The argument in one line.
The AI model is identical for everyone, and your competitive edge is the proprietary context you feed into it — building an AI operating system is the discipline of accumulating that context deliberately.
Read if. Skip if.
READ IF YOU ARE…
- You already use Claude Code for coding tasks but still switch to Claude.ai or other apps for writing, research, or thinking.
- You run a small business or solo operation and spend significant time hunting for files, messages, or data across multiple tools.
- You have built a Claude automation or two and want a mental model for scaling the system without causing accidents.
- You want to consolidate a sprawling SaaS stack into fewer, self-owned tools.
SKIP IF…
- You are new to Claude Code entirely — the creator explicitly recommends his free three-hour Skool course as a prerequisite.
- You need deep technical implementation detail; this video is mindset and architecture, not a code walkthrough.
The full version, fast.
The model everyone uses is identical. What differentiates your output is the proprietary context you build over time. The Four Cs framework (context, connections, capabilities, cadence) gives that accumulation structure: make sure the AI knows your business, can touch your tools, understands how you work, and eventually runs recurring tasks without prompting. The critical safety insight is that instructions are not capabilities — if the key is on the key ring, the agent can use it — so scope permissions tightly and phase trust incrementally using the bike method.
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 – 01:07
01 · Intro
Opus 4.8 launch as hook; claim that the system is a full OS; preview of Four Cs framework; GitHub repo teaser.
01:07 – 04:46
02 · What an AIOS Is
The default shift — reaching for Claude Code before Chrome. Tool-agnostic because it is all local files and folders.
04:46 – 06:31
03 · Context Is King
Central thesis: same model for everyone, context is the differentiator. Tokens as money. Stateless sessions require intentional memory.
06:31 – 08:20
04 · The Four Cs Framework
Context, Connections, Capabilities, Cadence — each layer depends on the prior one.
08:20 – 10:04
05 · Connections
Seven connection tiers to audit: revenue, customers, calendar, comms, tasks, meetings, knowledge. Connect APIs and MCP servers one at a time.
10:04 – 11:04
06 · Claude Code Insights
/insights slash command generates an HTML report of 30-day session history. Worth reviewing monthly.
11:04 – 14:32
07 · How to Organize Files
No single right way. CLAUDE.md changes almost daily. Reorganize quarterly. Only failure mode is disorganization neither you nor the AI can navigate.
14:32 – 15:24
08 · One Source of Truth
Everything in the AIOS; Claude Code navigates to other projects via documented paths. Eliminates the scavenger hunt.
15:24 – 20:12
09 · Agent Risk & The Bike Method
Real incident: agent sent 3 promo emails to 150K inboxes unprompted. Instructions vs capabilities. Bike method for phasing trust.
20:12 – 23:24
10 · Building Skills
Two paths: build forward or reverse-engineer. Session-handoff skill as a minimal example.
23:24 – 25:20
11 · AIOS as Mentor
Treat it as a mentor to consult, not an oracle to trust blindly on high-stakes calls.
25:20 – 28:08
12 · Do You Need a Dashboard?
Probably not. Visual dashboards only worth building if they improve decisions.
28:08 – 28:57
13 · Final Thoughts
Productivity = moving the needle on the goal, not hours worked.
Lines worth screenshotting.
- The AI model is the same engine for everyone; your proprietary context is the only fuel that is actually yours.
- Claude Code sessions are stateless — CLAUDE.md and skill files are the only memory that survives a context reset.
- Instructions are not capabilities: telling an agent never to send emails means nothing if the SEND key is still on the key ring.
- An agent sent three promotional emails to 150,000 inboxes because it picked up a task list and interpreted it autonomously — no one said go.
- You can outsource your thinking to an AI, but you cannot outsource your understanding.
- The default shift is the prerequisite: reach for Claude Code before Chrome, and context compounds over time.
- A skill can be as simple as a repeated prompt you are tired of typing — the session-handoff slash command is just a saved prompt.
- Phase trust the way you teach a kid to ride a bike: walk alongside, then hands off, then watch from the porch — never hand it a bike and go inside for a nap.
- Everything is files and folders, which means you are not locked into Claude Code — the same OS runs in Codex or any other agent that reads local files.
- The only organizational failure mode is disorganization so severe that neither you nor the AI can find things.
- Build skills two ways: forward from a brainstorm, or reverse-engineered from an output you already liked.
- Your first skill should be the boring thing you do every single day.
- A visual dashboard is only worth building if it demonstrably improves decisions — if you can pull the metrics on demand, the dashboard does not move the needle.
- The Four Cs are sequential dependencies: you cannot have cadence without capabilities, capabilities without connections, or connections without context.
- Productivity is not hours worked; it is distance moved toward the goal.
Your context is the only edge the model cannot commoditize.
