Bio
Head of product for Claude Code and Co-work at Anthropic. Runs one of the fastest-shipping product orgs in the AI industry — timelines collapsed from quarter-long to week-long with the help of an "evergreen launch room" cadence and a heavy reliance on engineers shipping end-to-end. Vocal advocate for replacing PRDs with metrics readouts plus a written team-principles doc.
Operating themes
- Process-light, ritual-tight. Replace PRDs with metrics readouts plus a written principles doc; keep tight cross-functional rituals around dog-food → marketing announcement.
- Engineers with taste over more PMs. As code becomes cheap, the scarce skill is deciding what to write.
- Build for the next model, not the current one. Ship features that almost work today; remove crutch features when the new model lands.
- Research preview as default mode. Lower commitment cost so the team can ship in days, then iterate publicly.
Cards
- Brand work as "research preview" to compress timelines from quarters to weeks — Label new work "research preview" to ship in days and iterate [Tier A]
- Replace PRDs with weekly metrics readouts plus a written team-principles doc — Replace PRDs with weekly metrics + a written principles doc [Tier A]
- When a new model lands, re-read the system prompt and remove crutches — Each model launch is a deletion event for crutch features [Tier A]
- Hire engineers with product taste rather than adding more PMs — Hire engineers who can ship end-to-end without PM input [Tier B]
- Taste is the scarce skill in an AI-native team — Taste is the scarce skill in an AI-native team [Tier B]
- Build products at the edge of what does not yet work — Build at the edge of what does not yet work; ship when the model lands [Tier B]
- An automation that works 95% of the time is not an automation — A 95%-reliable automation is not an automation [Tier B]
Sources captured
- 2026-04-27 — Lenny's Podcast, "How Anthropic's product team moves faster than anyone else" (
raw/podcasts/cat-wu--anthropic-product-team--2026-04-27.md)