Cat Wu

Sources#

• Fable's judgement
• How Anthropic's product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code)

Summary#

Head of Product for Claude Code and Cowork at Anthropic. Engineer for many years before a brief VC stint, then Anthropic. Interviewing "hundreds of PMs" trying to enter AI; channels that exposure into a strong opinion that the PM role is mid-restructure (see Engineer PM Convergence).

Notable claims and stances#

• Cadence shift. Anthropic feature timelines went from 6-month → 1-month → sometimes 1 day. Achieved by removing process drag, branding most launches as research preview to lower commitment, shipping internally before externally.
• Mission > coordination. "If there's two competing priorities, we'll talk about which one is more important for Anthropic's mission." This is what removes friction at scale, not headcount or process.
• Just do things. Personal life motto. "Jobs are fake. If you understand the constraints, you can figure out what you can do and then just like try to do it quickly." Sees agency as the rare skill startups should hire for.
• Build for the current model. Easy to build for super-AGI strawman; hard to elicit max capability from today's model. Hardest PM skill is defining what the product should look like a month from now.
• Ask the model to introspect. Underrated debug technique: when Claude does something unexpected, ask why. "A lot of times just like being very curious about why the model made the decision that it did will show you what misled it so that you can fix the harness" (see Model Introspection Feedback).
• Judgement over rules. At an AI Engineer World's Fair fireside with Thariq Shihipar hosted by Simon Willison (2026-07-03, relayed second-hand in Fable's judgement): let Fable "use their own judgement rather than dictating how they should work." The worked example is testing — a user can write "only use automated testing for larger features, don't update and run tests for small copy or design changes," but "it's better to just tell Fable to use its own judgement when deciding to write tests instead." Her prompt-pruning discipline pointed at the user's prompt rather than the vendor's, and in its delegation form rather than its deletion form (see Harness Shrinkage as Models Improve).
• Build evals. Ten great evals beats a hundred mediocre ones; eval-writing is "underappreciated" and should be done by more PMs/engineers.
• 95% automation isn't automation. Push to 100% or don't bother. The last 5% is most of the work but is what makes the workflow trustable.
• Build apps you use every day, not prototypes. Customizing setups beyond a point becomes distraction — "I think the simple setups actually work better."
• Character is product. Claude's personality (low-ego, positive, lighthearted, bias-toward-action) is core to product success — Amanda's role of molding character is "harder than coding" because the task is so ambiguous.

Operational details from interview#

• ~30–40 PMs at Anthropic across research-PM, Claude Developer Platform, Claude Code, Enterprise, Growth.
• Hires engineers with strong product taste over PMs with weak engineering — many engineers on the team can take feedback on Twitter to shipped product end-to-end with no PM involvement.
• Designers on the team have front-end engineering background.
• Nightly use of Cowork to draft 20-page slide decks from Slack/Drive/Twitter context (anthropic design system pre-loaded).
• Internal stack: heavy Claude Code + Cowork, Slack ("the OS of our company"), custom internal apps built by the team for personalized workflows.
• "Applied AI" team is second-largest token spender after engineering — technical go-to-market role making prototypes for customers.

Key quotes#

• "It's very easy to build the product for the super AGI strong model. The hard thing is figuring out for the current model, how do you elicit the maximum capability?"
• "As code becomes much cheaper to write, the thing that becomes more valuable is deciding what to write."
• "If a product isn't successful, as long as it's not blocking the core use case, it's okay."
• "We can remove a lot of prompting interventions every time the model gets smarter. We actually do this every time we launch a model."

Connections#

• Claude Code — co-leads product
• Cowork — co-leads product
• Boris Cherny — tech lead partner
• Anthropic — employer
• Engineer PM Convergence — central voice
• Harness Shrinkage as Models Improve — to-do list crutch removed at Opus 4 is her canonical example
• Model Introspection Feedback — her named debug technique
• AI Native Product Cadence — embodies the practices
• Claude Character as Product — articulates why character is load-bearing
• Fiona Fung — complementary senior leader on the same products (Fung frames the engineering org; Wu is Head of Product)
• Dogfooding as Product Discipline — her lunchtime vibe-checks are the eval-form of the same dogfooding Fiona Fung calls "ant food"
• Verification as the New Bottleneck — her "ten great evals" / "push to 100%" stances are the product-side of verification becoming the bottleneck

Sources#

• How Anthropic's product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code) — Lenny's Podcast, 2026-04-23
• Fable's judgement — Simon Willison, 2026-07-03 (practitioner-opinion): the judgement-over-rules advice from the AI Engineer World's Fair fireside, relayed second-hand — no transcript or recording is in the corpus, and the attribution is joint with Thariq Shihipar without saying who said what