Demo · Show, don’t tell
Same AI tool. Same task. Different answer.
This is the simplest way to understand a Company Brain: the model is not the main difference. The approved business context it can reach is.
The ask
“Draft a reply: customer wants a full refund on a £340 order, two items arrived damaged.”
Without the brain
Generic helpfulness
“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 full refund above the approval threshold.
- ×Invented a delivery date your team could not guarantee.
- ×Missed the per-brand split shipment and refund rule.
- ×Sounded helpful, but was wrong.
With the brain
Grounded operating knowledge
“I’m sorry two of your items arrived damaged — that’s on us. I’ve flagged the £340 refund for approval now, because orders above £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 orders above £200 need a quick human approval.
- ✓Does not invent delivery dates or operational facts.
- ✓Explains split refunds because your company context says the brands ship separately.
- ✓Starts from approved policy and leaves the human to judge edge cases.
What changed
The better answer came from better context.
The agent could reach approved policies, decisions, team norms, source links, and ownership. The human still reviews the reply. The difference is that the first draft starts from your company’s own context instead of generic guidance.
Example context files the agent used
- support-policy.md — refund thresholds, escalation rules, tone guidance
- brand-operations.md — which brands ship separately and what customers should be told
- decisions/2026-03-12-refund-approval-threshold.md — why £200 requires approval
- team/member-support.md — priorities, ownership, languages, and service-level norms
Same brain, different teams, different asks
The refund example is one proof point. The system is the reusable context layer.
“Prep me for Monday's trading meeting.”
Context used
Trading cadence, open commitments, KPI definitions, previous meeting digest, ownership map.
Output shape
A brief with what changed, where decisions are needed, and which follow-ups are stuck.
“What did we decide about pricing?”
Context used
Decision log, launch notes, margin guardrails, customer objections, owner and review date.
Output shape
The decision, why it won, what it commits your team to, and when to revisit it.
“Draft the PRD for this feature.”
Context used
Product context, prior decisions, customer evidence, engineering constraints, house style.
Output shape
A first draft grounded in the real business instead of a generic SaaS template.
The brain supports more than one workflow.
The same context layer can support support replies, trading prep, PRDs, handovers, decision lookup, and weekly reviews. Workflows are where the Company Brain proves itself. The Company Brain is not the workflow.