WHAT TO LEARN
Every builder has access to the same model — what separates a generic AI response from a genuinely useful one is the proprietary context accumulated over time, and the Four Cs is a discipline for accumulating it deliberately.
02What an AIOS Is
- Sessions are stateless: every context window starts blank, and only deliberate investments in CLAUDE.md files, skill files, and connected data sources give the model anything to work with.
- Context compounds: the more tasks you route through a single system rather than scattered across apps, the more the system knows, and the better every subsequent output gets.
09Agent Risk & The Bike Method
- Instructions are not constraints: telling an AI agent not to do something is a suggestion; removing its access to the tool is the only reliable constraint — scope permissions before you need to.
- The bike method is a trust accounting system: every run is either a withdrawal if it goes wrong or a deposit if it goes right, and autonomy is the interest you collect after enough deposits.
10Building Skills
- Skills start with the repetitive: the highest-ROI skills are built around the things you already do every day, because they eliminate repeated friction immediately rather than solving an exotic future problem.
- Reverse-engineering a skill from a good output is often faster than designing one from scratch — do the task end-to-end first, then extract the pattern.
12Do You Need a Dashboard?
- Productivity is not throughput; it is directional movement — building a beautiful dashboard that does not improve decisions is the same as not building it.
Terms worth knowing.
- AIOS (AI Operating System)
- A Claude Code project configured with enough context, connections, and skills that it functions as the primary interface for running a business, replacing tab-switching between SaaS tools.
- The Four Cs
- A framework for building an AI OS: Context (the AI knows your business), Connections (what it can touch), Capabilities (how it works, expressed as skills), and Cadence (automation that runs without prompting).
- The Bike Method
- A trust-phasing approach to agent autonomy: walk alongside the agent on every run, then hands off the handlebars as it improves, then watch from the porch — earning autonomy in stages rather than granting it upfront.
- Instructions vs Capabilities
- The distinction between telling an agent what not to do (a suggestion it can ignore) versus physically removing its access to a tool (a hard constraint). If the key is on the key ring, the agent can use it.
- Default Shift
- The mindset change of reaching for Claude Code first before Chrome or any other desktop app, so that every task contributes to the shared context rather than scattering across separate tools.
- Session Handoff Skill
- A slash command that generates a structured summary of what was accomplished, decisions locked, open questions, and next steps before clearing context, so the next session starts with continuity.
- MCP Server
- A Model Context Protocol server that gives an AI agent programmatic access to an external service without copy-pasting data manually into the chat.
Things they pointed at.
06:31linkGitHub AIOS Starter Repo
08:20toolClickUp
08:20toolQuickBooks
08:20toolFireflies
25:20toolObsidian
20:14toolExcalidraw
Lines you could clip.
04:57
“AI isn't king. Everyone has access to the same AI models. So if AI is king, then wouldn't everybody be king? Context is king.”
17:39
“Instructions are not the same as capabilities. There's a difference between saying don't ever use that key and saying you don't get to put this key on your key ring.”
25:19
“You can outsource your thinking, but you cannot outsource your understanding.”
18:02
“You don't just hand the kid a bike, put a helmet on him, and say go ride. You walk with them, hold the handle, feel if they're adjusting too much to the left, and you adjust them back.”
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.
metaphoranalogy
00:00So Claude Opus 4.8 just dropped, so I turned it into my own AI operating system. It basically is my second brain. It's my executive assistant.
00:07It's everything. It runs all of my different businesses, and I literally live inside of this operating system. As you can see over here on the right hand side, we have a visualization of my actual second brain and all of my skills and other projects and everything that I'm working on in my business being built out in real time.
00:22So in this video, I'm gonna show you guys exactly how I have this set up. I'm gonna talk about my four c's framework for building your own AI operating system, context, connections, capabilities, and cadence. I'm gonna go over the mindset of building one.
00:33I'm gonna go over how you make it better and better and smarter over time, how you're gonna save tokens, and how you're able to organize all of this so that you can scale it up as your business evolves. I'm also gonna give you guys this free GitHub repo, which you can literally clone into your own project and start to just feed in and pump in context and skills and stuff like that so that you can feel more confident about the way that you're setting up your AI operating system based on my two frameworks here, the three m's, mindset, method, and machine, and then the four c's like I just mentioned for the actual architecture of how you build this automation system.
01:03So if that sounds good to you, let's just get into it. By the end of this video, you're gonna know exactly how I actually use this AI operating system every single day. Okay.
01:10So this is my perk two project. If you guys have been following my channel for a while, you've probably heard me talk about this quite a bit.
01:16It kind of started off as my executive assistant and my second brain, and now I just realized it's my operating system. And the reason I say that is because I try to work out of this tab right here in Versus Code more than opening up Chrome or opening up other desktop apps.
