Write your business down once

Every AI employee needs to know your business. Write it down once, in one place, and let every employee read from there. That place is the Founder Brain.

The problem: every skill gets its own brand folder

Here is what happens without it. You install a hook writer. It asks for your audience and your voice, so you write a short brand file next to it. A month later you add a caption writer. It wants the same, so you copy the file. Then an email writer. Another copy.

Now you change your offer. You update one file. The other two still say the old thing. Your hooks talk about the new offer, your emails about the old one. Nobody notices until a customer does.

Copies always drift. The fix is one source that every employee reads, and no copies.

What the Founder Brain is

The Founder Brain is a folder of plain text files (markdown, the .md files you already met). It holds what is true about your business. Every employee reads from it. Nobody keeps a private copy.

It comes as a template. You copy the folder, open it in Claude Code, and run /start. Claude then interviews you in five rounds: your company, your topics and who owns them, your sources and which one wins, who may see what, and your first sources. After each round it asks if you want to go on.

You can also read and edit the files yourself, in any text editor. Many people use Obsidian, a free note app that opens a folder of markdown files.

What goes in

Five topics. These are the ones every marketing and sales employee needs.

Topic	What you write	Who reads it
Company	What you do in one sentence, since when, team, market, why you exist	Every employee
Offer	Your products and services, what makes them different, prices with a date	Marketing, sales, website
Audience	Who you help, their pain in their own words, their objections	Marketing, sales, website
Voice	How you talk: sentence length, words you use, words you never use	Everyone who writes
Proof	Real cases, reviews and results, with permission and a source	Marketing, sales, website

Two more are useful later: your visual identity (colours, fonts, logo) and your market (who you compete with).

Example: a fictional yoga studio, “North Room”

North Room does not exist. I made it up to show what a filled page looks like. Here is its voice page, short version:

---
type: knowledge
domain: brand-core
owner: Anna (founder)
status: current
sources: [interview with Anna, website copy]
updated: 2026-10-25
review-by: 2027-01-25
---
# Voice

## How we talk
- Calm and plain. Short sentences.
- We talk to one person: "you".

## Rules

### Always
- Say "class", never "session" or "workout".

### Never
- No body-shaming, no "summer body".
- No exclamation marks in a row.

## Example line
"Your back hurts from the desk. One slow class a week helps more than you think."

And its audience page, the part about pain:

## Pain, in their words
- "I sit all day and my back is stiff by 4 pm."
- "Gyms feel loud. I don't belong there."
- "I tried yoga apps but I never keep it up."

Notice what is not there: no invented numbers, no “95% of members feel better”. If you don’t have proof, the proof page stays empty. An empty page is honest. A made-up number ends up in a post, and then in front of a customer.

Common mistake: writing your voice as adjectives. “Friendly, professional, warm” tells Claude nothing. Write rules and one example sentence instead.

The loop: how knowledge gets in

Nothing goes straight into the brain. Every change takes the same route.

raw input  →  proposal  →  you approve  →  knowledge

1. Raw input. A document, a meeting note, an email, a thought. It goes into the raw/ folder as it is. Raw is never the truth.
2. Proposal. Claude reads the raw input and writes a proposal: what changes on which page, and why. The page itself does not change yet.
3. You approve. You read the proposal. You approve it, change it or reject it. The brain has an approve command for this.
4. Knowledge. Only after approval does the change land in the knowledge/ folder. That is what employees read.

Employees follow the same route. When the caption writer learns a new objection from a comment, it does not edit your audience page. It writes a proposal. You decide.

Four rules that keep it true

These four rules are the difference between a brain and a folder of documents.

1. An owner per topic. One person decides what is true about the offer, one about the voice. In a team of one, that is you, for everything. Write it down anyway.
2. Source and date on every page. Each page says where the information came from and when it must be checked again. When that date passes, the brain asks you if it is still correct.
3. Decisions with the why. Don’t only write “we stopped the 1-on-1 offer”. Write why, what else you considered, and when you look at it again. Six months later, that why is the part you need.
4. A human between AI and truth. Claude proposes. A person approves. No employee writes directly into your knowledge.

Check: open any page in your brain. Can you see who owns it, where it came from and when it expires? If not, the page is not finished.

How employees read from it

Each project folder in Claude Code can have a file called CLAUDE.md at its root. Claude Code reads that file at the start of every session. It is the briefing your employees get every morning.

You point that briefing at the brain. CLAUDE.md can pull in other files with an @ and a path. Claude Code then loads those files too, every session.

Say your brain and your marketing project sit next to each other:

my-ai-company/
├── company-brain/
│   └── knowledge/
│       └── brand-core/
│           ├── company.md
│           ├── offer.md
│           ├── audience.md
│           └── voice.md
└── marketing/
    ├── CLAUDE.md
    └── .claude/skills/

Then marketing/CLAUDE.md starts like this:

# Marketing

Before any task, read what is true about the business:

@../company-brain/knowledge/brand-core/company.md
@../company-brain/knowledge/brand-core/offer.md
@../company-brain/knowledge/brand-core/audience.md
@../company-brain/knowledge/brand-core/voice.md

Never edit these files. New knowledge becomes a proposal in the brain.

A few things to know about these imports:

• The path starts from the folder where CLAUDE.md sits. ../ means “one folder up”.
• The brain sits outside the project folder. The first time, Claude Code shows a dialog that lists these outside files and asks if it may load them. Approve it. If you decline, the imports stay off.
• A path with spaces needs a backslash before each space, like @../My\ Brain/knowledge/brand-core/voice.md. Easier: no spaces in folder names.
• Every imported file loads in every session. Keep pages short and to the point.
• Inside a session, type /context to see which memory files Claude loaded.

The file names above show the idea. Use the names of the files in your kit.

Now the hook writer, the caption writer and the reviewer all read the same voice page. Change it once and all three change with it.

Keep it alive: a weekly check

A brain goes stale if nobody looks. Once a week, ask the brain what needs attention. It has a weekly check command that lists, per owner:

• pages past their review date
• pages that contradict each other
• pages without an owner or a source
• proposals still waiting for you

Done every week, the list stays short. Approve what’s right, reject what’s wrong, update one date. Put it in your calendar on the same day each week.

Common mistake: loading your whole archive on day one. Start with the five topics. Add one client, one process or one decision when someone asks about it twice.

Where to get it

The Founder Brain template is BX-001 in the library. Comment or DM the keyword BRAIN to get it. It is also in the starter kit.

Next

Your business is written down. Now give it its first employee: Install your first employee.