if you cannot see what's going on in your project. And this is why, like, Claude CoWork was made for beginners. I want you to understand this.
1:18:58They have hidden everything from you. They care a lot less about helping you really, um, and they care more about you being wowed by this technology so that you pay for it, which is, you know, it's fine.
1:19:11But if you've made it this far in this course, you were you you do not want to be average with AI. Right?
1:19:16You want to be above average. And so if we want the insane edge that AI gives us, we need to go deep. We need to not just hide everything so that it's simple.
1:19:25We want to understand how the junk works. We need to see. We need to understand.
1:19:30We need to apply the human perspective to our projects. So what does this mean?
1:19:37Essentially, we need a better file viewer. That's literally it.
1:19:41So if we were able to see what Claude was doing better, we would work with it a lot easier. And so right now if you're not using a file viewer, how you would view Claude's files is in the Claude app, which is pretty inefficient right now, having a folder open and then opening the text file in this ugly hard to read form.
1:20:00K? Or you can do the exact same thing inside of a file viewer like Obsidian or anti gravity or cursor or Versus Code where you can have the exact stuff.
1:20:11Right? You can have Claude. You can have the file.
1:20:13You can have the folder structure in one window, and it looks incredible. That's all a good workspace setup is. So this is why we work in an IDE, which just means integrated development environment.
1:20:24Um, usually, just like it's kinda like a browser for your files. It's like using Google Chrome, but for viewing your files and working with Claude. Why are IDEs valuable?
1:20:34Well, they have a great user interface. That's the whole point. Right?
1:20:37To make this easier for you to work with Claude. And they have custom plugins to make it fun. They integrate with AI models really well, and you have the ability to read, write, and edit files and folders easily.
1:20:50That is the value prop here to give better context to Clot. So in this course, we're going to be using Obsidian.
1:20:57This is my personal favorite file viewer. That is because it is just the easiest in my opinion to edit files. It's got a lot of good core plugins.
1:21:05It's got some great community plugins. You can sync this with team members so that they can actually sync with your projects, and it's built around information not code.
1:21:14And so I'm not doing a bunch of coding. I mean, I do write code like Claude does it for me. I'm prompting Claude.
1:21:21And I find that this is easier than Versus Code or something like that because it's able to I'm able to give it better context, really. But you can use whatever you'd like. It's all the same thing.
1:21:31Right? Versus Code, anti gravity cursor. It's all about you being able to see what Claude is doing and help him do it better.
1:21:37That's it. So in part two of this course, the execution chapter, we're gonna cover my exact setup. I'm gonna give you guys literally everything of five hundred plus hours of building, like how I set it up, how I build.
1:21:48But I wanted to explain kind of why we do this, why I'm giving this to you, and how I made it. So these are the principles that I used to make my workspace.
1:21:57This is regardless of whatever application you choose to do with. Okay? Number one, stay focused on output.
1:22:04No one cares how cool your setup is. No one cares about your Notion database, and they especially don't care until you've gotten insane outputs.
1:22:13Think about it. Elon's a trillionaire, And when is the first time you ask, I wonder what his Notion setup looks like, or I wonder what his Obsidian setup looks like? Probably not because people don't care about the process.
1:22:25They care about the value you put out. You care about his cars. Right?
1:22:29His rockets. And so, um, stay focused on output. Now I still want you guys to try stuff out though.
1:22:35Like, have fun, but resist bells and whistles for the sake of bells and whistles. Make sure it's valuable. If something isn't adding value, meaning isn't helping you create better stuff, delete it.
1:22:47Because the goal is clean thinking and clean inputs to Claude, not a bunch of random plug ins and crazy themes. Right? Keep it clean.
1:22:55So, yeah, that was workspace setup. Now there's one more thing to cover as far as principles, and this will help you fully understand everything I'm gonna give you in the template section. And that is understanding different Claude roles.
1:23:08What is an AI agent? We talked about this in the Cowork Masterclass two point o inside of the free school.
1:23:13It is free if you wanna check it out. If you're basic, you should probably watch that, but you're already deep, so let's keep going. An AI agent is basically this.
1:23:20It's the brain model, so the actual code that reasons. It is context, and so how you prompt that code, what context you give him, and what instructions you give him, and then his tools. So what does he has access to do?
1:23:35So he can run commands, he can access different connectors, API calls, and then all this happens inside of a loop. And so if you set a goal and you have the agent continue working until it actually gets that goal, that is kinda what makes it an agent.
1:23:52It's what makes it problem solve and think for itself without you needing to be in a loop. Spoiler alert, you still need to be in the loop. Right?
1:23:59Like, that is the whole point. But this is what makes an agent. And so, um, like I said, most people see AI agents, but I'll be honest.
1:24:09Most of it is hype garbage. Like, these people do not get many valuable outputs.
1:24:14They get a lot of clicks, a lot of attention on how cool their process is, but they're not actually making incredibly valuable stuff. Remember, boring process, incredible outputs is what we want.
1:24:26We do not want incredible process, boring output. We wanna be providing value. And so beware of the people who are building an agent army in paper clip or whatever.
1:24:36No. As, uh, it's not not what we mean by this. But why is having these different roles or agents valuable?
1:24:44Like, why would you even do this? Well, remember, the goal is to get the exact inputs we need to Claude to get the exact outputs we want.
1:24:52And so if you have different outputs you want in different parts of a system, you want to give Claude a different role. So let's say we're trying to heal people.
1:25:01Right? Well, we would want to have a doctor Claude for that. So a doctor suited model, okay, a good brain, doctor context and instructions with the Claude, giving it a stethoscope and surgery tools, and the doctor goal of what its goal is.
1:25:17Let's say we want to ride bulls. Well, then we would give that Claude a cowboy suited model.
1:25:23Probably a I'm not even gonna make that joke. Probably a brain that's more for trucking through stuff than intelligence, giving it context for cowboy stuff, cowboy tools, giving him a lasso.
1:25:35Anyways, you get the point. By having separate cloud instances work on different specific systems in your build, you're able to get better outputs more efficiently. So instead of having all of this inside of one clot and him getting confused, we separate it into different roles, different instances, different chats, and we're able to get better outputs.
1:25:56And once again, this tech is only getting better and better, and so you do not have to separate everything you do into a new chat. Honestly, you don't anymore.
1:26:04But I wanna show you how to max out this skill so that when you get in a situation where you should be separating things, you'll know what to do. Okay? So these are the different roles that I use.
1:26:16I do not use all the hype the hype y tools. Mine is actually really, really simple, and I'm gonna show you guys how to use it in part two. But these are my Claude roles.
1:26:25So the first role is builder Claude, and he is using the best model I can get. He needs macro context of the full build, which is usually inside of the Claude MD, and I run this agent inside of the Claude app, typically co work or code.
1:26:39Doesn't really matter whichever one your your preference is. I start a fresh chat every time. I hit around 50% context window.
1:26:47And so I start a new collide, and I give him the exact same context every time. And you'll see in the execution phase how I do this using my daily logs. That's how I give him kind of some memory.
1:26:58Um, and if I have to start a new chat, like a a deep one in the middle of a project, I will have him write a handoff prompt, which is very simple. You just basically say, hey.
1:27:07Write yourself a prompt for the next Claude. Open a new Claude. Give him the prompt.
1:27:11And the goal here is to keep the context clean. This is the main Claude. This is the one where we literally want to give him exactly what we need, nothing more.
1:27:21And he only needs context for the problem you're solving. So he doesn't have to know exactly every single detail of every feature of your app. If I'm just building the notes section of my app, I only want this Claude to really understand kind of what the macro app is for and how the notes feature fits into that so that he can build the notes feature.
1:27:43I don't wanna go in-depth on everything else. And so that is a skill, but you'll you'll learn it. Okay.
1:27:48Next is the research and VA quad. So this is using a cheaper model, typically using like Sonnet, but he doesn't really need any context besides the initial prompt.
1:27:59And this is just a separate chat that I use for researching stuff, explaining things to me, helping me understand, creating context for the main agent to use.
1:28:09And so let's say I want to use a bunch of examples for my main builder Claude in like an email system, I will have this Claude read 250 short form scripts and then distill those into a two page list of email ideas.
1:28:25And then I'll give that list to the main Claude. So instead of that main Claude doing all this work in the middle of our project, I'll do it in a separate Claude and then just give him the output. Right?
1:28:36I typically run this Claude in Obsidian, which I'll show you guys how to use in the Claudian plugin. Or if I'm working in Versus Code, I will use the Claud code plugin in Versus Code. The main value here is keeping the main agent's context clean.
1:28:49So this the goal of this research, Claud, is just to basically do low do low tiered tasks outside of the main agents that I keep this one clean as possible. That's it. That's all I'm doing.
1:29:01Um, I don't give him any extra context. It's literally just a Claude chat. Like, I just open Claude app, do the chat, or open Claudian.
1:29:07It's our new chat and just ask him questions. And then last one is crash dummy Claude. So I build a lot of stuff for you guys, and I want to test it as if you were using it.
1:29:19And so I cannot test it using the Claude that I'm using to build it. And so I will start a new Claude instance in usually Cowork because most of you guys use Cowork, um, and I will test it as if I'm you guys.
1:29:31I don't have no fresh context. It'll just be the plugin that I've given it or the instructions I'm giving it. And I'll try to use a cheaper model because I know I know a lot of you guys will be trying to do a cheaper model.
1:29:43Also, I use it to test stuff for my application. And so in my application, I'm not using my Cloud app.
1:29:50It is API calling typically a cheaper model than Cloud. And so I want to use whatever model I'm calling in the API to test it in a fresh chat to see if I get the outputs that I want. And so, yeah, the value here is you're able to test your automations in real scenarios that your users will be using it in.
1:30:08Then you can bring back the feedback and give it to your builder, Claude. This is how I made the Cowork starter pack. This is how I made the finance dashboard.
1:30:15This is how I made the second brain. It was using a crash dummy Claude to test stuff and then give the results back to my builder. So that's it.
1:30:23Really simple. Once again, bro, you ain't that g yet. None of us are.
1:30:29If you can't make one clod make you money, you can't make five clods make you money. Keep it simple. Once again, output over process.
1:30:37Do not do this just to feel cool. Right? Start with one clot and work your way up until you really need all these clots.
1:30:45It's crazy. So, yeah, that is all of the three goals. You now understand the principles which underlie the entire actionable part of this course.
1:30:53I know this was kind of a long one, but I wanna congratulate you. Like, you probably understand AI better than point 01% of people when it comes to actually creating valuable things.
1:31:03So, yeah, congrats. We are done with principles.
1:31:06It is time to have some fun and go execute. So, yeah, um, just know that you need both. Okay?
1:31:12I've been harping on tactics a lot here, but really you have to have both. Principles alone just leaves you in your head. You have to go do stuff.
