Cat Wu

Cat Wu is Head of Product on the «équipe Claude Code» at «Anthropic», and one of the founding engineers on a product that began in September 2024 as a Boris Cherny prototype able to identify what music was playing. Four source notes track her from that origin story to a public argument about what the role of a builder now is.

Her most concrete methodological advice concerns restraint. At a Fireside Chat at AIE, moderated by Simon Willison, Wu and Thariq Shihipar argued for letting Fable and, to a degree, Opus exercise their own judgement rather than dictating rules. The worked example is testing: you can write an explicit rule about when to run automated tests, but asking Fable to judge when tests are warranted works better than the hard-coded version. Willison applied the same logic to model delegation and reported his Fable allowance shrinking more slowly.

The same instinct shows up in how she describes using «Claude Code» internally. In the Every podcast with Cherny, plan mode is the single practice she puts first: do not try to one-shot complex work. Shared settings files, stop hooks, and subagents that challenge each other's findings all push determinism into specific steps rather than into a global instruction.

Wu and Cherny's July 2026 video pushes the claim further: product and engineering roles merge, and shipping becomes cheap. Commenters accepted the thesis but reframed it. If everyone can build, someone still has to be accountable for saying no.