01:30I try to do everything through here, and that's kind of like the core default shift where I reach for that first before I reach for anything else. Because a lot of people might think like, okay, what is an operating system anyways? Well, think about it like, you know, your Windows operating system or your Mac operating system or, you know, your iPhone, your iOS.
01:45It's basically just the way that you're interacting with the different systems and the way that you're getting work done. So I'll show off some of my files and folders in here a little bit, but really what I want you to wrap your head around is that first idea is that this thing has all the context that I want it to have, which is basically everything.
02:01It sees my meeting transcripts. It sees all of my posts in school, all of my YouTube video transcripts, all my LinkedIn posts. It can read through my Slack threads, my ClickUp threads, my email.
02:09It knows everything about my business to the point where I have to come in here sometimes and ask it to remind me of things because it can recall it better than I can and faster than I can. And the reason why I wanted to test this out with OPUS 4.8, obviously, it's a new model and, you know, the benchmarks compared to OPUS 4.7 look better.
02:25But if you think about the actual models, I liked OPUS 4.6 more than I liked OPUS 4.7.
02:31It was just kind of a feel thing. And so far with OPUS four point eight, it feels more like 4.6 to me, is great. If you think about some of the things that OPUS four point seven wasn't doing great, Sometimes it had a bit of an attitude.
02:43Honestly, it sounds funny, but it did. Sometimes it would lie to you, and they have a big improvement here, OPUS four point seven, with honesty. Sometimes it was spending too many tokens, and it was being way too sort of like out of bounds of what you asked it to do.
02:54And that creativity was nice, but sometimes obviously it wasn't. So now that I've worked in Opus 4.8 into my entire operating system, it has already felt a little bit better. And by the way, I did just do a whole deep dive video on the launch of Opus 4.8, which I will tag right up here if you wanna check that out.
03:08So anyways, let's just start going through this stuff because I've got a lot of things to talk about, and I wanna get through it quick to not waste your guys' time. So the first thing I wanna talk about is the default shift about an AI operating system because it's not going to be valuable to you, and it's not going to feel like you're getting the ROI on it if you don't actually use it.
03:26So this is where I realized that I needed to use an iOS and where I really started to build it up was when I was opening up, you know, Cloud Chat. I was opening it up on the web all the time.
03:37Had I a bunch of projects in there for helping me write YouTube outlines or LinkedIn posts or school posts or whatever it is. And even though at that point I was a frequent Claude code user, I realized that I was only using Claude code specifically to write code, to do things like building a Python script or building some sort of automation.
03:52But when you realize that Claude code has the same underlying model, Opus 4.8 or whatever it may be, as Claude chat, then it's like, why would you ever use not Claude code? Even if the use case is just brainstorming or thinking or writing, you know, content, nothing to do with code, use Claude code because you start to build up all of that context.
04:11Right? And what I started to do is as I transitioned everything over to Claude code, my Cloud Skills, my Cloud Projects, other SaaS tools that I might have been paying for, I just dwindled down my tech stack because Cloud Code, my AIOS, just does everything.
04:24And it's not only more efficient, but it's also potentially cheaper because you're not paying for the tools, and it just removes the amount of context switching. And so this default shift of thinking, I have x y and z task to do today.
04:37Let me try to do x y and z task without opening up Chrome, without opening up whatever other apps I have. Let me just try to do everything from my Cloud Code. And once you start having that mindset shift, it's very, powerful because AI isn't king.
04:51Who cares OPUS 4.8 benchmarks? Who cares GBD 5.5? I don't care.
04:55The AI isn't the king. Everyone has access to the same AI models. So if AI is king, then wouldn't everybody be king?
05:02Context is king. And the more you use the thing, the better the context gets. So if you take the model, which is the engine, right, and you're feeding in your own fuel, your own context, That's how you get actual useful output.
05:13That's how you get an AI assistant or executive assistant or OS that doesn't feel like you're getting generic stuff. You know? If everyone has access to 4.8, right, and 4.8 can write really good LinkedIn posts, Then wouldn't everyone's LinkedIn posts be going viral?
05:27But that's not the reality. So it's your context.
05:30Right? Context is king, not the AI model. And you wanna be .
05:36You know? You wanna think about the fact that your that your models, your AI models are stateless. So if you open up a new session in Cloud Code, what does it have?
05:44It it loads in its, you know, global rules. It loads in the cloud.md, things like that.
05:48And then from there, it's gonna load in other files or instructions or memories that you've given it. Otherwise, it would be a complete beginner every time. So if you right now open up your Cloud Code product and said, hey.
05:58Based on what's going on in our business, what should I do next week? If it gives you an answer that's horrible, you probably need to give it better context and better, you know, hands to pull in the data from the right sources. And if you get a really good answer, then that's a good good starting point.