1:31:21But if you don't do stuff with the understanding of principles, you're going to be confused when you don't get the right results. But if you have both, man, you get results. So it's time to go put these principles and actions.
1:31:31I'm gonna share with you the latest and greatest tactics that I found from, uh, helping 10 plus business owners automate real businesses. Five hundred plus hours, lots of helpful automations. Let's freaking get it.
1:31:43Alright. Welcome to part two of this course execution. It is time to start building, my friends.
1:31:48I'm gonna give you guys my entire building process and the template, how I work with Claude, all the context structuring, literally everything. It's taking me five months of building and five hundred hours of working with AI to actually perfect this setup and workflow.
1:32:03And I want you guys to judge it by the outputs, not how cool it looks. Like, what did it produce?
1:32:08And I managed to accomplish honestly some stuff that I've only ever dreamed of, including my first ever custom paid software with lots of paying users. Shout out to who bought the yearly package this morning.
1:32:18You're an absolute beast. But, yeah, you can check out the app if you want to in the description to see just how legit it is. And then also tons of custom plugins, automations, client work.
1:32:27If you want us to build custom software for you and you don't wanna do this yourself, then hit us up, uh, the agency in the in the description. But, yeah, custom thumbnail software dashboards, client dashboards for my one on one clients, all kinds of amazing creations all through this setup that I really think is incredible.
1:32:45And let's get straight into it. I'm give it to you guys for free. No strings attached.
1:32:49There is one string attached though. If you're someone who skipped the principal section, I wanna warn you, you will not get the most out of this until you watch principles. I know that it's kind of boring.
1:32:59It's not shiny, but I cannot explain enough. This stuff did not just come from having a cool template and a good Claudine D. It came from understanding principles.
1:33:09So I cannot recommend enough. You go back and watch it if you're someone who skipped it. But with that being said, if you watch principles, man, you are ready.
1:33:16Let's get right into the sauce. So what is included in this template? What am I actually going to give you today?
1:33:23Well, the first thing is my Obsidian slash IDE setup. So how do I view what I'm working on with Claude?
1:33:31And this allows me to prompt him well to to see what's going on, to see why what's causing problems, and actually fix the context so that he can get better outputs. The second is how I set up my Cloud Agents, how I work with Cloud. Like, what does my workflow with an agent look like?
1:33:49Third, probably most importantly, is my folder and context structure. So how do I structure these projects cleanly to build these robust automations and have Claude work on it without having to kind of start fresh every single time, which leads to some ball in context engineering, which is the whole point of this course.
1:34:08And then last but not least, the workflow. We're gonna go over three levels of how I go from start to finish with automations depending on whether it's beginner, intermediate, or advanced. I'm gonna show you guys all of it.
1:34:18So even if you're a brand new person, this should be able to help you a ton. And all this with a goal of not to look cool, not to get Internet points, but once again to build ball in systems with ball in outputs.
1:34:30This stuff is built for value. I built it for myself not to share. I built it to build things, and so I hope you guys, uh, get a lot of value from it.
1:34:39And also, a lot of stuff I'm gonna cover right now in this module, it will do automatically, but I have to explain how it works so that you can personalize it.
1:34:47You can make this template your own and really get maximum value from it, and you can under stand like we talked about in the principal module. So first up, my Obsidian slash IDE setup.
1:35:00Like I mentioned before, I use Obsidian. I absolutely love it.
1:35:03I've tried like 10 IDEs, 10 different softwares, and I always come back to Obsidian just because it is the cleanest and easiest in my opinion.
1:35:12Here is how I set up my Obsidian. So if you just download Obsidian fresh, this is what these are the settings that I change, but they're always they're gonna be personalized in my template already, so you don't have to change them.
1:35:22If you use my template, they're already there. So let's go ahead and take a tour of my Obsidian and kinda see what it looks like. Alright.
1:35:28So here's the Obsidian setup. I'll run you guys through it real quick. We are going to do live builds with this, so you don't have to mat like master this right now, But I'll give you guys a tour.
1:35:37So the only plugins that I really have enabled are some of the base plugins, so bookmarks and search. That's literally it.
1:35:45And then I have Claudian, which is enable you to have Claud on the side. And then also terminal.
1:35:51So I can spawn the terminal, and I can spawn Claude code if I wanna work with Claude code on the right hand side as well. And so if you if you're new to Obsidian, it really is just a file viewer and kind of a note taking system, which makes it perfect for what we're doing.
1:36:06And if you're brand new, have tons of other courses about this. It's not gonna be a beginner course, um, but it's very simple. You can close the tabs here.
1:36:14I can close this. And so on my left hand side, we have folders and we have the different files. And so I can click in the files.
1:36:21I can edit this stuff. And the way that I actually have this laid out is it has a translucent window on the appearance, which is what makes it look a little cool.
1:36:30But you can turn that off if you want, and it'll look just the same as normal Obsidian. But I've built this to be very, very simple. I just want to prompt Claude.
1:36:39And, um, we'll go over the Claude MD here in a little bit, but this Claude MD actually sets up the entire project for you. You don't have to really touch any of these things, but this is my Obsidian setup. The only setting that I will change is if I'm going to be sharing this with my team.
1:36:54Also, I will check on automatically check for updates. That usually helps. But if I'm gonna be sharing this with my team, I will also turn on Obsidian sync, and then I will sync this to my Obsidian account, and it can share it with my team if I want to, if that's a project that I'm gonna share.
1:37:10So, yeah, let's jump back into the Miro board. So, yeah, that is the Obsidian setup. And the way that you want to personalize this template yourself, which I'll show you guys how to download the template here in a little bit.
1:37:19All you have to do is open up the template in Obsidian, make changes, and then whenever you go to to do a new project, you just wanna copy the folder of the template and then just, like basically, you have this folder that is your template. You can change that Obsidian.
1:37:33And then whenever you want a new project, you just copy it, and it has all your settings. Um, and if you wanna actually see the settings, you just do command shift period in the folder to open up the dot files, so hidden files, you and can see all your settings in the dot Obsidian folder.
1:37:48Yeah. You guys can look that up. Second part of what makes this template so special is how I work with Claude.
1:37:54So what are the different agents that I use, and what are their roles? First Claude is just in the app. So I'm using Cowork or I'm using Code, and I will basically open up the folder that I'm working on inside of Cowork or inside of Code.
1:38:08So that's really it. And this is the main agent that I use. I use the latest and greatest brain usually because I have a good subscription, and then I give it macro context usually inside of the Claude.
1:38:19I run this inside the app like I said, and then I will start a fresh chat every completed like subsystem or system. Or if I hit around 50 to 75% of the context window limit.
1:38:31And if I need to kind of continue working on the same subsystem across different clouds, I will have him check out the daily logs, which I'll show you guys how to use in the execution phase, and I'll have him write himself a handoff.
1:38:44So I'll say, hey, Claude. We're running out of context in this window. I want you to kind of write yourself a handoff for picking up with the next Claude, and it's super, super easy.
1:38:54An entire goal here, like I said in the principles, is to keep this main clod as clean as possible. I only want the context needed for solving the specific problem we're tackling. So if I'm working on my application and I'm working on the note taking feature inside of my app, if I'm working with a a brand new Cloud Agent on that, I only want him to have context around the note taking feature and whatever he needs to make that great.
1:39:22So sometimes that includes a little bit of macro context about the whole app, but most of the times he doesn't need to know that. But let's say the notes play into a feature that we have with the actual video player inside of Signal?
1:39:35Well, he needs to know that context. And so it's all about giving it just enough that they need to solve that problem and nothing more. The second way I work with Claude is with like a VA slash research agent.
1:39:48So what I do is inside of Obsidian like you just saw, I will spawn in the Claudian plug in, or I will spawn in the terminal plug in, and I will open Claud code in terminal. Now this Claud, I just use for doing research. I use it to explore different tools.
1:40:04So let's say I'm going to connect something through an API. I will use it to search the web and kind of summarize what different tools are out there. I will get it to build context for the main agent.
1:40:17So let's say that I have I'm building an email system, and I want the main agent to take some email ideas and rough them out. If I want to take a ton let's say I'm gonna take, like, 200 past emails. I'm gonna feed them to this VA agent.
1:40:31He's going to make me an idea document from all those emails to then hand to the main agent. So it's keeping the main agent as clean as possible. And so, um, yeah, you guys will learn how to do this with time.
1:40:42But, basically, the entire point of this kind of research agent in the Obsidian is it can do these small tasks that I don't want to fill up the main builder agent's context with. So I'm working on a complex feature.
1:40:56I don't want to just start asking questions about this one API documentation in my main thing because I'm just clouding up the context. So I will ask the research agent to just, hey.
1:41:06I'm working on this. Go find this documentation. Give me a summary.
1:41:10Teach me about this thing that I don't know about. So it's like, you know, walk me through it, and then I'll go back to the main builder. So super cool.
1:41:17Another thing here is you you have to have Cloud Code, like, installed on your machine. So not just the application, but Cloud Code in terminal.
1:41:26And so there's tons of tutorials on that. Literally go to ClaudeCode website, copy and paste their bash command, open up the terminal app, and then paste it in, and you're good to go.
1:41:35Once you have that, you can open up Claudian, or you can open up ClaudeCode in terminal app in Obsidian. So like I said, um, this uses a cheaper model.
1:41:44I don't really give it any context besides just the initial prompt that I send in, and it's just for researching, explaining things to me, creating and curating context for the main agent, organizing files, renaming, moving them around the project. And, um, I use it inside of whichever ID I'm in because it's just so so simple.
1:42:01And the value of this, like I said, is to keep the main agent's context clean, creating better context, better systems, you know, better outputs. Clot number three, test dummy clot. So if I'm using an LLM inside of my system, so for example, the plug in I gave you guys for a co op, the plug in I gave you guys for the second brain, the finance plug in, the system I'm creating is gonna going to use the LLM that you have access to.
1:42:29So basically, it's using your Claude. And so I have to make sure that it will work for you on your machine. And so to do that, I have to have a test dummy Claude.
1:42:39Basically, it's very simple. I just open up a brand new Claude chat in Cowork because that's probably the one that most of you guys are gonna be using. Um, I'll use a cheaper model, and I will just test my plug in.
1:42:50I'll test my skill. I'll test my application, my automation, whatever. And, uh, I will not give it any other context besides the context that you would have.
1:42:58I want to test it as if I'm a user. And so, yeah, it keeps the build clod clean, and I'm able to bring notes back in from the crash dummy clod to give to the builder Claude to improve the plug in to then test it again. So that is kind of the workflow.
1:43:15Number three, folder and context structure plus my juicy Claude MD. So if you've not seen the principles modules yet, what are you doing, bro?
1:43:24Go watch them. They're literally the secret to all this. But the way I've structured my context is with systems thinking in mind.
