A company brain every AI on your team reads from

What is a company brain, and how is it different from a second brain?

A second brain is personal. It’s the notes app where one person keeps what they read and think. The term comes from Tiago Forte’s method, and it works for one head.

A company brain is the same idea with two changes. Many people write to it, and most of its readers are now machines: Claude, ChatGPT, Gemini, a coding agent, a bot in Slack. Each one starts every conversation knowing nothing about you.

So the job is not storing notes. The job is that every one of those readers starts from the same facts, and that the facts are still true.

Why does every AI chat start from zero?

Because your context lives in ten places and none of them is the answer. The decision was in a Slack thread. The client brief is a Drive doc with three versions. The price change went out by email. The AI sees none of it unless someone pastes it in.

Each tool also built its own memory. ChatGPT has memory and projects, Claude has projects, Gemini has its own. All useful, and all separate. Switch tools, or let a colleague open a new chat, and the context stays behind.

• The cost is not the AI being wrong. It’s the AI being right about an old version of your company.
• The second cost is you. You become the copy and paste between your tools, every day, for every chat.

Where should the single source of truth live?

Somewhere every reader can open, where every change has a date and an author, and where a wrong edit can be undone. For people only, Notion or Drive does that. When several AIs write to it, I use Git.

$ ls company-brain/CLAUDE.md      rules every session loads first, under 150 linesAGENTS.md      the same rules, for Codex, Cursor and Copilotdecisions/     one numbered file per decision, with its whyclients/       a folder each: plan, voice, what was promisedmemory/        one fact per note, datedinbox/         raw material from Slack, mail and Drive$ git log -1 --format=%sD-028: weekly client report moves to Monday 

• Plain text. Every model reads Markdown. No export, no connector, and no vendor in the middle.
• History by default. Each line knows who changed it, when, and why. It’s the question Google Ads change history answers for an ad account, answered for your company.
• Review before it counts. A change to the rules can go through a pull request, like code.
• One file loads first. Claude reads CLAUDE.md at the start of every session. Codex, Cursor and GitHub Copilot read AGENTS.md. Same rules, whichever model is on duty.

The rule I hold it to is one fact, one place. Everything else is a pointer. When the same fact lives in two files, one of them is already wrong.

How do Slack, Outlook and Drive connect to it?

Through connectors. Claude and ChatGPT both connect to Slack, to Microsoft 365 for Outlook, Teams and SharePoint, and to Google Drive. Most of these run on MCP, the open standard explained in what is MCP.

The trap is to connect everything and call it a brain. A connector lets the AI read the Slack thread. It doesn’t tell it the thread was overruled two days later.

So the connectors feed an inbox, and the brain keeps the outcome. Raw material lands in inbox/. What is true moves to its file, with a date. The thread stays in Slack, where it belongs.

Before you switch a connector on, read what it can write, not only what it can read. Some only search. Some draft and post.

I don’t run my own work on Slack or Outlook. Mine comes in through Notion, Gmail, Google Drive, Search Console and my own ad-data connector. The pattern does not change with the tool.

Can several AIs share one brain without stepping on each other?

Yes, if they share the rules and not the desk. This is how mine runs on 29 October 2026:

• Three Claude setups on three computers. One runs the ad accounts, one builds the software, one runs the websites and SEO.
• Eleven project chats, one per website plus an assistant. All of them load the same rules file: 102 lines, with a hard cap at 150.
• One shared memory of 311 notes, one fact each, every one of them dated.
• 27 numbered decisions. If a change has no decision entry, behaviour does not change.
• A ticket desk in Notion where the three ask each other questions. What one AI writes is information for the others, never an order.

The repository behind the web side took 691 commits in its first fifteen days. That’s what a living brain looks like: small corrections, every day, each with a reason.

What breaks, and what did I change after it broke?

• A rule remembered is not a rule read. An AI that knew three of my rules broke all three in one session. The rules that matter now run as checks, not as text.
• Long rules files stop working. Past a couple of hundred lines, the important line drowns. Mine has a hard cap, and what does not fit moves to the file where it belongs.
• Memory goes stale. A note that names a file was true the day it was written. The top of my memory says it plainly: check before you trust it.
• Two AIs in one folder overwrite each other. One session once committed another session’s half-finished work. Now each one works on its own branch and folder.
• Read-only is not a perimeter. Agents told to only read still read the secrets in the folder. Connectors get the narrowest scope that does the job, and keys never live in the brain.

How do you start one this week?

1. Pick the home. A private Git repository, on GitHub or your own server. One per company, not one per tool.
2. Write the rules file first. Who you are, who your clients are, how you write, and what nobody may do. Under 150 lines. Give the same file to the other models as AGENTS.md.
3. Add one folder per client. The plan, the voice, and what was promised and when.
4. Start a decision log. One numbered file per decision, with the date and the why.
5. Connect the sources last. Slack, Outlook, Drive. Reading first, and an inbox folder for whatever comes in.
6. Correct it every day. Each time an AI gets something wrong, fix the file, not the chat.

The skeleton takes an afternoon. The value comes in the weeks after, one correction at a time.

Want one built for your team?

I build this for agencies and small teams that already use Claude or ChatGPT every day: the repository, the rules file, the connectors to Slack, Outlook or Drive, and the habits that keep it true. My own company runs on it first.

Tell me how your team works today and which tools you use. I’ll tell you whether a company brain is worth it for you, or whether a shared Drive folder will do.

Write to me about your team

Last updated: 29 October 2026

Questions people actually ask

Write to me about your team

Related reading.

• What is MCP: the standard most of these connectors run on.
• Connect your AI to your ad data: the same idea, for one source.
• My Marketing OS: the ads half of the system this brain feeds.
• You’re deciding from memory: why the record matters more than the recollection.
• Letting AI run your ads: every failure was a missing brief, not the model.
• Segundo cerebro para tu empresa: this guide, written for Spanish searches.