06:12There's always room for improvement, though. And so one thing I did wanna call out real quick is if this concept already seems a little bit out there and you don't kind of you know, you feel a little bit lost already, then maybe you want to upskill first with Cloud Code at a basic level before you try to build your whole AIOS.
06:27And I think a good place to start would be inside of my free school community. Link for this down in the description. It's all completely free.
06:32I've got a classroom section, and right here I have a full course about building your own AIOS, the four c's that I talked about. And in here, I basically walk through everything.
06:41This is about a three hour course. So if you wanna start here and then come back to a video like this or watch this video first and then jump into that course, that would be a great way to do it. So this community is linked in the description, and it is completely free.
06:53It's also the largest AI automation community on school, so hop in there. But, anyways, the whole idea of those four c's, context, connections, capabilities, and cadence, is the way that you wanna mentally think about what do you need to give your model access to in order to actually be useful.
07:08So let me hop into the GitHub repo real quick to show you what I mean by this. So the one liner of context is that it knows your business. Right?
07:15You open up a fresh session, and you should be able to say, what does this business do and who works here? And it should be able to answer that. Connections is what stuff can it actually touch?
07:24What's on your calendar tomorrow? What tasks do you have? What messages did John send you yesterday?
07:28What is the you know, what's going on in the general team chat right now? Can it see that kind of stuff, or are you copying and pasting stuff into it in order to give it that context? Then we have capabilities.
07:38How you actually do work. This is usually basically just like the skills. The instruction files of, hey, whenever Nate wants to write a LinkedIn post, you should do it with in this style, you know, use analogies like this.
07:51Here is his writing guide. Here is his framework for writing LinkedIn posts. When it starts to be able to understand how you work, that's when you have true capabilities, and usually those are skills.
08:01And then finally, we have cadence. So turning all of this stuff that we just talked about into things that actually happen while your laptop is closed or when you don't explicitly ask for those things to happen. And each of these layers can't happen without the previous one.
08:14So contacts, connections, capabilities, and cadence. And like I said, that's what I go over in that free course.
08:20So how do you actually think about the connections? Well, for example, in mine, I'm connected to ClickUp, all of my local files, obviously, my Google Workspace, my QuickBooks, YouTube, Fireflies, and so many more things. And the way that I like to think about what to add as connections, because I know at the beginning, you might be like, okay.
08:36I have so many things. Right? Think about on your week to week, Where do you go to look for things, and where do you go to look for these specific seven things as a really good starting point?
08:45Where do you go to look for your revenue figures? Where do you go to look at customer data or customer communication? Where do you go for your calendar, your communication internally, your tasks, your project management, your meetings, and your knowledge.
08:57And, obviously, there's more than this in my business, but this is just when you first sit down and you think about your day to day or your week to week and you do a little audit on yourself, what desktop apps do you have? What software do you pay for?
09:08What are things that you actually are constantly, like, clicking on? You know, what what are your bookmarks on your Chrome? What are things that you're using so often that it would be helpful if your AI operating system had access to those things?
09:19And that's just, like I said, a good place to start. Write them down, and then just start one by one connecting to, like, school's API, Stripe API, QuickBooks API. Just start connecting to the different API endpoints or MCP servers one at a time, and you'll start to build up a really good bank of connections.
09:35And also this GitHub repo will walk you guys through a lot of this. It's it has like a whole onboarding skills, so it'll interview you. It'll have you, you know, connect things.
09:42It'll audit you. It's gonna be really helpful if you're starting from zero right now. Now another cool thing that you guys can do is you can run slash insights right here, and it says that it's gonna generate a report analyzing your Cloud Code local sessions.
09:55So that will give you an HTML file, and then when you open that up in a browser, it will look like this. So it'll show you the date range that it looks through. It will do thirty days, but the only reason mine has twelve is because I recently got a new PC.
10:06So there's only days of local data. But it will show you at a glance what's working, what's hindering you, quick wins to try.
10:15So here are, you know, some skills that you can do, ambitious workflows, and it'll show you different things that are going on. Right?
10:21So what you work on, how you use Cloud Code, impressive things, you know, where things go wrong, features to try, new usage patterns. So looking through this stuff is going to be hopefully very helpful for you to figure out how you can actually streamline the way that you use your AIOS by implementing this feedback.
10:37But not only implementing it once, but checking this every couple weeks, checking this every month, and seeing how the way that you're using it has changed and seeing if your sessions are increasing and seeing if your, you know, different recommendations are changing and things like that to make sure that you're always getting better the way that you're using your AIOS.
10:54And, of course, you can have Cloud Code brainstorm with you on the report findings and just keep iterating. So that's just a pretty cool feature that I wanted to tell you guys about, and it's definitely something to utilize, especially as you're getting your AIOS up and running.