1:43:33And so at the top, we have archive. This is where I will just throw in old project versions or old context, or let's say I just have a big, like, research PDF that I don't want cloud to read every time, but I wanna have it in archive and and so I can reference it. I'll throw it in archive.
1:43:49It's just a folder of old stuff. Next, have attachments. And so Obsidian can reference pictures throughout the notes, and so I store the pictures in attachments.
1:43:59And so anytime I bring a picture into Obsidian, it knows to just put it in attachments and then reference it in the note that I'm talking about. And then journal logs, this is so so beneficial.
1:44:10Um, and if you're not lazy with this, this will change the way you work with Claude. And I'll show you guys, you know, examples here in a little bit. But as I'm building a complex thing with Claude and automation, I will take notes about what I'm learning, the tests I'm running, what failed, what did good, what I'm building that day, what are the problems are.
1:44:29And so that way whenever I spawn in a brand new Claude instance, I can literally tell him, dude, go review the last three days of journals. We're still working on the same feature. Look at the open problems, and then let's just kinda get to work.
1:44:42So it allows me to prompt him with all some, like, the memory of the last couple days without having to, like, re prompt him every single time. It gives him a little bit of, like, short term memory, uh, to get that context in faster. Then the build out folder is where we're actually working on these steps of the system.
1:45:02So inputs and then process steps and then the output. And so this CloudMD, once it interviews you on your project, it will create these folders for you.
1:45:12Um, these are just kind of placeholders to show you what is inside the build out. But if I'm gonna build like an email system, I might have an ideas folder. So how do we come up with email ideas?
1:45:23Then I might have like a rough draft folder, which includes how Claude creates the rough drafts and I was iterating on it. Then I might have, like, a finished draft folder, etcetera, or, like, a uploading folder. It's just the steps in the process.
1:45:36And the source folder is for when you're creating plugins or applications. And so if you're building an automation, you're probably going to work with Claude in the project folder.
1:45:45So it is the source. But if you're building a plugin that other people are gonna download or you're building an application which is going to be hosted, you know, on a server on a website, um, I will put the, like, finished source in that source folder.
1:45:59So, like, the finished plugin version will be in that source folder. Um, that way I always know, like, what I'm building and then what is actually shipped. Like, what is the current live version?
1:46:09Um, you'll learn that as you go. And something I wanna explain is that this is a template. So every one of my projects doesn't look just like this.
1:46:18I've got small nuances. That's why it's essential that you get the principles and learn the skill of working with AI. This is going to take time.
1:46:25It is a skill. But hopefully, I'm saving you hundreds of hours by explaining to you, uh, the good stuff of what actually works. So then we have important info.
1:46:35This is just a document where I save, I don't know, passwords or things that I just need to easily find. And then lastly, we have the build process, which is just a good little reminder of how to build efficient and good systems. Um, it's kinda like a very short synopsis of part of the principal modules, so it's really good to remind yourself where you're at and how do you build good stuff.
1:46:57Finally, the claude.md. So how I've structured this claude.md is really to help you get the best start to your project as possible.
1:47:07And same for me because this is what I use. And so all it does is it basically says, hey. The user is going to build one of these three things, an automation, a plug in, or an app.
1:47:17So first thing we wanna do is interview the user. So get the best inputs as possible out of the user, figure out what their process might be, like what they're thinking about. Offer to help them do research.
1:47:30Um, and then second is to poke holes in it. And so offer to review the plan that you just created. Third is to write the spec.
1:47:37So it'll go through and write, like, which tools we're gonna use, how we're gonna build this. Four, it'll build out the folders. So like I said over here, it will change the names of these to match whatever you agree on in your plan.
1:47:49Then it'll confirm with you. Then it will rewrite this entire Claude. Md.
1:47:54So it'll rewrite the Claude. Md with context of your project based on the answers that you just gave it. Super, super cool.
1:48:01I know it sounds complicated, but like I said, this is a skill. Just get in there. Build with Claude.
1:48:06I'm gonna give you guys three different levels of difficulty to get you up to speed so you can do this. Speaking of the three different levels, let's go over how I work through these automations.
1:48:16Now something I want you to understand is that it doesn't really depend on your skill level. It's more about the complexity of what you're building. And so I will use kind of like level one workflow if I'm building something that is very simple and doesn't require a ton of, like, you know, effort and thought.
1:48:33And so I'll show you guys what that looks like. Level two is going to be, you know, an automation or a plug in or even an application that you're going to build over the span of weeks.
1:48:45So something that's gonna take a little bit longer. It's gonna require more iteration. This is what I think of when you're going to actually give things to your team or give these to your audience or sell these.
1:48:57So I had to spend weeks on the finance plugin because other people were going to use it. You guys are gonna use it.
1:49:04I had to make it simple. I had to test the edge cases. I had make it as valuable as possible with different features.
1:49:09And so there's different, you know, features inside of the plugin that took me weeks to build, and that's when you would want to use kind of a level two setup and workflow. Lastly, level three. This is something that you're gonna work on for months or years.
1:49:23So I am like, uh, like five months in to building my first paying web application called Signal, which I absolutely love. And so I've learned a lot through this.
1:49:33And building something that users are going to use and, like, you're gonna store their data, it requires a whole different level of work.
1:49:42It is not just building a simple Claude skill or an artifact. It requires serious iteration, serious thought, and that's when you want to use the maximum kind of level three workflow that I will show you guys.
1:49:54And what's funny is like level three systems are really just made up of level two subsystems. And so this application has like, you know, 25 different features that are each their own system that we had to combine together and, uh, do that in a way that's safe and user friendly.
1:50:12So, yeah, it's really tough to do. So level one, like I said, it's kinda just for stuff you're exploring or it's really, really simple. Um, I don't know what that's doing there.
1:50:21So all you have to do is kinda just, like, use the CloudMD and rip it, which I'll show you guys examples. Level two, how you do this is you're actually going to use my template.
1:50:31You're gonna have Cloud interview you. You're gonna kinda build a v one and ship it. And then the only difference between level two and level three is you're not gonna do that much iteration.
1:50:40Like you'll test it out, you iterate it. But just because you're the one using it or your team's using it, it doesn't have to be perfect per se. Then lastly, level three is when you want to go all out.
1:50:53Heavy thinking, lots of pruning, keeping up with your journals, exploring different options, restarting everything over again. And, uh, I'll walk you through what a a, you know, process looks like with this. But really, this is how you build amazing stuff, amazing systems with Claude.
1:51:09The first is, you know, you want to get as deep as possible on the inputs, the process, the outputs. So getting as much thought as you can inside of the prebuild context. Like I said, we'll go over it.
1:51:20Phase two is building the v one, so just getting an output even if it's not great. Like I said, watch the principles if you haven't. We go over these phases in-depth.
1:51:30And then phase three is the big phase. It is iterating. It is making it work.
1:51:35It's pruning the context, rethinking the systems, rebuilding all the little subsystems inside of everything, starting over, struggling, thinking, rebuilding.
1:51:45I know you guys are probably like, Katie, this is a, like, cop out step. This is, like, great instructions here, but I can't show you this. Like, you have to struggle through this.
1:51:53You've got to build these automations. They're like, there's no way I could tell you how to think clearly and struggle on a notepad with how to structure things.
1:52:04You just have to do it. You have to do it. So anyways, yeah.
1:52:07Then phase four, leveraging your automations. We're gonna cover that in an entire module. Yeah.
1:52:12Those are the three levels of workflows. And like I said, we'll go over each one here in just a second. I'll give you guys examples.
1:52:17But some important reminders, keep it simple. Um, Iteration is the key here.
1:52:23Improving your process, improving your automations over time. And the good news is AI gets smarter and smarter. Meaning, like, some of the the parts of this process might not be necessary in the future because of how smart AI has got.
1:52:36But here's one thing I can guarantee you. You have to be the one who has creative, clear, and innovative thinking.
1:52:43You have to be the one who is guiding Claude, pushing, you know, the agency, the human agency of, like, I want to do this. How do we think outside the box to get this outcome?
1:52:53And over time, you should spend less time in front of Claude and more time in pen and paper. The better Claude gets, the more he can execute your vision. So the more you should spend time cultivating your vision, and that is a hard thing to do.
1:53:07That's why I gave you guys the principles modules because that's the only way I know how to think about it. And once again, the final reminder is what to build matters so much more than how to build it.
1:53:17So just like the, you know, this is really cool stuff about building, but what matters so much more is like what you're actually working on. You could use the level three automations to build something that like, you know, I don't know, just track something random that doesn't help anybody.
1:53:32I would highly encourage you guys to just try to build things that are as useful and valuable as possible. Solve problems that really, uh, are plaguing people and like you, you know, you want to solve.
1:53:42So without the way, let's go build something. Enough yapping. Let's go build.
1:53:45Let's get in Claude. Let's get in the the template, and let's go do it. Alright.
1:53:49Welcome to level one automations. Let's start building some stuff. So here's the process for building these simple, um, beginner automations.
1:53:58Also, if you've never even touched Claude, if you've never made anything with Claude, I highly recommend starting at level zero. I have a brand new Claude CoWork one zero one course completely free. It goes over all these things about settings, use cases, skills, plug ins, uh, your first automation, usage, lots of bonuses as well.
1:54:16Go check that out if you're brand new. I'm serious. All this is gonna make way more sense if you actually have a little experience inside of Claude.
1:54:24So check that out. It's in the school. Uh, probably link in the description, but if not, just search up useful AI school.
1:54:30Now level one automations. The entire goal of this level is just to just to build something kind of valuable with Claude, to test out this new systems thinking you've learned, to build something cool. So let's do it.
1:54:44Here's the process. We're gonna create a new folder just like that. Um, we're gonna drag in my level one CloudMD.
1:54:50So not the entire template yet. We're just gonna drag in a CloudMD file. Super simple.
1:54:56You're going to rename it to just Cloud. Md so that Claude actually recognizes what it is. Then you're going to open Claude Cowork, select that new folder.
1:55:05So, like, select the folder you just renamed. And then let's say get started. Then you're gonna answer the questions and build with Claude.
1:55:12So let's go do it right now. Alright. So step one, we're gonna create a brand new folder wherever you want to.
1:55:17So I'm gonna call this, um, YouTube research level one.
1:55:23So now I have my new folder. I'm gonna go to the link in the description of the school. Um, it's inside a classroom, then go to start here, and then go to YouTube resources, and then click on the level one claud dot m d.
1:55:35I'm going to download this. It says no preview available because it is just a text file. There's no pictures, but you download it by clicking the top right or clicking the big yellow download button.
1:55:45Then I'm going to drag that to my desktop. I'm going to, um, actually drag this into our folder. I'm gonna open the folder.
1:55:54I'm going to rename this to just Claude dot m d so that Claude knows exactly, um, what this is. Now the Claude app will actually register this as a Claude MD, all caps. Um, so now what we're gonna do, we're going to open up Claude and get to work.
