Most AI tools have a memory now. Claude remembers things about me between chats. So does ChatGPT. That's real, and it helps.
But it's their memory of me — built automatically, tucked away where I can't see it, and locked inside one product. What ChatGPT picked up doesn't follow me to Claude. What either of them decides to remember, I don't really get to steer. And every time I opened a fresh tool, or a new kind of session, I was back to typing out who I am and what I'm working on.
What actually changed how I work wasn't leaning harder on that built-in memory. It was the opposite: I started keeping a file about myself that I wrote and control.
Where it started: a file the tool reads every time
I stumbled into this inside Claude Code — the coding tool I use most days. It reads a file called CLAUDE.md at the start of every session automatically. Think of it as a standing briefing note: whatever's in there, the AI has read before it answers your first question.
So I started putting things in it — on purpose. Nothing lands there on its own; the AI won't quietly decide to remember something for you. I either open the file and type, or I tell Claude "save that" and it writes the line. How I like to work — direct, ship fast, show me the simpler path. What I was building. Little corrections I was tired of repeating.
And that deliberateness turned out to be the feature, not a chore. Every session got sharper. The AI stopped handing me generic advice and started building on what I'd actually told it. I wasn't reminding it who I was anymore; it already knew, because I'd written it down once in a place it always looks.
That was the whole lesson, and it was smaller than I expected: the useful move isn't hoping the AI remembers me. It's writing down who I am, on purpose, in a file I own.
Then I got carried away (on purpose)
Filling in that one file taught me how much leverage was sitting there — so I went further and built a whole second brain out of the idea: linked notes on every project I've shipped, the skills I'm working on, decisions I've made and why. One map file at the front points to the rest, one line per note.
I mapped 34 projects into it. It felt like a lot. It was a lot. But once it existed, I could test whether it actually worked.
The test that told me it was real
I opened a completely fresh AI session — no history, nothing but those notes — and quizzed it about me like a stranger cramming before a meeting. Who is this person? What has he built? Why is he doing it this way?
It answered 6 out of 10 questions fully — pulling the right notes, in one or two hops, and citing where each answer came from.
The four it missed weren't the system breaking. They were things I simply hadn't written down yet — and it flagged them as gaps instead of inventing an answer. That was the tell: the memory was exactly as good as what I'd actually fed it. The hard part was never the software. It was writing down what I know about myself.
What I'd tell you to do this week
Skip the second brain. Start where I actually started — one file.
Open a note, or make a plain text file, and put in it:
- Who you are — your work, your field, the thing you're genuinely good at.
- What you're working on right now — the current project, the current problem.
- What you've done before — a few real things you've built or shipped, so it has receipts to draw on.
- How you like to work — short answers or full reasoning? Blunt or gentle? Say it plainly.
Then paste it in as your first message before you ask an AI for anything — or, if your tool has a "custom instructions" or memory box you can edit, put it there so it loads every time.
You'll feel the difference in the first reply. The advice gets specific to your situation instead of a generic one. And the file compounds: every time you catch yourself re-explaining something, that's the thing to go add to it.
The reason a file you write beats the memory the AI keeps for you is simple — yours is portable (it works in any tool), legible (you can read and edit exactly what it knows), and deliberate (you decide what goes in). The built-in kind is none of those.
Where I'm headed next
The thing I'm still figuring out: making that memory follow me everywhere without me lifting a finger — same context at my desk, on my phone, in a browser, in whatever tool I happen to open — so I never introduce myself to my own AI again.
That's the next hill I'm climbing, and I'll write up how it goes. But the part that already changed how I work was embarrassingly simple: I wrote down who I am, once, in a file I control — and stopped making the AI guess.