3 takeaways from our last workshop: Build your Product AI Operating System

Oct 30, 2026

Zihan Wang

Last week, a room full of PMs built their own AI operating system in one day. Here's what we took away.

1. Provenance is key to building trust.

The biggest worry in the room was AI slop. Output that looks finished. Reads well. And nobody can check it.

Much better: a wiki page where every line links back to the interview or ticket it came from. You click, you see the source, you decide for yourself.

Try this: if you wouldn't defend it in the room, don't share it. It doesn't need to be bulletproof. It needs a backbone: data, an insight, or a point of view you've thought through. And if AI wrote it, make it show its sources.

2. A good skill starts as a team agreement.

One participant had built a PRD agent. It's the one their team actually uses.

Why? The PM built it with their tech team. They first agreed what a good PRD looks like. The agent just holds that agreement.

We saw the same thing live. The room started designing a skill to prioritize bugs. Before anyone wrote a line, the room debated criteria: how often does it happen, who reported it, what else do we know?

The skill was the easy part. The agreement was the work.

Try this: before you build a skill, sit down with the people who use its output. Agree the criteria. Write them down. Then automate.

3. Don't install a team OS. Build it together.

Once people had their own OS running, the next question came fast: how do we roll this out to the team?

We get it. But a team OS pushed onto people who don't use AI every day ends like the company wiki. Nobody opens it.

AI fluency comes from daily use. You only know what's worth sharing once you've felt what helps in your own work. That's how it went for us too. We each built our own OS first. Only then did we see what belonged in the shared one.

So start personal. But don't start alone.

Put the whole team in one room. Everyone builds their own OS, on the same shared knowledge. Along the way, the agreements from #2 happen: what a good PRD looks like, how we score a bug. Those become the team's shared skills.

And people adopt them, because they helped build them.

Try this: one room, everyone builds their own. The shared parts follow.

What participants told us, and what's next

The top win: the thinking behind the architecture, plus a template to start from. Enough to go home and extend their own setup.

The next session gets even better:

  • More time to build on your own use cases.

  • Sharper ideas. The AI grills your idea before you build it.

Join the next one

Next public session: 4 December. Grab a seat.

Want this for your whole team? We run it in-house too, built around your product and your tools.

Book a 30-minute call with us

See you soon.

Back to Overview