1:56:08Alright. So now that I have Claude open, all I'm going to do is select that folder to work in.
1:56:13So I'm gonna go to choose a different folder. Make sure you do folder and not project. I'm gonna select a folder, go to desktop.
1:56:21I'm gonna select that folder we just made, which includes the Claude. Md, which I gave you guys. I'm gonna change this to act without asking because I feel like Claude is pretty safe these days.
1:56:31I'm just gonna say, um, let's begin. And now what it's going to do is it's going to read that CloudMD that I showed you guys, and it's going to interview you about your first automation or your beginner automation, and then it's going to build it after it's done.
1:56:46It's gonna create the folder. It's gonna do everything for you. All you need is just a folder, the Claude MD, and then Claude.
1:56:53Like I said, this is for beginner automations, but this is the best way to start. Because you can always take a beginner automation and upgrade it to like a level two or level three if you find that it is really valuable to you and you want to make it more robust and better.
1:57:07Alright. So Claude now has the first question for us. He says, what are you building here?
1:57:11I'm just going to build an automation. So basically, I'm just gonna work with it in Claude. It's not gonna be a plugin or it's not gonna be an application.
1:57:18It's gonna be just a a folder. So I'm gonna say automation plug in or an application.
1:57:23Sorry. Alright. The next question is what comes out of it.
1:57:26So it wants to know what is the output, what's the point, and what does the perfect version look like? So I'm gonna tell it to it real quick. I use something called Super Whisper.
1:57:34Uh, you can get the free version. I have links about this in my school. I'm not affiliated or anything because it's free, but you can check it out.
1:57:41Okay, Claude. So what I want to get out of this research, um, automation is I want to give you a video idea that I'm working on, and then I want you to give me all the competing videos based on the keyword demand that you can find.
1:57:57And so we can probably use like the vidIQ plugin for this or connector, I mean.
1:58:03And then I want you to basically to give me all the keywords around that idea that are doing well, then give me some tags for the video, and then also find some videos and thumbnails that are a little bit broader than the actual video that I'm doing. So find me direct competitors and then find me, like, broader niche competitors.
1:58:22And then based on all that data, kinda give me some YouTube titles and some thoughts about the thumbnails for my video.
1:58:30And all of this should be in like a nice HTML file as the output. Okay. So that is kind of some really rough just giving hits some stuff.
1:58:40But the better you give it on this context, the better you answer these questions, the better your output is going to be. I'm telling you guys. But this is a good way to start just to make something get an output.
1:58:51Alright. So Claude has taken that context, and he has actually searched for some tools of what he could use.
1:58:57There's a connector called vidIQ, which I've already using. So I'm just gonna connect that.
1:59:02It's going to find the keywords. And say, what is your channel and niche? The input when you bring me a video idea, what form is it?
1:59:11Okay. So let's do this. So I don't have any examples of research done before, but the idea that I'm working on right now is a full AI automations course using Claude.
1:59:23And so the value of this video is gonna teach someone how to build valuable systems and get crazy outputs, um, using Cloud AI and using my templates.
1:59:33It's gonna be a long course, so a couple hours long. It's gonna be awesome. So that is the input.
1:59:37I want you to find kind of competing videos directly to that, that value proposition, and then also find videos that are just kind of in the general AI or claw niche.
1:59:48Alright. Cool. So now we gave him a little bit more context, and we'll see what he comes up with.
1:59:52Alright. So he's asking a few more questions. He's want me to run the real research on your idea.
1:59:57I'm gonna say yes. Um, actually, I will build the system first, then run it.
2:00:02How deep's the research go? I'm gonna go 15 each just to see if we can really juice this thing. Alright.
2:00:08So now Claude is actually kinda taking this plan, and he should be building the system. And so we should start to see some folders and things pop up. Like I said, this is level one, so you really don't have to look at this at all.
2:00:18You can just kind of run the system, um, and kinda get an get an output. And he's using this plugin that we found, but this is a little note here. In level two and level three, you would want to do way more research about what tools are available.
2:00:33You might want to test different tools. Like, is vidIQ good, or should we use an API? Like, these are all decisions that you have to make as the system designer because Claude's just gonna head in the direction you say.
2:00:45It's just like we talked about in the principles. The like, how you start with Claude is so important because that's where he's going. Like, it's it's autocomplete on steroids.
2:00:54And so you have to be the one to pump the brakes, to slow down and say, hey. Let's actually test this. Should we use vidIQ, or should we use the YTJS API or Supadata API?
2:01:05Like, you have to ask questions, go deep. Like I said, for level one, it doesn't matter, but this is a point here. This is where, like, things differ from just rolling with Claude and actually freaking thinking about the system and building something super dialed.
2:01:19So as we can see here, Claude is actually building out the system in the forms of folders. These things don't matter much because like I said, you're not really going to use, uh, too much context and use Obsidian.
2:01:30You're just literally gonna build a system in here in a Claude chat. But you can check it out if you want. You can see kind of all the steps that it's doing.
2:01:37You can see the actual outputs, the HTML report. And so, yeah, let's see, um, if he can get some decent outputs with a very, limited context that I actually gave, bro. Alright.
2:01:47So what he's doing now is he's actually rewriting the CloudMD. So the CloudMD I gave you was prompted to, you know, collect context from you, ask questions, figure out the build.
2:01:57But now that it's built, he's actually rewriting the CloudMD to have context about what you've already done and what what it's built. Pretty crazy, but it's really cool. Like I said, you don't have to know how this stuff works in level one, um, but just know, like, that's what it's doing right now is it's actually rewriting its own Claude so that if you open a brand new chat with this folder, that Claude instance knows what's going on.
2:02:20It knows what the system is. It knows how to run the system, which is just really, really cool. And just like that, he is done, and he actually has us an output.
2:02:28Um, so here we have prebuild context, the template, the Claude MD. So I wanna actually see this template report and see what we get. Let's pop that open and see.
2:02:38So keyword demand, direct competitors, broader niche, tags, there's titles, thumbnails. Dope. So that's a pretty good looking template, I would say, to start with.
2:02:46So let's actually run this. So let me tell Claude. That looks great.
2:02:51I want you to actually run, um, this search and get me actually finished output on the course idea that I talked to you about.
2:03:01Cool. And while he's doing that, we'll talk more about how to progress your builds into each level and how to kind of work your way up to, a finished live web app that people can pay for. But this is why you wanna start with Claude first instead of immediately building an app is it allows you to go through and actually iterate really quickly.
2:03:22Because I can be like, oh, I didn't like that. Get you know, move the tags down or change the titles like this or do this before you actually make it into a plug in.
2:03:30Because once you make it into a plug in and then to an application, it's really hard to make these changes. But while doing it at this early stage just with Claude, you can dial it in and then wrap it into a plug in.
2:03:42And just like that, Claude has an output for us. So let's open up this research in Google Chrome and see what we got. Let's actually let's take a peek.
2:03:52Alright. Just like that. Full Claude I AI course.
2:03:55Okay. The read Keyword demand.
2:03:58Okay. Cool. So we have all our keywords here.
2:04:00That's actually saved me so much time. Y'all have no idea how long that would take me to just type these in manually, come up with these based on every video. This already saves a lot of time.
2:04:09Here we have some direct com competition, and so not bad. I don't like the formatting of the thumbnails or whatever, but it did find me videos, um, that are compete competing with the video that I'm going to make that you're watching right now.
2:04:23How how crazy is that? And so I've got about 20 of these. Pretty cool.
2:04:28Then I have some broader niche competitors, so things that are just in the AI space. Once again, not perfect, but I do have some thumbnails that I can look at for inspiration and packaging inspiration from these videos.
2:04:42So, honestly, you know, this is valuable. It's got tags for the video already. Amazing.
2:04:46Super valuable. We have some suggested titles here, and they're ranked based on kind of vidIQ. I don't love these, but some of these are okay.
2:04:55Not bad at all. And then some thumbnail ideas. So here are some concepts on that.
2:04:59And so, literally, the automation is working.
2:05:03It is going from the input to an output, and it is valuable. Now you can already start to notice the problem with level one automations.
2:05:11And this is why people probably like yourselves use Claude, and you're like, you know what? This is cool, but it's really not life changing yet.
2:05:20And that is because co work hides a lot from you. Like, it's hard to go through and read these.
2:05:26It's hard to change them. Like, you can't edit the text.
2:05:30You can if you, like, click on the code, I think. Yeah. You can't even do that.
2:05:35You have to tell Cloud what to change. You can't really see the structure. If you wanna open this up in a folder, it's honestly not that much better because now you have to go through and sort through all these folders by clicking into them and then trying to edit with text edit.
2:05:48This is the problem. So level one is great for systems that you don't have to dial in, but it becomes a huge issue when you're trying to really iterate on something.
2:05:57And that's where the value comes because you need that human taste. And so what I wanna take you guys through is we're actually going to rebuild this same automation, but using the level two process.
2:06:08That way we can compare the outputs, and you can see just how much a little bit of tweaking and preparation can really change an entire system. And you probably wonder, KJ, is this an automation?
2:06:18Like, don't you have to build these things and make and whatever? Not anymore, guys. AI does it all.
2:06:23So if I want to run this again so let's say I have an input, I have a new video, and I want this research. All I have to do is click on new task. I just select that folder, and all I do is say, you know, alright, Claude.
2:06:37I want you to do research for a new idea. This video, we're gonna go over how to decrease the usage used in Claude, so how to maximize usage as possible. I do that.
2:06:48It's using that folder, which has already written the CloudMD. It's already told itself what to do when I ask this, and then I just run this off. So I'm not going to, obviously, but that's it.
2:06:58That's the whole automation. It knows what to do. All you gotta do is select that folder when you want to run the automation, and you're good.
2:07:04And in level two, we'll talk more about this. But when you dial this in, that's when you create a plugin. In that instance, you don't even have to use a folder.
2:07:12All you gotta do is go new task, and then you just unselect this, and you just choose the plugin called, like, know, KJ thumbnail research, and I can run it. And the cool part about that is you can then give that plugin away as a free value gift, like, to an to your audience or even give it to your teams that your team members can run that and do research for you.
2:07:33So super cool. Let's move on to level two. Hopefully, you guys can see this is dope, but you can already see the limitations that level one has.
2:07:41Another important note on these level one automations is that if you were to go through and try to edit this system, you're now using a ton of usage because everything that you run-in this chat, it's having to reread all of it.
2:07:55We we learned about this in the principles section. And so if you're gonna edit this, it's really hard to edit it, and it takes a lot of usage versus using Claude and the app.
2:08:06So having a separate instance of Obsidian, I mean, not not Claude. Having another Obsidian using Claudian or using Claude coated Obsidian, we can open up this folder and go get tactical.
