Free company brain playbook
A practical guide for turning scattered docs, meetings, tickets, decisions, and repo knowledge into agent-readable company context.
It shows what to capture, how to curate it, where to store it, how agents access it, and how approved learnings get written back — so your team can use it from Claude, ChatGPT, Copilot, Cursor, or any tool they already trust.
Inside
The company brain operating model: capture, curate, store, use, write back
A source-system inventory for Drive, Confluence, Jira, Slack, meetings, CRM, support, GitHub, and BI
The recommended folder structure for a curated agent-readable context store
The metadata model that keeps the brain trustworthy: owner, source, last verified, confidence, and permissions
A worked team-context example, reusable shared skills, connector-assisted ingestion guidance, the MCP access pattern, and where a company agent runtime fits
A 30-day pilot plan: one team, three use cases, measurable before/after results
A 10-point readiness self-check that tells you whether to DIY, get the kit, or book the sprint
Why it matters
If your business context is scattered across tools, every agent starts from zero. People paste background, repeat decisions, and keep the best operating knowledge in their heads. Buying more AI tools does not fix this.
The playbook shows how to turn that raw material into a reviewed, versioned context system agents can search, pull into any tool or codebase, and improve through approved updates.
Start faster with the kit. The Company Brain Kit turns the playbook into a ready-to-use repo: templates, examples, and specs. £299, updates included.
See what’s inside ↓What you get
Map the knowledge your agents need: strategy, org, teams, products, workflows, decisions, meetings, and repositories
Choose what belongs in the brain, what stays in source systems, and what should never be ingested without review
Turn messy docs, meetings, tickets, and Slack threads into clean context packs with owners, source links, and freshness dates
Design the access layer so any agent — Claude, ChatGPT, Copilot, Cursor, OpenCode-style CLIs, internal assistants, or workflow automations — connects through MCP or a context API instead of people pasting background into every chat
Understand where a company agent runtime fits: persistent agents in Slack, Discord, CLI, webhooks, or cron can load the brain, use tools, run recurring workflows, and propose reviewed updates — without making the runtime the brain
Shared skills
The brain is not only facts and decisions. It also stores the approved methods your agents should follow: how you prep a trading meeting, write a PRD, capture a decision, or turn a workflow owner’s knowledge into an SOP.
1
Captured from a strong operator's way of working
2
Drafted into a reusable method with inputs, steps, and output format
3
Tested on real work, then approved by the owner
4
Shared through the brain so any agent can follow it
5
Improved or deprecated as the workflow changes
The operating loop
Find the raw material
Identify the systems where useful context already lives: Drive, Confluence, Jira, Slack, Granola, Intercom, Salesforce, GitHub, BI, and customer feedback.
Make it usable
Decide what matters, clean it up, dedupe it, redact sensitive material, and turn it into structured, agent-readable documents with owners and freshness dates.
Put it to work
Connect the brain to every agent through MCP — the open standard tools like Claude and ChatGPT use to reach your systems — give coding agents local context where they work, and route updates through review instead of letting agents freely rewrite the brain.
Agent runtime
The brain should not become a giant automation platform. It stays the trusted, reviewed context layer. A company agent runtime sits around it: loading context, calling tools, watching channels, running scheduled jobs, and proposing write-backs when new knowledge appears.
Agents can live in Slack, Discord, CLI, webhooks, cron, or internal tools instead of waiting for someone to open a chat window.
The runtime loads approved brain context and shared skills before acting, so every workflow starts from the same company reality.
MCP servers, APIs, and connector platforms help the runtime fetch live facts or take approved actions in source systems.
When work produces a decision, learning, or workflow change, the runtime proposes a brain update with source links and owner review.
Hermes is one open-source example of this pattern. You can also use another runtime, internal tooling, or lightweight scripts. The important separation is brain for trusted context, runtime for execution.
Source-system ingestion
The Company Brain does not ask you to bulk-ingest Slack, Jira, Drive, Confluence, meetings, support tickets, and CRM records. Durable knowledge gets curated into reviewed brain documents. Live operational facts stay in the tools that own them.
1
Find where decisions, workflows, customer truth, product context, repo knowledge, and team operating rules live.
2
Turn selected threads, docs, epics, transcripts, tickets, and PRs into reviewed brain documents with owners, source links, confidence, and freshness dates.
3
Leave ticket status, customer records, dashboards, and sprint state in Jira, CRM, BI, support, or GitHub — agents fetch them when needed.
