Figma built an internal AI tool called PM OS, designed to cut busywork for product managers and surface time back for high-stakes decisions. The project came from a small volunteer group, not a dedicated team, and grew fast enough to become a daily workflow staple across Figma's PM org. This sits alongside similar internal AI bets at Ramp, which built Glass, and Stripe, which built Minions.
The real story here is not the tool itself but the build process. A volunteer-driven, experimentally-minded team shipping something that sticks is a meaningful signal about how Figma operates internally. The talk frames this as a culture story as much as a product story, which makes it worth watching for anyone thinking about how AI tooling actually gets adopted inside organizations.
The full talk was delivered at Config 2026 and targets PMs and designers, but the lessons apply to anyone deciding whether to buy, borrow, or build internal AI. The specific contrast with Ramp and Stripe is worth unpacking in the original, as is the framing of PM work as a problem of cognitive overhead rather than raw productivity.
[WATCH ON YOUTUBE →]