2:08:16We can read the MD files. We can have another instance of Claude, change them, and reword them, and then test it again without having to use a ton of extra usage. Anyways, I know that's complicated, but you will figure this out over time.
2:08:28I promise. But right now, you should just build something. So if you have not built anything yet with Claude like this, I want you to stop right now.
2:08:35Don't go to level two. Go download the Claude MD, open a new folder, run a new project, and get an output. Get something that is somewhat valuable to you even though it's not perfect.
2:08:46And, yeah, then you can move on to level two. So I'll see you guys over there. Alright, ladies and gents.
2:08:51Welcome to building level two automations. Now the goal of level two is to build something a little more complex, to have your hands kind of working with Claude a little bit more, getting used to working with Claude and Obsidian, having kind of two different Claude instances for different, um, functions. This is the point of level two.
2:09:10So if you haven't seen level one, go watch that one. Do it first. It's pretty simple.
2:09:14Now let's get into level two. So here's the process of building level two automations. First, you're going to download my full template.
2:09:21So the entire folder template inside of the school, which a link will be in the description, of course. You're going to unzip that folder on your desktop. So just right click extract here.
2:09:31Then you're going to duplicate that folder. I'm a say that again. Duplicate the folder.
2:09:36The folder is the template. You don't want to go into it unless you're trying to modify the template. If you're trying to make a new project, duplicate the template.
2:09:44It is a template. That way you have all my Obsidian presets. Anytime you wanna start a new project, you just duplicate it.
2:09:51Then we're going to open that folder as a vault in Obsidian, uh, then you're gonna be able to actually see that new project that the template that you used. Then we're going to open up that folder also inside of Claude, and we're going to start building the automation level two.
2:10:06Um, but this time, you can see what is going on because you have Obsidian, and you can actually work with a separate instance of Claude to improve the context, to build things, to clean things up as Claude is working on the main application. And then we're going to kind of iterate this with one good iteration and see what we can come up with for level two.
2:10:24So let's go build out that YouTube research automation again, but this time on a whole other level. So like I said, first step is downloading the templates. You wanna come to the classroom in My Useful AI School.
2:10:34Click on start here. Click on YouTube resources, and then click on KJ automation template folder. So it is a folder, so there's no picture, but, um, you can download it by clicking the big yellow download button.
2:10:46Please just, like, rewatch this and don't ask me questions. I get so many questions about this. Just click download or click in the top right on download and boom.
2:10:54There you go. Now you have the zip file, which is just a folder that is compressed. Drag this to your desktop.
2:11:00I want you to right click and extract here. And so boom. Just like that, we have the folder outside of the zip.
2:11:08You can now just delete this zip file. You don't need it, and you can download it again if you ever need it. So this is the project template.
2:11:14You can put this inside of your second brains in your project folder. You can put this wherever you have the things that you're building with Claude. Just stick this in there.
2:11:23Anytime you want to do a new project, it's very simple. You just right click and duplicate. That's it.
2:11:29Watch that again if you guys need to. Right click, duplicate anytime you wanna start a new project. Now I'm going to rename this to the actual project we're gonna build today to YouTube research level two.
2:11:42Awesome. Now what I'm going to do is I'm going to open this up in Obsidian. So I'm gonna grab my Obsidian, and I'm going to actually, um, go to manage vaults if you already have a vault open.
2:11:53I'm going to open folder as a vault. I'm going to choose that folder that I just created. So YouTube research level two.
2:12:01I'm gonna hit trust author enable plugins, and I'm going to just do automatically check for plugin updates, which I think is just, uh, cool to add. And now we can see this project, which is really awesome.
2:12:14So now what's going on if you're new to Obsidian is I am actually inside of this folder in Obsidian. So here I can see all the folders, the subfolders.
2:12:23So, like, this is, what you see in the file viewer. Same exact thing as what you're viewing over here. It's just an easier way to work with Claude and to see your files.
2:12:31And so what I'm gonna do is, um, check out the build process, import info, all this good stuff. Now before we actually start working with Claude, how we're going to make this level two better is I want you to spend some time really filling out the prebuild context.
2:12:47What I like to do is I'll actually fill in as much as I can in here, and then I'll go with a note, like a pen and paper, and I'll write out what I think the system should look like. I'll get really clear about what I want. Because like we talked about in the principles section, if you don't know what you want, it's very hard for Claude to read your mind and give it to you.
2:13:07But the clearer you can get on what exactly you want to see, the better Claude can make it. And that's why in level two, we're gonna spend more time than just a few questions, um, answering it here.
2:13:18So now how do we actually start building with level two? Well, same reason same way that we start with level one. We just go and open up Claude.
2:13:25But before that, like I said, if you're wanting to use Claude in here to do, like, the research and VA tasks, you can do it by just by popping up in Claudine, or you can click over here on the terminal, and you can spawn in a terminal instance and then type in Claud. But make sure that you have Claud code downloaded, the actual CLI.
2:13:44So like I said in the original overview module of the execution, go download Cloud Code in your actual terminal app. So this will be able to give you both of these options to work in Obsidian.
2:13:55So let's go start up this project. So we're gonna do is we're going to go to Cloud Cowork or Cloud Code, whichever one you like. They work the same.
2:14:02Um, I'm going to basically turn off all my connectors to start with, and I'm going to choose this brand new folder we made to work in. So I'm gonna go to desktop, choose research level two, and I'm going to say let's let's start, I guess, is what I'll say.
2:14:18Now this is the exact same way that you start with level one. The same exact thing.
2:14:25Same process. The only difference is I want you to spend some time researching on your own how you should build this thing. So a couple examples are what tool are we going to use to actually find the YouTube videos?
2:14:40And so something that I might do is I might come inside of of Obsidian here. I might open up this Claudian, and I will say, I'll just do this.
2:14:51Hey, Claud. I'm going to work on this build here in a second, so I don't want you to start the build quite yet. But what I do want you to do is figure out what are the different ways that we could actually pull YouTube videos and find YouTube videos.
2:15:03Is this, like, through an API? Is there connectors? Like, how can I basically search YouTube by using Claude to do it for me inside of an automation?
2:15:13So I might spawn this off and do some research on that, and it's actually not connected great. So I have to start a new conversation. Try it again.
2:15:21But you get the point. I would actually so there we go.
2:15:24We're actually building now. What I would do is I would send that prompt off of like, hey.
2:15:29I want to actually go research this before I start. That way I can fill out this prebuild context better. And then I would spawn in another instance, and I'd be like, okay, Claude.
2:15:37I want you to, um, let's work. I want you to figure out, like, how can I find keyword demand?
2:15:44Like, what tools can I use to figure out where demand is on YouTube? I would spawn that off. And so I would do research to fill in this pre build context, and that is the difference.
2:15:54So what I'm gonna do is I'm gonna go through, I'm gonna fill out this project. I'm going to just work on it for like, you know, a couple hours. And we'll just see the difference in a level two automation and a level one automation, and I'll kind of show you what I did and what the project looks like.
2:16:09So yeah. Let's go do it. Alright.
2:16:11So I now have kind of the level two of building this YouTube packaging research automation. The first thing I wanna do is I wanna show you the output. So if you watched level one, you saw the level of output that we had from these level one system.
2:16:24Here's level two. So here we have a little bit, uh, cooler looking thing off the rip.
2:16:30We have it's like big title, which is honestly a way better title than the last ones. We've got kind of some thumbnail things here, and then the best kind of golden competition. Who had the best video that's similar to mine?
2:16:42We've got the keywords most relevant to me, so it kind of separated the keywords into most relevant to the video and ones that aren't relevant. Next, we have the competitor section. Wow.
2:16:51These look way better as far as just like, you know, the formatting looks better. Also, have the ability to see which ones are outperforming the other videos on each channel. So, like, this video is a huge outlier for this channel, which is really cool to see as well.
2:17:06Um, also, these videos look like they're a lot more on point than the original level one automations videos. So these are a lot more accurate.
2:17:16Pretty cool. And then we look at the broader inspiration. So there's actually a lot more here.
2:17:21They're in this smaller format, which is nice, and they're separated by what type of broader it is. This is just the Claude AI. So broad as far as Claude is concerned.
2:17:31Then we have AI automation. It's just the keyword there for all of these. And AI automation agency, so all of these are falling in that bucket.
2:17:41Overall, this is way better inspiration. So there's a lot better inputs for me to come up with thumbnails from this clot output. So overall, we're already off to a way better start than level one.
2:17:53Like I said, I'll show you guys how I did this in a little bit. Um, and it looks really cool. And I I can actually click 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. Next section is title elements and ideas.
2:18:05So I really like how it's done this by talking about the different, um, elements of titles I can use. Then it gives me some titles based on my voice, so it actually knows how I like titles, which is huge.
2:18:19This is like a big these these titles are way better. Like, go back and watch level one. Look at the titles, and look at these.
2:18:25These are a lot more like a KJ video. I might even use literally some of this, which is awesome. Um, experiments worth testing.
2:18:32So Claude actually gave me some out of the box titles that aren't like mine that could be worth looking into or stealing inspiration from. And then, uh, which ones vidIQ like the best? We have some tags pretty similar to the other one.
2:18:45And, um, yeah, that's pretty much it. But overall, way better output.
2:18:50Like, way better. Like I said, it looks similar, but it's not about the looks here. Like, why am I building this automation?
2:18:56It is to come up with a the best packaging for the video you're watching right now. So what I can guarantee you is I will be using what you see right here to make the title and thumbnail that you just clicked on, which is crazy. So this is very valuable to me, and it's a lot more valuable than the last one because these are a lot more accurate to the video I'm trying to make, and it gives me better insights, and I can make better judgments on my own stuff.
2:19:19So pretty, pretty cool. Now how do I do it? How do we build this level two?
2:19:23Well, this is what the obsidian looks like. I know it looks a little crazy, but I promise don't don't worry about this.
2:19:29I built this in one day. So here's the pre build context. I took time to really answer those questions.
2:19:36So instead of just firing off like a couple short prompts to Claude like we did in level one, I actually took the time to sit down and figure out what am I building? Why am I building this? What do I wanna see come out of the system?
2:19:48What does a perfect output look like? Like, why am I doing this? What is a process that might work?
2:19:55Also, what I did was I tested out different things. So these are like I tested, like, where's a good place to search these videos?
2:20:03Do I use the vidIQ MCP, SupaData, or YouTube JS? I tested different tile variations, um, based off of my own skill, and I did all this in putting it into these different folders.
2:20:15Um, this looks complicated, but the only difference here seriously, the only difference is that I took way more time to fill out the prebuild context. I know that sounds crazy, but that's literally it.
2:20:28The closer you can get to where you wanna go, like or the clearer you can get the output for Claude of what you want, he can help you do the rest. And this log, I just kept manually.
2:20:39So I literally said, like, okay. What input will I give Claude, and what is the output I want out of this system?
