Cat Wu

Head of product for Claude Code and Cowork at Anthropic. One of the principal architects of Anthropic’s product-shipping culture.

Background

• Engineering background; worked as an engineer for several years.
• Brief stint as a VC before joining Anthropic.
• At Anthropic, leads the PM organisation for Claude Code and Cowork alongside Boris (tech lead and product visionary).
• Rock climber; aspiring reader (targeting 1–2 books/week from a current ~0.5/week).
• Twitter: @_catwu.

Talks in this wiki

Recurring themes

Product taste as the scarce skill. As code gets cheaper to write, the scarce resource is knowing what to write. Cat Wu consistently returns to product taste — aesthetic and strategic judgement about what to build — as the most durable PM skill and the hardest to automate.

Mission as decision technology. Anthropic’s mission is not just branding; it is an operational protocol for resolving competing priorities. Cat Wu attributes much of Anthropic’s cross-org execution speed to the willingness of teams to sacrifice their own KRs for Anthropic’s mission.

Calibrated expectations of models. The hard skill is not knowing what future models will do, but knowing what the current model can do — and building harness and product features to elicit that capability and patch its weaknesses. Being the “right amount of AGI-pilled.”

Bias towards action. Cat Wu’s personal motto: “just do things.” Jobs are not as rigidly scoped as organisational charts suggest; first-principles reasoning about what the team needs, combined with willingness to act across boundaries, is more valuable than role adherence.

Key concepts associated with Cat Wu

• Product Taste — her candidate for the most durable PM skill in the AI era.
• Evals — her argument that PMs should write them; even 10 good evals add concrete value.
• “The model eats the harness” — her framing of the harness simplification dynamic as models improve.
• “The right amount of AGI-pilled” — calibration between current and future model capabilities.