11:07Anyways, I wanted to now talk about how you organize this stuff because I think this is the most requested topic that I get inside of my communities and comments is how do I organize my AI OS. So let me start off by just saying this.
11:22Don't stress it because there's not a right way. There's not a one and only best way to run your AI operating system.
11:30You don't have to do it exactly the way that Nate does it or the way that your other, you know, creators that you watch do it or your friends do it. It all is just folders and files. And when you start to actually realize that, it makes things a lot simpler.
11:42Right? Because when you realize everything's files and folders, the first thing it does in my mind is it makes me realize, okay. Cool.
11:48I could open up these files and folders in Codex or in OpenCLaw. I'm not locked into Cloud Code. I'm tool agnostic here, which is great.
11:55You might see over here, I've got a dot agents folder. I've got a dot Cloud. I've got a dot Codex.
12:00So I can use this entire operating system with multiple different types of coding agents, which is beautiful. And the other thing is it's a bunch of local files and folders, which means AI can look through everything. AI can crawl through it.
12:13AI can reorganize it. AI can search through it. So the point I'm trying to make here is don't stress it too much.
12:18If it gives you guys more comfort, think about this. I probably change my Cloud.
12:24M d file or my agents. M d file almost every day.
12:27I'm changing that thing so often because I live in here so much. So anytime I think, oh, that would be good to know, or, oh, let's, you know, make this more concise, I'm updating that Cloud. M d file.
12:37And I'm also moving around my projects and my files and my folders definitely on a weekly basis as well because every quarter, I have new priorities. Every week, there's something new that comes up.
12:47I'm making new files. I'm maybe working on new projects. I'm scrapping old projects.
12:51So things move around a lot. So the reason why I'm spending so much time there is just so so you feel more comfortable. You know, don't stress yourself out.
12:58I'm doing this the wrong way. I I truly don't think there's a single wrong way. The only time where you run into issues is if there's so much context and it's so unorganized that you can't find it, you can't find things manually, and that your AI can't find things.
13:10Because this is laid out in a way where to me it's very clear. I have decisions. I have audits.
13:14I have my archives. I have other worlds, which literally means these are entire full Cloud Code projects that I can open up in their own, and I use them on my own. So my scheduled automations, my YouTube OS, I've got the book that I'm working on.
13:27All of this stuff is other worlds, and if I wanted to add more Cloud Code projects in my other worlds folder, I could because it's nice that my main OS can look through this kind of stuff. But here's the thing. Even if I have other ClaudeCode projects, which trust me, have way more than just these four that live on my desktop or live in my documents, ClaudeCode can still get there.
13:46And it still knows they live there because I I have, like, documentation on, okay, you know, the ClaudeCode GitHub repo that owns, you know, like, my website, that lives at desktop slash blah blah blah. And so ClaudeCode can go find it if it ever needs it.
13:59You can see how many products I've gotten here. Right? Like, for example, this YouTube video, if I can just scroll down here and find it.
14:05Oh, yeah. Right here. In my YouTube videos folder, this one was called Opus 4.8 operating system, and all the diagrams you guys just saw right over here, right, in Excalidraw, all of these diagrams and the whole outline of this video was built for me in here by my Cloud Code OS, my outline, my visual plan, all of this, all these other YouTube videos, all these x articles, everything that I'm doing, I'm doing through here.
14:27And that's where I've got all the stuff set up because think about it. Now this thing has access to basically everything I've done.
14:35So it can update my skills later. It can make its memory files better and better. And, yeah, I think you'll find just how helpful it is if you're constantly context switching and doing a lot of stuff throughout your day, which I'm sure all of you guys are.
14:47Having just, like, one source of truth that has everything, and you kind of eliminate that scavenger hunt of, oh, where did I leave that file? Who did I send that to?
14:56Did I do that on chat GBT or Claude Code or Claude? Like, where did I do all this stuff? No.
15:01Just do it all here and give your system access to touch everything because then it can find things. Right? Like, I remember someone on my team sent me something the other day, and I couldn't remember if that was in Slack or ClickUp.
15:12So I just came in here and said, can you help me find this doc from this person? And it just found it in, like, ten seconds. So those types of use cases where you're not doing the scavenger hunt trying to find something, it's huge.
15:23Now when I start to bring that concept up, there's probably alarm bells that go off in a lot of your guys' heads, which I think if that happened to you, I'm glad. Like, that's a good thing that happened because the more autonomy you have, the more reach you have, as you move your way up sort of like the AI systems pyramid that I talk about, you know, like workflows, AI workflows, AI agents, teams of agents, as you move up, typically, risk goes up as well as cost.
15:47So that's why I like to talk about your keys. Right?
15:51Like, permission layer around the AI agents. Thinking about if you're building an AI operating system that has so much data and it has so many different API keys or MCP servers, you have to be careful.