2:20:45Like, what do I wanna see? And so by clarifying that, I came up with questions like, should I use vidIQ or should I use Subadata or is there something else I can use? And so guess what?
2:20:56I spawned in Claudine and I asked it, hey. Do some research on what kind of tool I could use to find these videos. So it's all about spending time to really make things clear, building the system with Claude as opposed to level one where we just let him make assumptions and run with it.
2:21:13Now we actually got a lot a lot clearer. And so, yeah, that is really cool. And now if if I wanted to make this into a plug in now, I could because I think it's really dialed.
2:21:22The input is literally just, hey, Claude. Here's an idea for a video. Do the research.
2:21:27The output is this amazing document, which literally does all the work that used to take me four hours at least a video, maybe even eight hours of video, and now does it in minutes.
2:21:37And it does it better than I did, which is absolutely incredible. So, yeah, this is ready to then make into a plug in or automation.
2:21:44We'll talk about this in, um, the leveraging your automations, um, module. But just know, like, this is when you get to the point where you're getting really good outputs with just a, you know, files and clod, you're ready to actually package this into a plug in that you could give to other people, and you can test that plug in first and then give it to them.
2:22:02They can do the exact same thing for themselves. Yeah. Let us move on to level three where we really get deep on the crazy automations and building full applications.
2:22:11Alright. Welcome to the final level, level three automations. Now the goal of this is really the end goal of everything we're building to build something valuable with multiple features, subsystems that you can eventually sell or give away like I do to people like y'all so that you can use them and create value in your life.
2:22:29Like, this is really kind of entrepreneurship one on one, but it's online. So it's online building building valuable solutions using the tools that we have.
2:22:39This is like max level. So let's get into it.
2:22:42The process, uh, for level three for setting up is the exact same as level two. Right? You want to download the folder template.
2:22:49You want to unzip it. You want to then copy and rename it. So if you haven't watched level two, go watch that.
2:22:54I won't redo it again. Then you're just going to open whatever your renamed project is in Obsidian, and then open that in Claude, and you're good to go.
2:23:03So what is the difference between starting at level two and level three? Well, level three is you really want to fill out the prebuild context in-depth, like do some serious thinking.
2:23:14When I build level three automations, I mean, I'm, like, literally on a whiteboard, like, mapping out stuff, testing out different processes, um, because I want to start in the best way possible.
2:23:27And, normally, I don't start by building a level three automation, honestly. It starts out as a level two. And then once I decide that, man, this is really valuable, I should, like, build this out into a plug in or, uh, like a like a full app.
2:23:41Like, I will then transition to level three. So a good example we'll look at here in a second is my finance plugin. I built that for myself.
2:23:49It was not a plugin. It started out as just a simple automation. Then I realized, wow.
2:23:53I could build this into an artifact. Like, let me actually step this up. It became a level two project, then it became a level three project.
2:24:00And I was like, man, this thing is in-depth. I worked on it for over a month. It has different features.
2:24:05I'm ready to package it up into a plug in and give it away to you guys. And so, um, yeah, you don't really start as level three. Start it off as level two.
2:24:13Fill out that prebuilt context as much as you can, but, you know, don't don't get too narrow thinking. Like, leave some room for Claude to kind of poke holes and stuff.
2:24:22And then you start it off by, once again, opening up that project in Claude as a folder and then just running, say, let's get started, do whatever. Now here's the difference between level three and level two. It's really on the back end.
2:24:34It's learning how to work in Claude, like in Obsidian with Claude and changing the context, like pruning the context, iterating, iterating, iterating.
2:24:47There's a reason why there's so many software engineers, but very few of them are billion dollar founders. It is not the technical ability.
2:24:55It is it is the judgment and the clear thinking and the ruthless iteration to to simplify and to make things better and to focus. That is something that I cannot teach you.
2:25:06I can make you aware of how important it is, which is what I hope this course has done, but I can't teach it to you. So I wanna make you aware of it.
2:25:13I wanna give you examples of how, like, I've started doing this in my life and the things that I've built, and then I want you to do it yourself and to learn this skill. So with that being said, let us, um, dive into what a level three automation actually looks like, um, behind the scenes. Alright.
2:25:29What you guys are seeing right now is my signal application. So this is a real application that is live, and we have paying users on this thing. So shout out to you guys who are actually, you know, subscribed and and doing this, um, using our app.
2:25:40But it is dense. I mean, we have everything here. Billing.
2:25:44Um, we have a big, like, settings. Like, there's so many features on this app, and I'll show you kind of what the core feature is.
2:25:50And that is to help you go from just a basic problem that you have that you wanna find videos on to having a ton of perfect videos for you.
2:26:01So I'm just gonna do, like, audio to learn. Oh, hold on. That's the developer tools.
2:26:06Let's try this now. I want to learn how to do YouTube thumbnails really well for my YouTube channel. And so I want to learn how to like, you know, use Photoshop, what other tools I could use to just make better thumbnails to get clicks and get more views and help more people.
2:26:21So I'll just throw that off, and we'll spam that here. And so I'm just gonna say some experience, how deep you wanna go.
2:26:28Let's just go deep, and this is going to reading your goal. So now this application that I built is going to give me this brand new project and give me videos to help me learn.
2:26:39It's like YouTube, but on steroids for people who learn. Yeah. Pretty cool.
2:26:44So I'm gonna show you guys, like, how I built this entire application, how I did the, um, UI for it, the actual, like, player UI, all that good stuff. It took a long time, took months. This is what I like level three project really looks like.
2:26:57So, yeah, the input was I had a problem I wanna learn more about, and then the system, the output presents me these videos that I can then save to my projects. So I can click on them and save them, um, and then it'll actually download the transcripts and kind of give me summaries and stuff for them. Then if I want to watch these videos, I can do it right here inside of the Signal app.
2:27:18And once again, don't you guys do not have to use or download this. This is it's not supposed to be like a sales pitch. Honestly, I'm more focused on my agency stuff right now, which I'll show you guys that in a second.
2:27:28So if you guys want us to build you custom software, me and my incredible team of engineers, if you're a business owner, hit me up. I'll build you everything we're talking about, um, for a really good price.
2:27:37So, you know, book a call here if you actually want that. But back to signal, um, I wanna show you guys how I made all this stuff.
2:27:44Like, you can literally do different notes. I can add notes in here. I can do different sections.
2:27:48I can uncheck these. I can skip directly to this section. I have the ability to Right.
2:27:54Change the speed. Have the ability to see a summary. I can control the player from the notes.
2:27:59I can see more from this channel. Like, this is a full live big paid application. So that's what I'm saying.
2:28:05This is what we're building in level three. It is it is a lot. Um, but hopefully you guys can see how cool this is and you're excited.
2:28:11So let's go take a look at how I did this. So as you can see here, this is the actual files, the context, the project of Signal v three. Um, so on the left, you can see, like, the archive, the journal, stuff like that, and then we see the process of going from the the first step.
2:28:27So context to query and then filtering batch one, like all this you don't have to know about. But this is how deep it can get. And so if you see my journal of like building this one feature for you can see my brainstorming, was all of my, like, all the research beforehand for this one feature.
2:28:43The feature that I'm building here is literally talking about just how to go from input, someone has a problem, to literally finding, ranking, creating tons of, like, um, results where people can, like, filter through.
2:28:57So I won't go deep in all this because honestly, it just it won't help you guys, but here's what it looks like. It's me every day, like, putting out my thoughts for version two, iterating, testing over and over again.
2:29:09And so from going from context to query, I have tons of variations of just this skill. And I won't show you because this is literally, like, the moneymaker is what I make money from. Um, but it's figuring out how do I go from that one step.
2:29:22So we talked about in principles. Like, this is where the principles get extremely useful because otherwise, you're gonna be very confused watching this. But if the input is user has a problem and the output is we give them every single video that's gonna perfectly solve their problem, the process is what we're building.
2:29:39How do we get there? How do we get from input to output? How do make that output better and better and better?
2:29:44Um, and so this is how I'm doing it. The first step is literally taking their their context and creating search queries to then push into APIs.
2:29:53So figuring out what API do I use? What queries do we actually create? How do I give the skill to Claude or whatever LLM I'm using the application to actually create those queries?
2:30:04Like I said, I can't show you because this is my my moneymaker, but this is me testing it out over and over and over again. And then once I actually have the results back, like, many videos do I collect? How do I filter them?
2:30:16And that's where the second step comes in of, like, the filtering batch. And then how I actually test this before going to application is I literally have it give me, like, a a result.
2:30:28So, like, here's the actual report, the filtered report. And so I'm I'm doing the main value of the application without it being in the application.
2:30:37And so here I can have it do an HTML where I say, hey. This is the input. I'm a copywriter.
2:30:42I wanna get clients through Cloud Outreach, but I don't know how. Boom. Now I'm able to see clearly how the algorithm that I programmed and how Cloud is ranking these videos, and I can see which ones that it's cutting off and not showing.
2:30:55I can see all the ones that it's bringing in. So this is behind the scenes of what's going on in the application. And here I can iterate it.
2:31:02I can get really, like, like, in the weeds. This is what I'm trying to only value I'm trying to give to you guys about this is just showing you, like, there's levels to all of this.
2:31:12If you're gonna make something that's really, really valuable, you have to spend a lot of time on even the simplest the simplest sounding feature. And so that's level three. It's going deep.
2:31:22It's taking days to iterate and redo and prune and try again and all this good stuff. And then I had Claude give me a customer view so that I can see how it actually shows these videos in the project to a customer.
2:31:35So if we go back to this project, this is using the exact same skill right here. So what it's showing here is what it's gonna show me in the HTML.
2:31:43But instead of having to, like, push this to a live application, I can test just the algorithms right here on my computer and then send it off to, um, my software engineers who actually implement all this. So super cool.
2:31:56I also tested the refresh feature. So I'm, like, testing, like, how much does it refresh, which ones does it refresh. Um, if they click refresh, like, what what happens at the top, and how does it actually choose which videos to show, it's really cool.
2:32:11And so all this is what it looks like to really go deep in level three. But like I said, you will get here over time. I just want you to understand how to work with Claude in the app and in Obsidian, how to see what you're working on with Claude so you can actually read it, understand it, and then make changes.
2:32:30Um, and you can see all this in here, brainstorming, old skill, questions to answer, test notes, problems to solve. You have to dig deep.
2:32:38You gotta understand. And then also don't be lazy. Take time to keep a journal of what you're learning each day so that your thinking is clear.
2:32:47And then Claude, whenever you spawn him in, you can just say, hey, man. Read day two. I want you to check out day two, and then let's start working on these problems at the end of day two.
2:32:57And so, yeah, hopefully, this is valuable. I know it seems like a little bit confusing, but that's the whole part of level three, man. Like, this is building billion dollar software.