4
When a task produces a new decision, learning, or workflow change, the agent proposes an update instead of silently rewriting the brain.
Connectors can help with the reading and the actions. Direct APIs, MCP servers, or Composio-style platforms can reduce the work of reaching Jira, Slack, Drive, Confluence, GitHub, CRM, support, and BI. But the connector is the pipe. The brain decides what becomes trusted company context.
Read before you sign up
Here’s the full table of contents. Three complete chapters are free to read online. If they’re useful, the rest of the PDF is one form away.
Why AI stalls without a company brain
What to capture: the source-system inventory
What belongs in the brain (and what never goes in)
How to structure it
How agents access it — MCP, tools, and runtimes
Keeping it alive: write-back and review
Free template
Decision records are one of the most useful documents in a company brain. They stop the same debate restarting and give agents the “why” behind every rule. Copy this one, write up your last big decision, and bring it to your next leadership meeting.
The other seven templates — and the filled examples, skills, and rollout assets — are in the kit.
--- owner: <the decision maker> date: <YYYY-MM-DD> status: active --- # Decision: <one-line summary> ## Decision What was decided, in one or two sentences. Unambiguous. ## Context The situation that forced the decision. What problem, what pressure. ## Options considered The realistic alternatives, each with the reason it lost. Two lines each. ## Why this option The actual reasoning. This is the section that saves the re-litigation. ## Consequences What this commits your company to. What becomes easier, what becomes harder, what should be revisited — and when.
Which path should you take?
Best if
You are still learning the model.
Understand what a Company Brain is, what belongs inside it, and why AI work keeps stalling without shared context.
Get the playbook →Best if
You have an operator or technical lead who can self-install.
Get the repo structure, templates, examples, skills, ingestion guides, and rollout assets. Open it in Claude or Cursor and install the first version with your agent.
Get the kit →Best if
The first build is clear, but DIY will stall.
A hands-on week for one ready team, owner, or operating loop. Heavier than the audit, narrower than the sprint.
Ask about Build Week →Best if
The problem is real, but the build shape is unclear.
The lower-friction consulting step: mostly Marc-led diagnosis of first team, source boundaries, access pattern, owner model, business case, and whether to DIY, Build Week, or Sprint.
See the audit →Best if
You want the first company-ready version built with you.
A 30-day implementation that installs the brain, access layer, skills, QA gates, owners, and rollout rhythm your teams can keep using.
Explore the Sprint →Company Brain Kit — £299
An implementation repo, not a document bundle: structure, filled examples, specs, agent skills, ingestion guides, and rollout assets. Open it in Claude or Cursor and it installs itself with you — your agent interviews you and drafts your first documents in about 30 minutes.
If you built this internally, the hard part wouldn’t be writing a few templates. It would be deciding the structure, the curation rules, the permissions, the examples, and the rollout. That design work is what the kit packages — the first 40–80 hours, often spread across 4–6 weeks of back-and-forth, done.
Get the kit — £299 →30-day money-back guarantee. And if you book the AI-Readiness Audit within 90 days, the £299 comes off the audit price.
One purchase covers your whole company · Updates included, forever · Reply with one question — Marc answers personally.
The brain structure, ready to clone — with CLAUDE.md and AGENTS.md so your agent understands the repo the moment you open it
The install skill: your agent interviews you and drafts your first five documents — strategy, principles, first team, decisions, a workflow — in about 30 minutes
Six more starter agent skills: ask-the-brain with citations, decision records, meeting capture, workflow docs, new-hire context packs, and the monthly freshness review
All eight document templates, plus a complete 14-document example brain for a realistic 400-person company — copy the shape, not the blank page
The MCP access spec your engineers (or coding agents) can build from, and connection guides for Claude, ChatGPT, Cursor, and Copilot
Source-system ingestion runbooks for Slack, meetings, Jira, Confluence, CRM, support, GitHub, and BI — what to extract, what to leave live, what never to ingest, extraction prompts, review checklists, and acceptance criteria
Connector guidance for Jira, Slack, Drive, Confluence, GitHub, CRM, support, and BI — use direct APIs, MCP servers, or Composio-style platforms as pipes, not as the brain
Runtime pattern guide: how Slack/Discord/CLI/webhook/cron agents can sit around the brain, with Hermes as one optional open-source example — not a required dependency or done-for-you deployment
Governance pack: permissions model, write-back review workflow, and a prompt-injection checklist
Workflow selection and ROI/payback worksheets, the 30-day rollout plan, and the team onboarding pack (launch email, checklists, weekly ritual)
Maintenance guide: the eight classic failure modes, the debug order for wrong answers, and the monthly review routine
Kit boundary
The kit gives you the Company Brain structure, examples, skills, specs, and implementation guidance. It explains how a runtime can operationalize the brain, but it is not a done-for-you runtime deployment.
