What I'm building
Product Brain — persistent, governed memory for coding agents, so product knowledge compounds instead of decaying every session.
Agents forget. Chat context compresses. Decisions evaporate. The agent that confidently violates a choice you made two weeks ago isn't broken — it had no governed memory to read.
Most teams patch this with scattered context files, wikis, and hand-maintained CLAUDE.md rules. Those are lossy mirrors. Nobody knows which parts are still true, how they connect, or what governs what.
Product Brain is not a document store. It's a governed knowledge graph — the Chain — where product facts have type, status, provenance, and relations.
Typed entries, not loose notes
Decisions, tensions, standards, insights, work packages — each with a prefix and a place in the graph. Not markdown pages in a folder. Structured entries agents can orient from.
WHY, history, and relations
Every entry carries its rationale, version history, and typed links to what it resolved, informed, or superseded. You capture not just what was decided — but why, and what it connects to.
Draft → ratify → serve
Agents capture drafts during work. Humans ratify what becomes truth. Nothing serves until it's committed to the Chain. Governance isn't overhead — it's what makes agent memory trustworthy.
Agents read the Chain before they act — via MCP, the pb CLI, or Cortex at work.productbrain.io.
One ratified brain, projected into each tool's native format and pulled live at the moment of work.
Product Talk
A voice-first surface in the Product Brain family — talk through your work and the Chain with Sage, grounded in the same governed memory. Early development; the web voice path works, public launch still ahead.
Product Talk consumes Product Brain via the gateway. It never shadows the Chain — drafts promote through the same human-gated ritual.
AI has changed what's possible. Non-developers can now build production systems.
The skill isn't prompting. It's governance.
If you can make reasoning explicit — capture the WHY, link the relations, ratify what's true — agents can execute without undoing decisions they never saw. The bottleneck was never typing code. It was clarity that survives the session.
I build Product Brain the same way I use it: orient from the Chain, capture drafts as I learn, commit what's durable. This site, the product, Product Talk — all dogfooded on the same governed memory.
My decade in organizational design taught me why governance matters. Product Brain is where that meets agent-native product work — not org charts in software, but the coherence layer for teams shipping with AI.
Reality
- Solo non-developer founder
- Side project with founder ambition
- Open beta — built in the open
- Zero traditional developer hires
The loop
- Orient from the Chain every session
- Capture decisions, tensions, learnings as drafts
- Ratify what becomes SSOT
- Agents read it on the next prompt
The Chain is the product. Chats and docs are lossy mirrors — I re-orient from governed entries, not from a stale markdown file I wrote last month.
What exists
- Governed Chain with hundreds of typed entries in daily use
- MCP server,
pbCLI, Cortex UI (Bridge, Review, Brain Chat) - Handshake projections into Cursor, Claude Code, Codex, Copilot
- Open community at community.productbrain.io
What's in flight
- Coherency engine — challenge work that contradicts governed reasoning
- Product Talk voice surface (early)
- Flywheel metrics — proving agents get work right first time with Chain context
The honest position
Product Brain is real and in open beta. The substrate works. The moat-grade experience — automatic contradiction catch, measurable beat-the-static-file baseline — is still being proven. I'm not selling certainty. I'm building in public and measuring what compounds.
I write about building Product Brain, agent governance, and what I'm learning on the way. I share the messy middle — what works, what doesn't, what I'm still proving.
Want the founder story behind this?