Agentic Product Management — with Jiri Kobelka, ex-founder of Tatum

AI writes code over a weekend now. So why aren't products 10× better?
Because the bottleneck moved. Not building — *knowing what to build.*
We're hosting **Jiri Kobelka** — founder and CEO of **Tatum** which sold for 46M USD, the developer platform he scaled to 150k+ developers, and now founder of **ClosedLoop AI**. Over the past year he's been inside US AI-native product companies, mapping how the best teams have completely inverted how they decide what to build.
ClosedLoop turns 100% of customer conversations — every call, ticket, and Slack thread — into structured product intelligence, and feeds it straight to the AI coding agents (Claude, Cursor) over MCP. The machine collects and clusters; the human investigates and decides; the agents build and ship. Customer conversations moved from the first step of discovery to the last step of validation.
**The talk.** The new operating model Jiri saw from the inside — and the patterns you never see when you read 5% of your feedback by hand: why 70% of churning customers had a *positive* NPS 60 days before leaving, why customers who never file a single ticket churn at 2×, why three of four lost deals name the same competitor weeks in advance. Plus a live demo — five of a real company's biggest problems, surfaced in ten minutes.
**The workshop.** Then we get hands-on. Everyone works the same fictional SaaS company and drills one move — *what the customer said → what's the real problem → what wrong assumption am I making → what do I actually build* — across three escalating cases, each with a trap engineered so the obvious answer is wrong. Then you hand your decision to an agent as a context brief and watch it turn your judgment into a grounded build plan. The point lands without a line of code: the agent does the fast part, you do the irreplaceable part.
Because the last bottleneck is the one that was always there: the customer.