Included
Not included
If you want runtime setup implemented with your team, that belongs in the Sprint.
Look inside — the actual repository you get
company-brain/
├── CLAUDE.md · AGENTS.md your agent reads these and offers to install
├── company/ strategy · org · ai-principles
├── teams/ one owned, dated context doc per team
├── products/ one per product or business area
├── workflows/ SOPs agents can follow
├── decisions/ dated decision records — no more re-litigating
├── meetings/ curated digests, never transcripts
├── skills/ 7 starter skills, incl. install-your-brain
├── repo-context/ business context per codebase
└── _kit/
├── guides/ 13 guides: setup → ingestion → MCP spec → runtime patterns → failure modes
└── example-brain/ every template, filled in for a 400-person companyUngated demo
Same ask, no brain
“Draft a reply: customer wants a full refund on a £340 order, two items arrived damaged.”
“I’m so sorry about your experience! We’ll process a full refund of £340 right away, and your replacement items will arrive within 3–5 business days…”
Promised a refund above the approval threshold, invented a delivery date, missed the per-brand split. Polite — and wrong three ways.
Same ask, with the brain
“Draft a reply: customer wants a full refund on a £340 order, two items arrived damaged.”
“I’m sorry two of your items arrived damaged — that’s on us. I’ve flagged the £340 refund for approval now (orders over £200 get a quick second check) and you’ll hear from me today. The two brands ship separately, so the refunds will appear as two credits…”
Knows the £200 approval threshold, the split-shipment refund rule, and the tone policy — because your team wrote them down once.
Read a finished document before you buy
This is an actual file from the example brain — a team context document for a fictional 400-person marketplace. Every template in the kit comes filled in at this level.
--- owner: sofia.lemos last_verified: 2026-05-20 confidence: high --- # Team: Member Support ## What we own Customer support across chat and email (Intercom), the help centre, and support tooling. 70 people across London and Lisbon, covering EN/DE/FR/PT, 07:00–23:00 CET. ## Current priorities 1. Cut cost-per-contact 25% by year end (currently £3.80) without CSAT dropping below 4.3 2. Automate triage for the top 5 contact categories (~55% of volume) 3. Bring Plus-member first-response time under 1 hour ## What agents should know - Refunds above £200 always need human approval — see decision 2026-03-12-refund-approval-threshold - Never promise delivery dates; quote the brand's dispatch estimate - Plus members: lead with the goodwill option; churn risk is 3× revenue impact - Legal threats, chargebacks, press: escalate same day, never answer directly
Thirty lines, and an agent drafting for this team now has the rules, priorities, and context it needs. The example brain has fourteen documents at this level — and the install skill drafts yours with you.
Yes — the full guide, no call required. Add your email and it will be sent instantly. You can also download it straight after submitting the form.
The playbook teaches the model and is enough to DIY. The kit is the implementation repo: the cloneable brain structure, a 14-document worked example, 7 starter agent skills (including one that installs the brain with you), the MCP spec, curation runbooks, governance pack, rollout plan, and onboarding assets — the 40–80 hours of design work, done.
No. Start with the pattern in the playbook and kit: source-system inventory, curation runbooks, extraction prompts, acceptance criteria, metadata, and review rules. For a pilot, selected exports, links, and read-only connectors are enough. For steady state, MCP/API access can combine the reviewed brain with live lookups into Jira, Slack, CRM, BI, GitHub, and other systems.
No. The Company Brain comes first. A runtime is useful when you want agents to run persistently in Slack, Discord, CLI, webhooks, or scheduled jobs. Hermes is one open-source example of that pattern, but the playbook and kit are not dependent on Hermes.
30-day money-back guarantee, no questions asked. And if you book an audit within 90 days, the £299 comes off the audit price anyway.
Start with the free playbook. Upgrade to the £299 kit for the full templates. If the shape is unclear, use the audit. If the implementation need is clear, book a call about the sprint.