# What Should a Small Business Put in Its Company Brain First? > A company brain is how your business works, written down once so people and AI tools read the same thing. The twelve pages to write first, in order, and the mistakes that make one go stale. - Author: [James Oldham](https://sentrysolutions.ai/about/james-oldham), Founder, Sentry AI - Published: 2026-10-22 - Canonical: https://sentrysolutions.ai/blog/what-to-put-in-a-company-brain-first Every business already has a company brain. It lives in the owner's head, in a few long email threads, and in the habits of whoever has been there longest. The trouble starts when that person is away, when someone new joins, or when you ask an AI tool to help and it gives you an answer that would be right for any business except yours. Writing it down fixes all three. The question is where to start, because "document everything" is how these projects die. ## Start with what a new hire asks The best guide to the first pages is the questions a good new hire asks in their first fortnight. They are the questions an AI tool needs answered too. In the order we would write them: 1. **What we do and who for.** One page. The customers you serve and the ones you turn away. 2. **What we sell.** Each offer in a paragraph: what the customer gets, how long it takes, who it suits. 3. **How work flows.** From first enquiry to paid invoice, step by step, with who does each step. 4. **Who decides what.** The decisions that need the owner, and the ones that do not. 5. **Which tool holds which truth.** The CRM for contacts, the accounting system for money, the shared drive for proposals. When two tools disagree, which one wins. 6. **The rules nobody wrote down.** The things a new person gets wrong in their first month. This page is usually the most valuable one in the brain. 7. **How we talk.** Tone, the words you use, the words you never use. 8. **Recurring jobs.** The weekly and monthly work and who owns it. 9. **The numbers that matter.** The handful of figures the owner actually watches. 10. **Recent decisions and why.** Three is enough to start. The reasons are what get lost. 11. **The glossary.** Words that mean something specific in your business. 12. **Who to ask.** For each area, the person who knows. Twelve pages, each short enough to read in two minutes. That is a working brain. ## Write it in your own words, then link it Each page should sound like the business, not like a template. Then link the pages to each other: the workflow page links to the roles page, the offer page links to the glossary. Links are what let a person, or an AI tool, follow a question from one page to the next instead of reading everything. If you would rather be interviewed than stare at a blank page, our free [company brain starter skill](https://sentrysolutions.ai/go/skill/company-brain-starter) has Claude ask you the questions one at a time and write the first twelve pages from your answers. ## The mistakes that make one go stale **No owner.** A brain everyone can edit and nobody owns is out of date within a quarter. Name one person. **Written once.** The brain is useful when it changes as the business changes. The trigger for an update is simple: someone asked a question it could not answer, or it gave a wrong one. **Too much at the start.** Importing every old document gives you a large brain nobody trusts. Start small and true. **Kept away from the tools.** If the team's AI tools cannot read it, the brain helps people and leaves the AI guessing. Put it where both can reach it. ## What it does once it exists The obvious gain is onboarding: a new hire reads twelve pages instead of shadowing someone for a month. The larger gain is that every AI tool the team uses can now answer in your business's terms, because it is reading your rules rather than inventing plausible ones. That is the difference between an assistant that sounds right and one that is right. We run our own business on one. [Inside ours](https://sentrysolutions.ai/company-brain/ours) shows how it is organised and how it is kept current. ## Related reading - [From Karpathy's LLM Wiki to the Company Knowledge Layer: Why Every Organisation Needs Its Own Knowledge Graph](https://sentrysolutions.ai/blog/karpathy-llm-wiki-for-organizations): Andrej Karpathy proposed an LLM wiki for personal context. The same idea applied at the organisational level is the missing piece behind every failing enterprise AI deployment. This is how a unified company knowledgebase plus a custom knowledge graph becomes the foundation of a company's AI transformation, and why the future is custom company LLMs grounded in your real data. - [Why Enterprise AI Fails Without Structured Context](https://sentrysolutions.ai/blog/why-enterprise-ai-fails-without-structured-context): Most AI projects fail not because the models lack capability but because the organisation's knowledge is fragmented, undocumented, and impossible for AI to traverse. The context problem is the real bottleneck of enterprise AI adoption, and almost nobody is solving it.