How I stopped retelling everything to AI with a second brain

Every new conversation with an AI starts with amnesia. It doesn't know who you are, what project you're on, how you write, or what got decided last week. You explain everything, it gets half of it right, you correct it and explain a bit more.

Getting real work out of AI depends less on the model and more on what it finds when it shows up.

A notebook the AI can read too

Second brain is the modern name for an old idea: take what's worth remembering out of your head and keep it where you can find it later. Niklas Luhmann did this with paper index cards for decades and called his card file a conversation partner. Seventy years later, the partner talks back.

Mine lives in Obsidian, a notes app that saves everything as plain text files in a Google Drive folder. No closed platform. It sounds like a detail, but it's what lets Claude open the same file I open to read, and get to work.

Everything in its place

The structure follows PARA, a four-drawer method: projects with a deadline, ongoing areas of life, reference material, and archive.

For me, it's organization. For the AI, it becomes house rules. At the top of the vault there's a note any AI reads before touching anything: where each kind of content goes, how to name it, what never to do. A portfolio case, for example, doesn't end up as a loose note. It becomes a document with a fixed structure: context, problem, decisions, outcome, what I learned.

I wrote that rule once. Since then, it applies to every conversation.

How things get there

On my computer, Claude loads the rules on its own when it opens the folder. On my phone, I set up a project in the Claude app with the same rules and Drive connected: I type or dictate an idea on the way somewhere, and it shows up filed in the right place.

Some of the content arrives without me. Scheduled routines bring in running, health, finance and site traffic data every day. When I open a conversation about training, the history is already there.

What changes day to day

The first gain is not having to retell everything. One command saves a decision in the right place, another records where I stopped, and the next conversation, on any device, picks up from there.

Quality goes up with it. Before writing anything in my name, Claude reads a voice guide built from about 140 of my real messages, plus a list of AI writing habits to avoid. The first draft already sounds closer to me, and I correct less.

And it compounds. Every note written for one conversation serves all the ones after it. A voice guide from June reviews articles in October. Knowledge that used to vanish into chat history now earns interest.

How to set up yours

You can start in an afternoon and adjust as you go. If you want the idea behind it first, Andrej Karpathy described the same pattern in LLM Wiki, a markdown knowledge base the AI itself maintains.

1. Create the vault in a synced folder. Install Obsidian and point the vault to a Google Drive, iCloud or Dropbox folder. That way it lives on both your computer and your phone.
2. Set up the four drawers. Projects, Areas, Resources and Archive. Nothing else is needed at first.
3. Write the rules note. At the root of the vault, a short note: who you are, where each kind of content goes, how to name files, what the AI must never do.
4. Connect the AI to the folder. On your computer, open Claude Code in the vault folder and, in a CLAUDE.md file, ask it to read the rules note before anything else. On your phone, create a project in the Claude app with the same rules and Google Drive connected.
5. Save at the end of every useful conversation. When a decision or a lesson comes out of it, ask the AI to file it in the right place. That's how the vault grows without extra effort.
6. Review now and then. Once a month, merge duplicates and delete what went stale. The AI helps you find them, the call is yours.

What doesn't run by itself

The system needs upkeep. I've had the same text in three copies without knowing which one was right, and documentation describing things that no longer existed. The AI follows the rules you wrote. Vague rules get vague results, same as a brief.

It doesn't replace reading either. AI delivers in proportion to what you know. The second brain just makes sure that doesn't get lost between one conversation and the next.