Rolling out a company brain
Adoption beats announcement. Launching to the whole company on day one creates a pile of unreviewed captures you spend months culling, and people stop trusting answers before the brain has earned any trust.
Phase 1 — Pilot (2–4 weeks)
- One workflow, seeded per "Start narrow" in
vault-config.md. - One or two people who live in that workflow daily and will actually run
/cb review. - Capture everything relevant to the workflow; don't capture outside it yet.
- Hold a weekly review pass. If it slips, fix that before anything else; it's the habit everything else depends on.
Phase 2 — Prove value
Move on only when all three are true:
| Signal | Bar |
|---|---|
| Queries answered | The pilot group asks it real questions weekly and gets useful answers |
| Review cadence | Weekly review has held for 3+ weeks; the unreviewed queue isn't growing |
| Trust | Answers cite verified sources often enough that people stop double-checking |
If a signal misses, narrow rather than widen: fewer dirs, fewer sources, a tighter workflow.
Phase 3 — Expand
- Add one workflow (and its dirs) at a time, with an owner who runs review for it.
- Add people in the order they'll contribute, not by org chart.
- Turn on auto-sync (call transcripts, Slack, CRM) per workflow only after manual capture has proven which content is worth keeping. Auto-sync on day one multiplies the review queue before anyone knows what good looks like.
- Set sensitivity levels before leadership or confidential content arrives, not after.
Anti-patterns
- The big launch. All dirs, all people, all syncs at once. Review can't keep up, so trust never forms.
- Capture without review. A growing
unreviewedqueue is the leading indicator of a brain nobody trusts. - Measuring pages, not answers. Page count is vanity. Questions answered without asking a person is the metric.