16:03There was an example, which you can see this is kind of built around on our team, like, real. An AI agent basically sent out three promotional emails that weren't supposed to go out to over a 150,000 inboxes.
16:16Like, it was bad. We had to apologize. We had to, you know, take down the page and whatever.
16:21But why did that happen? Because it's not like our team said, hey.
16:26Go send out these three emails. What happened was the agent proactively picked up a to do list, a task, and and it interpreted it as, okay.
16:33I need to make these emails and send them off, and it just did it. And, basically, the mindset shift there is you have to think about what can your system actually touch?
16:43You know? You have to assume that if your agent has access to read something or to do something, it will do it. And, like, that's not always the case.
16:51Right? Most of the time, it won't. But if you assume that it will, you change the way that you give your agent endpoints.
16:58You change the way that you give your agent MCP servers. You you start to have things that are scoped. You start to just be more careful about it, and obviously, that's the right way to do it.
17:06And I also wanna clarify, like, I was not mad at, you know, this person that when when this accident happened. Like, it happened. It was a really good learning experience for everyone on the team, myself included, and, you know, we grow from it.
17:18But anyways, as you start to just give your agent more and more connections and more capabilities, you definitely wanna be thinking about that because it's exciting. It's exciting to connect to a bunch of different data sources.
17:28It's exciting to build all these new skills, but you also have to bring yourself down to earth a little bit and think, okay. What is the worst case that could happen here?
17:36And the thing is instructions are not the same as capabilities. If you think about , there's a difference between saying, hey. Don't ever use that key and saying, okay.
17:47Give me that key. Like, you don't have you don't get to put this key on your key ring. It's a huge difference.
17:51So as much as you could say never send emails, if there's a send email key or send email tool inside of that agent harness, then it could do it.
17:59Right? Like, it it it actually physically could. And so that's why I like to think about it as the bike method.
18:05When you're building skills and when you're building automations, think about it like you're teaching a kid to ride a bike. Basically what I mean by that is you don't just hand the kid a bike, put a helmet on him, and say, okay, kid, go ride.
18:16You typically will walk with them. You'll hold the handle, you'll put your hand on their back, you'll walk with them, you'll feel that they're adjusting too much to the left or leaning too much left, and you'll adjust them back to the middle. You will help them out throughout the way, and every single time you use that skill or every single time you you and the kid like go up and down the driveway, it gets better.
18:36Like slowly it gets better and better and slowly there's more trust and slowly you can take off your hands, slowly you take off the training wheels. Slowly you let the kid just ride down the street and all you do is watch. You know?
18:46You probably don't immediately go inside and take a nap while the kid's riding a bike, but you still watch, and you make sure that things are feeling good and it's earned its spot. Basically, like, it's earned its next phase until the point where you get to autonomy.
19:00And I think, yes, the barrier to entry is getting lower to build systems. Yes. It's easier to evaluate and easier to push something into production same day.
19:09Whereas, you know, earlier on without all these AI models, it was harder to build automation so quickly and so accurately right away. But making it easier shouldn't be giving you that false sense of security.
19:22It shouldn't be, um, actually putting too much trust in your head because too much trust is bad if you are putting something out there too quick and it's not ready. So that's another thing to think about, phase trust in the bike method.
19:35Every time you run a skill, it gets better. It's not a waste of time. So you can see here in my dot Claude, if I open up this folder, I've got, you know, agent memory.
19:42We've got agents. We've got some plans, some rules, and here's my skills where you can see I've got quite a bit in here.
19:48Right? There's a lot of different skills that I've built out in here, and these have built obviously over time.
19:53These get changed all the time. Some of them I've even moved, like, globally instead because you can have skills that live only in a Herc two project in this local project, or some of the skills I've made I moved globally so that if I'm ever working in any other directory, I can still use those skills.
20:07But anyways, lots of skills here. I'm obviously not gonna dive into these because, like I said, there's just a lot. But let me talk about now how do you get to a place where you're building skills.
20:15So I think that you should first of all think about it like this. Think about your day. Think about your week.
20:20What are things that you do on a cadence? Right? What are things that you do every Monday, every single day, multiple times a day, whatever it is?
20:25What are things that you know you do often? And then build it forward. So say, hey, Cloud Code.
20:30I want you to use the skill creator. I want you to help me run this skill. Here's the end goal.
20:34Here are typically the different tools and the different things I think about. Let's try building it out. And it's gonna walk you through.
20:39You correct it. You get an output. You give feedback.
20:42Right? And you just kinda take that iteration loop of building the skill. And sometimes your skills might take, like, 50 tries until you get to a place where you like it.
20:49And then even then, every time you use the skill, you're gonna evolve it. Right? Like every single time I use my LinkedIn writing skill, I give it feedback.
