How to ship AI features users actually use
with Cemre Güngor
Cemre has worked on social sharing at Facebook and Instagram, AI workflows at Figma, and is now reimagining the browser at The Browser Company with Dia. We talked about the difference between AI features that ship and ones that stick — what he calls the gap between "demo candy" and durable product value. He showed how Dia uses browser-level context to do things no standalone app can, and we got into why internal usage is the single most important signal for whether an AI feature is ready.
What I took away
Cemre's test for an AI feature — would you use this yourself, every day? — is one I've applied in every product review since. If the team building it isn't using it, no amount of polish will fix the product. His point about mental model debt also clarified something for me: PMs who've only built in one paradigm often carry assumptions that silently break in AI-native products, and it takes deliberate effort to unlearn them.
Where this connects
The conversation about when to wait for model improvements versus when to ship has directly shaped how I'm thinking about AI investments on my roadmap.