2:33:04Like, it's really hard to explain in a course, but hopefully, I've given you at least some nuggets or I've changed the way that you think about this stuff so that, um, you can build these better. Like I said in previous ones in in levels, like, one and two, you work up to level three. These are things that you wanna spend months and years on.
2:33:21But just start with level one. Move on to level two. When you're ready, you will naturally progress into level three, and you'll use all these features from the template.
2:33:28But, yeah, I hope you guys enjoyed. So now I'm gonna actually go take a look at how you leverage these things that you're building. So I'll see you guys in the next module.
2:33:36Alright. So how do you actually leverage these automations to, uh, bring them to the market and to create value to generate leads or to make sales or, you know, do whatever you want? This is the most efficient progression that I found for bringing these things into the market.
2:33:51The first step is starting with just an automation. So this is you with a folder in Cloud. We just talked about this in the the previous modules about the different levels.
2:34:00But even when I'm designing an application, I start with just an automation, just project files, just me and Claude working locally on my machine. That is because when you start to worry about the user interface and other people, like, kinda using it and, like, how to actually wire up the the coding on a live site and implementing it where it doesn't frick with your existing features on application, things get messy.
2:34:26But when you build it locally and you get the exact outcome you want, the exact output, talking about systems, you know, when you get the the desired output, that's when you can go ahead and move on to productizing it.
2:34:40And so I highly recommend that you iterate heavily locally just with HTML files, just with getting the output you want with Obsidian, with Claude. And then once your value proposition is dialed, like the core thing you do, the core thing your automation does, that's when you can package it.
2:34:58You can make it into application. You can make it into a plug in. And that brings us to step two, the plug in.
2:35:04So when I'm doing things for signal, just like you saw with this, like, kind of signal, the new algorithm, what I'll do is once it's dialed locally, I will package it up into a plug in.
2:35:15That way I can send it off to my team or my friends or beta testers, and they can see what are the results without having to have a finished application yet. And so, basically, the cool part about this is you have people can can test it without it needing to have any kind of UI.
2:35:32And so the only thing I care about in this one feature of Signal is someone has a problem. How good is our application at giving them the exact videos they need to solve that problem?
2:35:44That's what I wanna figure out. And that's a complex problem if I'm really gonna solve it. And so by making it into a plug in after it's done low like locally working, I'm able to give it to more people who can test it and not just me.
2:35:56People with different problems, different perspectives, and they can tell me how good the actual algorithm and, like, you know, the decisions were from the system.
2:36:06And so the cool thing is real people, you know, test it. Now to leave this phase is real people actually use it.
2:36:13So I know, you know, we all have friends and family who are gung ho on, like, checking out our cool stuff. But before you make something into an application, um, I highly recommend that people actually use it a lot.
2:36:25So people actually use your stuff, not just test it, but they use it in their daily life. That's a good sign that, you know what? I should build this into an application and, uh, actually actually charge for this and actually spend the time putting this you know, deploying it, getting it live, doing the safety and stuff.
2:36:41Now I'm extremely blessed. I have a team of engineers. One of them is an incredibly talented ex Google engineer.
2:36:47Shout out to you, Kaden. He finally quit his job at Google to do this kind of stuff full time, which is amazing in working with other companies.
2:36:55But, um, he does all the back end to keep these things safe. So I will warn you, when when you go to do an application, do not charge people for it until you have made this thing super, super safe and you actually know what you're doing on the software, like, scalable side of things.
2:37:09And if you wanna hire us to do it, we just started our brand new agency. We're, like, launching it officially this week, July. We will build you amazing software like you've seen that we've already built for ourselves here, um, for your small business.
2:37:21We're really excited to do this. I'll be doing kind of the business thinking, and I'll have engineers implementing it, making it safe, and building you custom software. So, yeah, there's the sales page for our agency if you are interested.
2:37:32Link in the description. Yeah. That is the best way I found to leverage these things.
2:37:37Start with just a project, like a folder and Claude, move to a plug in, get other people to test it, see if they actually like it and use it, and then if they do, start moving it to an application. So come up with UI, build the branding, wire everything up to happen on a live site, and then you can actually, you know, send it to people, charge it, and, uh, yeah, make some amazing value builds and make some money.
2:38:00So, yeah, that is how to leverage your automations the best way that I've found. Alright. Let's break down one of my favorite automations I've made in recent months, which is my finance plugin.
2:38:09So I released a plugin which helps you do your finances from start to finish in Claude, helping you with everything. And, uh, it's pretty awesome because I did not think a finance video would get any views, and it's actually doing incredibly well because it's not just a video. I'm giving away a solution, an actual automation I'm giving to people for free to solve a real problem.
2:38:29And you guys' comments and emails have been incredible, so thank you so much for that. But I'll show you guys how I did this. Like, how did I actually manage to build this thing so that people can get this, like, software inside of their Claude and it worked for every single person?
2:38:44It have all the amazing features that it does. How did I do this? Well, let's dive into the project and, uh, let's take a look.
2:38:50Alright. So now we were inside my finance plugin project.
2:38:54So you can see this is kinda like an an older version of the template that you guys got in this course, but this is actually I made a lot of strides learning how to make this template through building this plugin. So here we can see my CloudMD. It's very simple.
2:39:07It's just explaining like this is the plugin that I'm trying to create. This is what I'm doing. This is my workflow.
2:39:13Here's what the plugin is supposed to do. Here's how we're gonna work on it. And then whatever the heck this is, I didn't even write that.
2:39:20He just, you know, did that in here. And then the folder structure layout. So pretty simple, then some rules here that I just came up with as we were working on the thing together.
2:39:30But what I would do is these are the source folders. And so this is like all the skills in the plugin.
2:39:36So I told Claude, hey. I'm gonna make a plugin, but I wanna be able to see the skills that you're actually working on before you package it into the finished plugin.
2:39:47And so that's what the source folder is for in the template I gave you guys. It's for seeing like what the actual version you're working on, all the files in here so that you can actually edit them instead of them only being inside of the Claude app, which is hard to edit.
2:40:01So what I would do is I would it would give me the plugin. So I would, like, see the plugin here. I would download this into my Claude, and then I would run it.
2:40:11I would test it as if I am a user. If I'm if as if I'm one of you guys, um, and I would test different cases and edge cases.
2:40:18And then what I would do is I would just continue to make notes of, like, what was going wrong and what I needed to change. And then I would keep like, you see all these tests here.
2:40:28Like, it's so messy, but real projects just get messy, guys. I would build the dashboards in HTMLs before I even, like, did like, I would just test it with test data before I even did the artifact. And so that's what I would do.
2:40:42And so from start to finish, really, I just wrote out a big, like, prebuild context of the purpose of this dashboard, what the features were.
2:40:52Like, I thought through all of it really before I touched Claude. And that's the secret. That's the kind of thing that everyone's scared to say, and I'm kind of scared to tell you guys.
2:41:03But you are the edge.
2:41:07You are the alpha. Like, you've got to think about what you want. Claude just builds it, which is really, really cool, and he's a good thinking partner.
2:41:15But the reason why I feel like people have have liked my stuff is because I did the thinking for the dashboard. I used it in my life. I figured out what features were useful to me and my wife with our finances, and then I came back and would build a brand new version.
2:41:29I literally start from scratch and be like, alright, Claude. Here's the process I think is gonna work. Here's what we should do.
2:41:37I I wrote out like what are the elements that we want. So the the template, the setup flow, and questions, like how I literally wrote these out by hand of how should the flow go for someone actually setting up the system.
2:41:51And this is all written by hand. And then I had Claude actually go and implement it to make it the actual plug in.
2:41:59Here's a handoff prompt. So I will show you guys example of that of I was halfway through working on something, and I was like, alright, Claude. We're running out of context.
2:42:07You know, write yourself a handoff prompt so that we can open up a new chat and keep working. And then a bunch of sample data for testing it, screenshots. Um, Yeah.
2:42:15And that's how I finally got to the finished plugin, which is has these different skills. So three different skills. Um, I explained to Claude how the skills should work.
2:42:24I gave references for how the dashboard should look like inside the plugin. I made sure and I read all of these skills just by hand.
2:42:32Like, I literally go through and read every line to make sure Claude isn't getting something wrong or adding something that shouldn't be in there. And, uh, yeah.
2:42:40So, I mean, it it took, like, over months of working on it and testing it to actually build a plug in that was worth giving to you guys in the video. But this is what makes these things worth it, guys.
2:42:52Like, it is not simple building scalable automations that actually add value. If it was, we'd all be millionaires.
2:42:59But I just wanna encourage you to take the time to dig deep. Go through every single subsystem, every single step of input to output.
2:43:09Figure out what is Claude doing. How do I make this better? Research tools.
2:43:13Research how you can solve problem in different ways. Um, find better examples. Do all that good stuff, and you'll be on your way to making some really, really cool plug ins.
2:43:24And so, yeah, basically, the workflow for this was, like I said, work in this Obsidian with a Claude co work chat, and then we would build a plug in.
2:43:33And then I would open up a new co work chat where I would download the plug in, and then I would test it in that fresh chat. Take away all I'd make notes on a notepad of, like, the things that were wrong. I'd go back to the builder Claude, and I would tell it, hey.
2:43:46These things were wrong. Reedit the the plugin inside of the source.
2:43:50I would read the source files to make sure I got the changes right. Then I would say, okay. Ship this into a plugin.
2:43:56It would package up to a plugin, and I would repeat. I'd give that to the test Claude, see what went wrong, take notes, fix it, pack a new plug in, repeat.
2:44:05And this is what I did over and over again to make this finance dashboard. So hopefully, that makes sense and helps you guys. But, yeah, let's move on to another automation overview.
2:44:14Alright. So welcome to this plug in and automation that I set up for one of my one on one clients. And so he's a custom home builder, and he wants to figure out how can I give blueprints to Claude and get an actual takeoff and estimate on the other side of things?
2:44:28And so I won't show you all this because this is his project. Um, so I wanna keep it, you know, proprietary to him. But basically, how we built this is we started by just having a project in Obsidian with Claude.
2:44:41And so it started out as not even a plug in. It was just an automation. And we said, hey, Claude.
2:44:45Here's some blueprints. Here's the process we think could work. So we know that we're going to have to give you a couple variables like the formulas and what the current prices are.
2:44:56And then we want you to kind of take the blueprints, apply the formulas, the prices, total everything up, and give us a takeoff and estimate.
2:45:04So I guess I'm sorry I can't show you guys, but this is, like, one of their builds. And, uh, so, yeah, that's what we did. From there, once that worked, we made it into a plug in so that he could test this in his Claude and basically, like, run, um, takeoffs on his own to see, like, what was going wrong and what we needed to fix.
2:45:22And then once it looked good, we made this into an actual artifact so that now everything has some UI to it. He can easily change the variables here instead of having to answer in Claude. He can start a new project.
