One repo for the whole product team. Meetings, decisions and research become shared context your AI reasons against, and every claim traces back to a source. Every PM does their work, saves sessions and runs research in the same place. Runs in Claude Code and Codex.
Every PM uses AI, each in their own way. One keeps a folder of prompts, another pastes notes into a chat, a third has built something impressive that only they can use. Company knowledge sits in dozens of places their AI can't reach, and the reasoning behind each decision is lost before the next cycle starts.
So plans reach review unchallenged, the same mistakes get made twice, and leaders see AI use going up but can't see what it has changed. The bill keeps growing.
OS-Intelligence is a shared folder the whole product team works in, with AI built in. Meetings, documents and decisions go in. The AI turns them into one current picture of each project, and every point links back to where it came from.
Everyone works from that same picture, in the AI tools they already use.
Three meetings, a research readout and a design review this week. You're back to back, so there's no time to work out how it all changes the plan. And you're the one on the hook for the direction.
The knowledge layer reads across all of it for you, and points out what's changed and what matters, with the source behind every point.
Someone researched this last year. The findings are somewhere in a deck, a doc or a Slack thread, and nobody remembers where.
The research agent works through your company's knowledge for you, starting with your riskiest assumptions, and saves every finding with its source so the next person can find it too.
The review is on Thursday. You asked your AI about the plan, and it said the plan looks great. It always does.
The sparring partner won't just agree with you. It raises the few issues worth discussing, challenges you with the evidence, and ends with a plan in your own words, with the biggest risks named.
OS-Intelligence is a team operating system for product work. It gives your team one shared store of what it knows, from meetings, documents, decisions and research, and puts it to work: better answers, research nobody has to repeat, and plans that hold up. Every claim traces back to source.
June to September 2026, at a PE-backed European marketplace, I applied OS-Intelligence to the product organisation as a team operating system.
Idea to production. A seven-week pilot took two roadmap features from a one-line idea to an evidence-backed PRD, a working prototype and a live release, with product, design and engineering setting the quality bar.
Research agent with sourced claims. An agent retrieved and synthesised company knowledge and wrote every finding as a sourced claim with a quote and a link. A PM ran it by hand and asked for the prompts.
Before claiming anything worked, I measured it against simpler setups: up to 21 questions with known answers, a control page sealed before any runs, and setups compared side by side. Losses included.
| Test | Result |
|---|---|
| Context comparison: the knowledge layer against a plain baseline | Mixed |
| Names the right people and says what it couldn't find | Held |
| Facts survive when the agent writes a summary | Didn't hold: two facts lost |
| Knowledge prepared locally beats leaving it in company tools | Held, and a re-test confirmed it |
The knowledge layer shows promise, but it didn't win on every measure. Next: more and larger datasets, starting with a public synthetic one that anyone can rerun.
Talks: Behind the Build (Happy Operators), AI Build Club.
The next milestone is a short video of the workflow end to end, then the V2 code. Want to hear when it's ready, or try it with your team?
Want to try something today? The open-source V1 is on GitHub.