Set up your agent.
Keep your workflow.

No. With a supported CLI/MCP connection and working instructions in place, daily lookups, capture and authorised acceptance can happen through your agent, in the environment where you already work.

Your agent handles the record; you supply the judgment and remain accountable. Keep its instructions, your roles and the recorded knowledge current as your team learns. Setup and connection support vary by agent.

See supported agents and connection guides →

Claude on web and desktop through the MCP connector. ChatGPT on plans with custom MCP connectors. Claude Code, Cursor and Codex through CLI or MCP; Copilot has instruction-file support. These connections are beta. Only Claude Code has opening briefing hooks; other surfaces look up context as needed.

See the connection differences and guides. Check your plan’s connector support; setup varies by surface.

Make your purpose, vision, strategy and objectives visible. Define roles, holders and decision scopes. Configure agent instructions and agree how you’ll capture, accept and maintain the record together.

Bring what already exists. You can start small and refine the foundation as you learn. Some entries have direction prerequisites. If your direction and roles already exist, use them. The optional foundation day is for teams that want help making them clear—not a prerequisite for trying the beta.

Yes. Your agent can accept any kind of entry within the authority of the role its session acts in, once the entry’s requirements are met. Holding several roles does not give one session all of their authority. It’s recorded as acting for you; you remain accountable and can confirm it yourself.

A draft isn’t accepted knowledge. Acceptance establishes the agreement; completing the work is a separate state.

The owning role decides. An unfilled owning role falls back to the nearest filled Team Lead role up that team’s hierarchy. An entry without an owner uses the applicable Team Lead fallback.

1. Your team’s lead
2. Then a parent team’s lead
3. Then the organisation’s lead

The organisation’s root Team Lead must always have a holder. This fallback covers an empty role; it doesn’t override a filled owner. Broken or ambiguous ownership needs correction before the change can proceed.

Not your role? The change is refused and you’re told who decides. Raise a tension or ask the owner.

Your configured agent can look up context, capture learning and accept knowledge within its acting role as you work. Your team decides what belongs in the record and keeps direction and roles current.

Possible contradiction flags invite review; they don’t decide or guarantee alignment. Changes aren’t pushed into every active session. Teach your agent to consult the record when it acts.

Use the workspace for records intended for its members. Role ownership governs acceptance and changes; connected agents need authenticated access and writes need the appropriate role/session. Private-draft access is changing, so current beta drafts should not be treated as a confidentiality boundary.

Product Brain does not train AI models on your workspace data. Service providers process data in the US as described in the privacy policy.

Complete workspace export is not currently supported. CLI and MCP let you read individual records, but that is not a complete export of entries, relations and history.

If a full export is a requirement for your team, discuss it with Randy before putting important knowledge into the beta.

You can start with the free beta tier and included usage. Final team software pricing is still being defined, separately from the founder-led service.

Ask Randy about your team’s requirements and software costs before committing to a paid engagement. The foundation day is optional; it isn’t the price of trying Product Brain.