20:55So every single time I write a LinkedIn post, I say, hey. This was good, but this wasn't as good. So change that in the skill so that next time that doesn't happen, and I just keep evolving it.
21:05Now the other way is when you reverse engineer a skill, which honestly is what I do more often. I do something with Cloud Code from end to end, and then I think to myself, okay, that would be a good skill to have just in case I do that again, or because I know I'm gonna do it again. So I build the thing end to end, and then once I have that finished output, I reverse engineer and say, hey.
21:22This is a really good output. Look back at our conversation. What did we do to get there?
21:27What did you think about? What tools did you need? What questions did you ask me?
21:31And build a skill around it that we wanna have that skill give us that output. Obviously, it's not gonna be perfect once again, but that's the way that I typically build them as I reverse engineer them. So I was watching this section of the video back, and I realized that this isn't completely true.
21:44Like, for the most part, yes, this is exactly how I do it, but I don't want to plant in your head the idea that a skill is only, like, a workflow or something that you do that's, like, a big process, like an SOP. Right? Like, skills can also be just as simple as, have you ever typed in this prompt before, and do you not wanna type that prompt in again?
22:04Then build a skill around it. And I think a really, really good example to show you guys is my session handoff skill. So, you know, this is a, you know, a Cloud Code session.
22:12I've been chatting with Cloud Code for a while. I ran a goal, and now I want to be able to maybe clear the context, or I wanna move this over from Cloud Code to Codex, or open up a new terminal, whatever it is.
22:23And so I built this skill right here called session handoff, and I installed this one globally for Cloud Code for me. So when I run this, it basically is giving me a full breakdown of, hey.
22:34Here's what we did. Here's the files that were created. Here are open decisions.
22:37Here's what's next. And that's really helpful for me because it's not a super complex skill. Right?
22:41It's just a prompt. But I was typing that myself every time. I'd say, hey, you know, I'm gonna clear my context.
22:46Can you give me a summary? Can you tell me what decisions are left to do? Can you tell me where we need to pick off or pick up?
22:52And I was just repeating that same prompt so many times a day. And so I realized, okay, why don't I just take that prompt and just put it as a skill?
22:59Put it as a slash command. And now I have this. I can go ahead and do a slash copy slash clear, paste it in, and then I'm basically right back where I was with completely fresh context.
23:09So I will also leave this exact skill in my free school community if you guys wanna grab that as well, or you can just build your Right? As you can see, it gives me pickup here, deferred open questions, where it started, decisions locked, key files, all of that.
23:23Super, super helpful. So just wanted to throw that in there as well. And you really wanna be thinking about your AI operating system as a mentor rather than just, like, you know, a chatbot or an automation.
23:36So what I mean by that is when you start to think about, like, I wonder if that's possible or I don't know how AI could do this, it's it's tough.
23:45Because when you have that idea or that feeling of doubt, your brain wants to default to what's comfortable, which is what you already know how to do. Which means, let's say you wanna pull some reports and do a data analysis for the quarter.
23:56If you've done that for every quarter manually by going into the software and pulling the report and then doing your formulas, your brain's gonna wanna do that because you know exactly how it works.
24:06But if you think of your AI operating system as a mentor and you say, hey. Here's a process that I do, you know, at the end of every month. Here's the tool that I need to use.
24:14How is this possible? Like, how can I actually do this? And it'll walk you through the options, and it will start to give you ideas, and it will test things out with you.
24:24And then you get to a point where you have that done. And it is tough because with AI, with building skills, with anything that you do new in your business or in your day to day productivity, there is basically a short term cost that you have to bear.
24:37Right? Like, the idea of learning something and building out a new skill, that one day, it might be slower than your manual process.
24:45But in the long run, having an automation around something is obviously way way quicker. So you kind of have like a 20% dip, and it's basically is that change, is the dip worth it for the long term climb and the long term upside that you will have rather than just remaining constant how you would have if you kept that process the same and didn't change anything.
25:03But of course, your judgment still needs to stay there. You still need to be the one who's reading everything. You still need to be the one who is putting your own spin on it.
25:11You can't just outsource everything. I think a really good quote that I've brought up multiple times is that you can outsource your thinking, but you cannot outsource your understanding. Really powerful quote.
25:20And then I know I showed you guys this earlier. Right? Like, the actual Obsidian thing where you can see the visualization of all of my files.
25:28This is cool. Right? And to be honest, for my Herc two project, I don't really use this too much.
25:33I think that when it comes to visualizing your actual operating system, it's basically just personal preference. Right? Like, you could open up the terminal and run Cloud Code from Obsidian, and you can, you know, do that if you want.
25:44I've got a few Obsidian projects where I have all my YouTube transcripts in there, and I've got a bunch of tools, and I've got different breakdowns, and sometimes it is really nice to see that visually. But, like, for my HER2 project, I don't really know what I'm looking at here, and I don't really care. And the thing about my OS is it's how I work, and it's not very visual.
