Section 1
Stop Chasing Models. Build the Thing Underneath.
Here's the trap. Every few weeks there's a new model that's supposedly going to change your life. You switch, you relearn, you start over. Then the next one drops and you do it all again. You're running on a treadmill.
The people getting real leverage out of AI aren't doing that. They built one thing that doesn't change: a system that holds who they are, what they're working on, and how they want AI to operate. Then they just point whatever the best model is at it.
The model is the engine. Your system is the car. Engines get better every month. You don't throw out the car each time. You drop the new engine in and drive.
That's what this guide builds. Not a trick for one model. The foundation that every model from here on plugs into.
Section 2
The One Rule That Makes It Model-Proof
It's a folder, not a Project.
This is the part almost everyone gets wrong. They build their context inside a Project (in ChatGPT, in the Claude app, wherever) and feel organized. But Projects silo your information. Each one is a sealed box. Your AI can only see inside one at a time, and it can't connect anything across them.
A folder is the opposite. Everything lives in one place, all connected, all visible at once. You point Claude Code or Claude Cowork at that folder and it reads the whole thing: who you are, every project, every note, all of it together.
And here's why it's model-proof: a folder isn't tied to any one tool or model. When the next great model shows up, you don't migrate anything. You just point it at the same folder. The system you built stays. Only the engine changes.
A Project
Sealed box. One at a time. Can't connect across them. Tied to one tool. You rebuild when you switch.
A folder (this system)
One connected brain. All visible at once. Any model can read it. You never rebuild. You just point the next model at it.
Section 3
What You Need
Two things, both free, no coding.
A folder on your computer
Just a regular folder full of plain text notes. I use Obsidian to view and navigate mine because it makes the links between notes clickable, and it's free. But it's still just a folder. You don't need Obsidian to start. Any folder works.
Claude Code or Claude Cowork
Either one can point at a folder and read everything in it. That's the whole reason we use a folder instead of a Project. New to Claude Code? Start with Claude Code Day One. Prefer Cowork? Set Up Cowork Like the Pros.
Section 4
The Prompt That Builds It For You
This is the whole thing. Make an empty folder, point Claude Code or Cowork at it, and paste this in. It interviews you about your actual work, then builds your system around your answers. It won't just dump a generic template on you. It builds the version that fits your life.
Paste into Claude Code or Cowork (pointed at an empty folder)
You're going to build me a personal AI system: a single folder on my computer that any AI model can point at and instantly understand my work, my context, and how I want it to operate. We're building one connected second brain, not siloed projects.
First, interview me. Ask these one at a time and wait for each answer before the next:
1. What do you do day to day? Your work, your business, what actually fills your week.
2. What recurring projects or clients are on your plate right now?
3. Who are the people you work with or reference most often?
4. What do you most want AI to help you with?
5. How should AI talk to you, and what should it always know about you?
6. What do you keep losing track of that you wish your AI just remembered?
Once I've answered, build this folder structure and fill it in with what I told you:
- CLAUDE.md (at the root): my operating system. Who I am, what I'm working on, how I want you to work with me, and the rules you always follow. The first thing any AI reads.
- USER.md: my background, role, voice, and preferences, in depth.
- MEMORY.md: a running list of facts worth remembering across sessions. Start it with what I just told you.
- 00-Inbox/: where I dump raw notes, ideas, and captures.
- 01-Maps/: one navigation note per big area of my life or work, each linking out to everything related.
- 02-Reference/: knowledge I want to keep and reuse.
- 03-Projects/: one folder per active project.
- 03-Templates/: reusable note templates for things I make often.
- 04-Archive/: finished stuff, out of the way.
Wikilink everything. Whenever you mention a person, project, or topic that has its own note, link it with [[double brackets]] so the whole brain connects to itself like a wiki. When I add a new note later, suggest the links back to what already exists.
Then write a short START-HERE.md explaining how I use this: I point Claude Code or Claude Cowork at this folder, and it reads CLAUDE.md first every time.
One rule for you: never put this in a Project. Projects silo information so only one is visible at a time. This is a folder so every model, today's and next month's, can see all of it at once.Section 5
What It Builds (And How Mine Is Set Up)
This is the same skeleton I run my entire brand on. I won't bore you with my private files, but here's the high-level shape so you know what good looks like, and why each piece is there.
CLAUDE.md: the brain's operating system
The single most important file. It's the first thing the AI reads every session: who you are, what you're working on, how you want it to behave. Get this right and every model you ever point at the folder shows up already knowing you.
USER.md and MEMORY.md: who you are, what to remember
One holds your background and voice in depth. The other is a running memory of facts worth keeping across sessions, so your AI stops forgetting things the moment a chat ends.
Numbered folders: capture, navigate, do, archive
An inbox for raw capture, maps that act as navigation hubs, a reference library, a folder for active projects, templates for things you make often, and an archive for what's done. The numbers just keep them in the right order.
Wikilinks: the part that makes it a brain, not a file dump
Every note links to related notes with [[double brackets]]. This is the move Andrej Karpathy popularized: instead of folders alone, your notes connect to each other like a wiki. Over time it becomes a web your AI can actually think across, not a pile of documents.
The prompt above blends two things: the structured folder setup I use to run a real operation, and Karpathy's self-connecting wikilink approach. You get the organization of one and the connectedness of the other.
Section 6
How To Actually Use It
From now on, you point Claude at the folder instead of starting from scratch. It reads CLAUDE.md, knows your whole world, and gets to work. A few prompts to start with:
Let it work from everything you've got
The core move. It pulls context from across the whole folder, not one sealed Project.
Read my CLAUDE.md and everything in 03-Projects, then help me with [whatever you're working on].Capture something and let it file itself
Drop a thought in the inbox, then have Claude sort it into the right place and link it up.
Here's a raw note: [paste it]. Save it to the right folder, give it a clear title, and wikilink it to anything related that already exists.Keep the brain learning about you
End a session by having it update your memory, so next time it's even sharper.
Before we wrap, update MEMORY.md with anything important you learned about me or my work today.The next "best AI ever" is already on its way. So is the one after that. If your whole setup lives inside one tool or one Project, you start over every time. If it lives in a folder, you don't. You just point the new engine at the same car and keep driving.
Build it once this week. Then stop chasing models and let them come to you.
Make the folder. Paste the prompt. You're already ahead.
Work with Me
Want me on your problem for an hour?
Book a 1:1 call and we build the fix live — the workflow you want automated, the tool you can’t crack, whatever’s stuck. Direct, hands-on, no pitch waiting at the end.