What is a second brain for AI?

A folder of plain text files that holds who you are, how you work and what you are working on, which every AI agent reads at the start of a session instead of you re-explaining it.

Where should the folder live?

Anywhere synced across your devices and readable by your agents: Google Drive works, and a private git repository is even better for code agents.

What goes in the Brain versus the Context?

Brain is stable: identity, rules, skills and preferences. Context moves: one folder per organisation, project and dated work session, each with a log.

Does it work with any AI tool?

Yes. Claude, a code agent or a phone app all read the same files, so you are not locked into one tool.

How do I keep it useful?

Every time you repeat yourself to the AI, write the answer down once in the right file. Keep secrets in a password manager and reference them, never paste them into the files.

← How-to/build-a-second-brain-for-your-project-work

Build a second brain for your project work

I stopped re-explaining myself to every AI by giving it one folder it reads first — my context, on tap.

▶ 3 min·Jun 2026·Business operation · Project management · AI·●●○ lvl 2·Free

By Jin Longyu·Published 8 Jun 2026·Updated 21 Oct 2026

TL;DR

Every AI chat used to start from zero. The fix is one synced folder, in Google Drive or a git repo, that the agent reads before every chat: a Brain with your identity, rules and playbooks, and a Context tree of organisation, project and dated work sessions with a log. Set it up once and any agent on any device picks up your work where it stood.

01 · Story

Story behind

Every AI chat, I used to type my context from zero.

I'd re-explain my business, how I work, the project I'm on.
The agent was sharp for five minutes.
Then I closed the tab,
and the agent forgot everything.

So what if it could read all that before every chat,
without me typing it again?

My Google Drive "second brain" solves it.
One synced folder the agent reads first, in two parts:

Brain — my rules, skills, and preferences
Context — the client, the project, the work at hand

Now I set it up once.
Since it's one drive, synced to my office and home laptops,
the AI picks up my work any time, on any device.

02 · Why

Why this helps you

Never brief the AI twice — your identity, rules, and project context load automatically. You start every session already understood.

Any agent, same context — Claude, a code agent, a phone app: they all read the same folder. You're not locked into one tool.

Resume cold — open a project after a month and the work log tells the agent, and you, exactly where things stood.

It compounds — each correction you capture makes the next task faster. The system improves as a by-product of using it.

No new software — it's just folders and text files in storage you already pay for. Nothing to host, nothing to break.

03 · Steps

How to

Grab both from "Try it yourself" — the second-brain setup and the log-session skill.

Connect one synced folder — I use Google Drive — that the agent reads on every device. No API needed.
Give it context about you and your work so it fills the files instead of inventing them.

Let it draft the structure, splitting the stable "Brain" (who you are) from your "Context" (what you're on).

Write who you are and how you work — your role and your rules.
Add your playbooks — the tasks you do the same way each time.

Guide the AI through your work context:
- Organization: a client or company — who they are, the relationship.
- Project: one body of work under it — its goal, scope, current focus.
- Work: a single dated session — what you did and decided that day.
Let it build the folders in that shape, with the right files at each level:
- Log: what you worked on (e.g. 2026-05-07, redesigned the client's site to the new brand).
- Environment: the ID references (e.g. which Notion page holds the client's brand assets).

code

Google Drive/
└─ Organization/
   └─ Project/
      └─ Work/
         ├─ 2026-05-07 Client site redesign
         └─ ...

Run the log-session skill at the end of a chat — it does the write-up for you.
Capture three things:
- Work done: a session summary in the Work layer.
- Environment: the setup IDs it touched (secrets stay in 1Password).
- Lessons: anything reusable, flowed up into the Brain.

Hand the agent the latest work log next time — it reads back up the three layers and continues where you left off.

04 · Notes

Good to know

Keep secrets out of it — store API keys in a password manager and reference them. Never paste real credentials into the files.

Stable vs moving — if you're editing your Core file every day, that's probably project context in the wrong place. The brain should change slowly.

Keep it alive — every time you repeat yourself to the AI, write the answer down once. That habit is what makes the folder smarter each week.

Start tiny — a single Core file already beats nothing. Add Rules and Skills only as real patterns appear; don't build the whole cathedral on day one.

05 · Try it yourself

Copy the skill

Set up your second brain

Paste it to your agent as the first message.

Log my session

Paste it to your agent as the first message.

06 · FAQ

Questions, answered

What is a second brain for AI?: A folder of plain text files that holds who you are, how you work and what you are working on, which every AI agent reads at the start of a session instead of you re-explaining it.
Where should the folder live?: Anywhere synced across your devices and readable by your agents: Google Drive works, and a private git repository is even better for code agents.
What goes in the Brain versus the Context?: Brain is stable: identity, rules, skills and preferences. Context moves: one folder per organisation, project and dated work session, each with a log.
Does it work with any AI tool?: Yes. Claude, a code agent or a phone app all read the same files, so you are not locked into one tool.
How do I keep it useful?: Every time you repeat yourself to the AI, write the answer down once in the right file. Keep secrets in a password manager and reference them, never paste them into the files.

07 · Keep going