2:45:32He can go through the measurements, um, formulas, all that stuff, and then get the output that he wants, which could save honestly a lot of hours per week because this stuff sucks manually. But, um, yeah, Claude's now helping out, which is really, really cool. Also, I wanna teach on if you guys want we actually just launched our agency for this.
2:45:51Uh, we are now offering to build custom software animations for small business owners. So through doing the one on ones, I learned that not everybody has a deep passion for AI and systems like I do.
2:46:04Like, I just I love it. I figured everybody loves it, but not everybody loves it as much as I do. And so we decided that launching an agency would be a great addition to the market to fill a hole of people who just wanna pay us to build really complex and amazing custom software and, like, AI agent stuff for them.
2:46:23So if if you're a business owner, you're watching this, you're like, dude, freak. Yeah. Like, I'll just pay you money.
2:46:28You build this stuff to make my business better. Then book a call with us here, and we can get to work on something epic for you, which is super cool. So, hopefully, we have slots by the time you're seeing this.
2:46:36But, yeah, that's what I wanna I encourage you guys. If you do want to get some client work done from us, this is just a plug in, but we obviously build things way more professional when it comes to, like, real applications. So, yeah, check that out.
2:46:47Let's move on to the next project breakdown. Alright. Welcome to the breakdown of my actual real live scalable paid applications.
2:46:55This is the first app I've ever built which has, like, real paying users on it who are getting value and subscribing every single month, which is just amazing. Honestly, a dream come true. And I'll show you guys kinda how I built this.
2:47:06So we went over in level three how I built, like, the core value prop algorithms and, like, what happens behind the scenes. So I figured in this module, we'd look at how I built the UI and how things actually operate and how we came up with this UI and, you know, basically how I built the app to make it look the way that it looks.
2:47:28So this is if you're like just viewing it as an outsider and you haven't signed up yet. Let's take a look at what it looks like when you've actually signed up and can use it. So this is like the home page.
2:47:38On the home page, you can add different topics. So mine are very simple. It's RuneScape and Claude.
2:47:43And you can click on different topics and add subtopics. If I wanna look at like Claude skills, I can type that in here. It'll then change the search.
2:47:51I can go through and see all these videos like a custom YouTube algorithm for me.
2:47:57I can sort them based on views or upload date. I can add a brand new topic. So I'll add something like US World Cup.
2:48:05Do that. So shout out to the American gang. Hopefully, we win going into next week.
2:48:11Yeah. I can search this, and it will then just bring me all the best videos, um, for myself by doing that. So super cool.
2:48:18I can add these videos to different projects. So the entire goal of this app is to learn. And so I have all these different projects for learning about YouTube thumbnails or actually there's one I was working on a lot, which was, um, YouTube lighting.
2:48:31And so figuring out what to light. So I lighted this intro by learning from this stuff, which is pretty cool. So these are the videos that are inside my project that I actually saved to them.
2:48:41I can save them by just clicking on this button. And so everything you see here, I spent a lot of time designing exactly how it looks.
2:48:48And so that's what we'll kind of cover how I did that, um, right now in this project. So this is the Signal v two.
2:48:55This is how I designed all of the things you see here. And so the way that this project works is I create the HTMLs, the designs, and I hand them off to my developer, Kaden, who works with me with our agency and with this app.
2:49:09And, uh, he actually puts them into the live application. And so what I'm trying to do here is just to get the actual, um, looks right.
2:49:20Like, that's the whole goal of this system is to get the UI. So I have a separate project and system for getting the things to work, and this is how I get the to actually look cool. And so what I do is, um, every single page has its own folder.
2:49:34So it has its own kind of sub project. And then in there, just work on the different versions of what I'm looking at, what the goals of the homepage are. And so I want Claude to know, like, whenever we're working on the homepage, I want it to know what the point of the homepage is, what the point of the player are is, what are what are the features.
2:49:52And so this is just a different way to structure your projects. Also, have the design system. So, like, all the things that make, um, Signal's branding are in here so that Claude can reference.
2:50:03And so if I want to design a new, like, page or I wanna iterate a page, what I'll do is I will spawn a new Claude instance.
2:50:11He will knows he knows in the CloudMD like what we're doing here, what the purpose of signal is, what the folder structure is, um, like our tech stack. And then I will just say, hey, we're working on the homepage today.
2:50:22I want you to look at the homepage folder, look at the homepage goals, look at where we left off on v three, and then, um, let's start working on changes for v four. That's it.
2:50:32I'll literally start and then hand it off to Kaden whenever the thing is done. And so then I'll, you know, pull up the HTML. I will go through and test different things.
2:50:43So I'll test like the brightness of the chips. I'll test, um, what like the one click UI looks like. So whether it's a a chip or if it's if it's a ribbon or a magnet, I will just iterate on each little part of the feature like crazy.
2:50:58And when I find something that I like, what I'll do is I will get Claude to say, alright. I think this version is done. What I want you to do is to create a handoff for Kaden.
2:51:09And so based on the CloudMD, it knows who Kaden is and what the purpose of, you know, this workspace is, and then it writes a handoff. So it will tell Kaden, hey.
2:51:18Here's the v three. Here's what your Cloud should do. And so what Kaden does is he takes this into his Claude and actually builds it on the live site, which is epic.
2:51:29Now, um, I also do build applications like without other developers. And so the way that you do this is you literally just hand this off to another Claude that is connected to your GitHub that actually builds the app.
2:51:40If you guys want a full course on, like, building applications, that is a whole different beast. But let me know in the comments or let me know in the school if you want that. And, uh, I might can convince Caden, my my Google developer, to actually teach you guys the real sauce, like the back end stuff that no one talks about on YouTube.
2:51:56I can show you guys how to make these graphics and branding and features and UI and all that jazz. But, yeah, hope you guys enjoyed, and, uh, I'll see y'all in, um, I guess, the next course.
2:52:07Alright. So last thing I wanna say to you guys is thank you so much for watching this. I hope this course has been insanely valuable.
2:52:12If it is, and you you want me to keep these things free, then please send this to your friends. I really just want to share as much knowledge of as God has given me as possible.
2:52:21I consider myself to be extremely blessed beyond belief. God has blessed me more than I could ever even say.
2:52:29I mean, uh, just the fact that he sent his son Jesus to die for me for my sins and how broken I am, I am just grateful every day to be alive, and I'm so honored that he's allowed me to make this stuff free for you guys. I'll ask that you share this with somebody and help them get value from it as well. And a couple of last things of encouragement is is this.
2:52:50This is a skill. And just like how you cannot make a course on like a basketball or make a course on carpentry that's going to immediately give someone thousands of hours of skill, the same thing applies here.
2:53:04You're seeing some of this course, you're a little discouraged about how complicated this is and everything. I want you to know, I didn't touch cloud six months ago.
2:53:11I had no clue how to use AI. I cooked my usage rate in about two minutes of using cloud. I did the dumbest stuff.
2:53:18Trust me. I made dumber mistakes than you can make, but that was the whole point.
2:53:22I made mistakes quickly. And so if you wanna learn this stuff, if you wanna be a a a master of AI, like, we have to put in hours.
2:53:30So just build stuff. Have fun. Don't get caught in perfectionism.
2:53:34Build, build, build, and you will get better. You'll learn how these machines work. You'll learn how you work and how you can prompt them better.
2:53:42So hope this was encouraging. If you like this course, you know, then, uh, share it with somebody. Don't don't comment or don't like.
2:53:48Just just share it. You know? Share it with somebody who this could help, and I'll see you guys in the next one.
The Hook
The bait, then the rug-pull.
KJ Rainey opens by promising the most in-depth AI automations course on YouTube, then spends the first hour refusing to touch Claude at all, arguing that principles, not shiny templates, are the real sauce behind automating ten-plus businesses.
Named ideas worth stealing.
12:41model
The math-to-art slider
Judge whether AI will be good at a task by placing it on a slider between math and art. The closer to math (one right answer), the better AI performs; the closer to art (subjective, meant to be experienced), the worse. Crucially, rate subtasks, not the macro task.
Steal fordeciding which steps of any workflow to hand to AI vs keep human
24:26model
System = inputs → process → output + feedback (in an environment)
- Inputs
- Process
- Output
- Feedback
- Environment
Every app, plugin, skill, or automation is a system: inputs get processed into a desired output inside an environment, with feedback used to improve it. The process is usually smaller systems chained together.
Steal fordesigning any automation before touching Claude
31:40list
Three forms of leverage
- Labor (other people's energy)
- Copies (reusable systems, information, code, AI)
- Money (buys more labor and copies)
All wealth comes from getting more output per input. AI is a hybrid of labor (it reasons) and copies (someone built the model and everyone accesses a copy).
Steal forspotting where a business is under-leveraged
50:00list
The build phases (manual first, then leverage)
- Phase 0: system doesn't work
- Phase 1: it works but takes all my effort
- Phase 2/3: add AI leverage to reduce input
- Phase 4: add money and scale
You must reach a working output by any means (manually) before adding AI. There's no point automating or scaling a system that doesn't produce value.
Steal forsequencing any automation project so you don't scale slop
41:36concept
The 'worth it even if it stays at one' heuristic
Only build something if it would be worth the upfront cost even if no one else ever used it. This forces low-cost, fast zero-to-one ideas that you'd personally use, and the ones you'd actually use tend to be useful to others too.
Steal forfiltering which product or automation to build first
1:07:22concept
Context = context clues + instructions
Context is just files (the stuff you give it) and folders (how you group them). Split it into context clues (explaining the situation so Claude knows what you mean) and instructions (specific direction on how to do the task, i.e. skills). Structuring files into labeled folders is what unlocks good outputs.
Steal fororganizing any Claude project so it stops giving average results
1:39:44list
The three-Claude agent setup
- Claude #1: main builder (Cowork or Code, latest model, macro context in CLAUDE.md)
- Claude #2: research/VA agent (cheap model, Claudian or terminal in Obsidian, curates context)
- Claude #3: test dummy (fresh chat, only the context a real user would have)
Run three instances with distinct roles to keep the main builder's context clean: the builder builds, the VA researches and curates context, and the test dummy verifies the plugin works for an end user.
Steal forany multi-session Claude build
1:47:09list
The Obsidian folder/CLAUDE.md template
- Archive (old versions and heavy PDFs)
- Attachments (images)
- Journal-logs (daily notes of what failed/worked)
- Build-out (00 pre-build context, then process steps 01, 02, 03 to output)
- Source (finished shipped plugin/app)
- Important-info
A duplicable Obsidian vault whose CLAUDE.md interviews you and scaffolds the project. Journal-logs give a fresh Claude short-term memory; build-out mirrors the input-to-output process; source holds the shipped version.
Steal forstarting every new automation with clean, reusable structure
How they asked for the click.
MENTIONED ON CAMERA
Visual structure at a glance.