26:00And so I know a lot of people might think, oh, well, to have an AI operating system, you need, like, a fancy looking dashboard. You need to see all your agents. You need to see all this uptime.
26:08You need to see these things. I don't. I don't care.
26:11If you do, then that's perfectly fine. There's nothing wrong with that. But I don't need a dashboard for my operating system.
26:17Right? Like, because I come in here and I talk to different agents. I have different tabs.
26:21That's how I work. That's how I'm really productive. I was thinking about building one, but I thought to myself, okay.
26:26If I had a dashboard, what would I wanna see? And when I started to think about what I wanted to see, there really wasn't anything that to me was too important that I can't see already, or I can't just have Cloud Code pull in the data for me. So if you want a custom dashboard to build, certainly grab Cloud Code, have it build you a dashboard.
26:43But for me personally, I just honestly don't see too much value in that. The way that I like to think about if I add features to my AI OS or if I add a new skill to my brain is I think about my North Star.
26:55What is my ultimate goal at the end of each month that I wanna say that I made progress on? And if doing something, if spending time on something doesn't actually move me closer to that goal, then I'm not gonna do it.
27:06And so I think, like, if you start to think about metrics, right, like, what are what are the metrics that are important to me? Maybe that's my free school members. Maybe that is our monthly recurring revenue.
27:16Does having a a dashboard, a pretty visual AIOS dashboard that shows me these metrics going to improve the metrics? Some could argue yes if you're, like, a visual person and you need that to help with your decision making and your brainstorming.
27:30But the fact that the metrics exist and the fact that I can pull those in immediately, that's what actually matters to me. So just wanted to sort of, like, throw out there.
27:38Once again, I don't think that there's anything necessarily wrong with that. I don't think that there's a right or wrong answer. I think a lot of people are making up their own AIOSs, and however they wanna work is perfectly fine.
27:48I think that's the beauty of it. But just think about productivity.
27:53Productivity isn't how many hours did I work today. Productivity is, did I actually move the needle closer to my goal?
27:59And that's how you could stop getting so overwhelmed with all these AI tools and, you know, switching between things. That is the beauty of it. So anyways, guys, I know this one was a little bit shorter, but I wanted to keep it short.
28:10I wanted to keep it very, very, like, mindset oriented because I think the mindset stuff about the way that you use this, the way you set it up, and the way that you actually feel ROI is really, really important. But just remember in my free school community, you can come in here and get that full three hour course on how you actually build this out in Cloud Code, and you also can use this exact free GitHub repo, which I will link in the free school community as well in order to just clone this in, onboard into your AIS OS, and then just start pumping that thing full of those four c's we talked about, which were once again, hopefully, you guys remember, context, connections, capabilities, and cadence.
28:44So, anyways, that is going to do it for this one. I hope that you guys enjoyed, and I hope that you learned something new. If you did, please give it a like.
28:50It helps me out a ton. And as always, I appreciate you guys making it to the end of the video, and I will see you all in the next one. Thanks, guys.
The Hook
The bait, then the rug-pull.
When a new model drops, most people benchmark it. Nate Herk wired it into the thing he runs his entire business through. The operating system framing is not a metaphor — he means the first tab he opens every morning, the tool that reads his calendar, his Slack, his QuickBooks, and his YouTube transcripts, and the interface through which he does essentially every task that used to require switching apps.
Named ideas worth stealing.
06:31acronym
The Four Cs
- Context
- Connections
- Capabilities
- Cadence
A layered architecture for building an AI operating system. Each C depends on the one before it.
Steal forAny pitch or course on building AI into a business workflow
17:39model
Instructions vs Capabilities
Instructions are suggestions an agent can override; capabilities are physical access constraints. Remove the key, do not just write a rule.
Steal forAny safety briefing for a team deploying AI agents with real API access
18:02model
The Bike Method
Phase trust as the skill earns it: walk alongside every run, then hands off the handlebars, then watch from the porch, then full autonomy.
Steal forAny framework for delegating to an employee, contractor, or AI agent
08:50list
Seven Connection Tiers
- Revenue data
- Customer data/comms
- Calendar
- Internal comms
- Tasks/project management
- Meeting recordings
- Knowledge base
A starting audit checklist for identifying which data sources to connect to an AIOS first.
Steal forAIOS onboarding checklist or consulting deliverable
How they asked for the click.
VERBAL ASK
28:08link
“In my free school community you can come in here and get that full three hour course on how you actually build this out in Claude Code.”
Soft sell delivered twice — once mid-video at the four Cs introduction and again at the close. Free community as CTA rather than a paid product. GitHub repo as second pull.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
AFFILIATECommission earned if you click.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
