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

Episode Summary

Executive Summary: Kat Wu, head of product for Claude Code and Co-work at Anthropic, explains how AI is radically reshaping product management: PMs must prioritize speed, taste, and clear goals over traditional long-range coordination. She describes Anthropic’s low-process, high-agency culture, the role of evals and human judgment, and how products are evolving as models get smarter and old scaffolding gets removed.

Main Topics: AI-native product management is about speed and taste (Priority: 5/5): Kat argues that PMs must shift from multi-quarter planning to rapidly defining goals, shipping weekly, and deciding what to build as code gets cheaper and model capabilities change quickly. How Claude Code and Co-work are organized (Priority: 5/5): Kat explains her role versus Boris’s: Boris is the technical/product visionary, while she drives the path to shipping, cross-functional coordination, and removing blockers. Anthropic’s shipping process and operating system (Priority: 5/5): The team uses research previews, tight launch-room workflows, weekly metrics readouts, and clear principles so engineers can move features from idea to user quickly. The future of PMs and role convergence (Priority: 4/5): Kat believes PMs, engineers, and designers are increasingly overlapping; the most valuable skill is product taste, not rigid role ownership. Using models well: harnesses, evals, and feedback (Priority: 4/5): She emphasizes using introspection, targeted evals, and trusted human feedback to understand model failures and improve product reliability. Product evolution with stronger models (Priority: 4/5): As models improve, Anthropic removes prompts, workflow crutches, and unnecessary features; newer models unlock entirely new capabilities like reliable code review. Product ecosystem and user workflows (Priority: 3/5): Kat describes when to use Claude Code, desktop, web/mobile, and Co-work, plus internal use cases like deck generation, customer briefs, and custom tools.

Key Arguments: AI has compressed product timelines from months to weeks, days, or even one day, so PMs must optimize for rapid iteration and shipping rather than long coordination cycles. Great AI PMs define the intended user, the exact problem, and the success criteria clearly, because general-purpose LLMs create ambiguity without strong product direction. Research previews reduce commitment and let Anthropic ship early, learn quickly, and iterate without overpromising on stability. The PM role now includes setting frameworks for engineers and cross-functional teams so that anyone can move an idea to launch with minimal friction. Product taste is becoming more valuable than raw coding ability because code is cheaper; the key decision is what to write and what UX to choose. Engineering background is still useful because it helps PMs understand implementation difficulty and make better prioritization decisions. Roles are merging: engineers do PM work, PMs do engineering-like work, and designers increasingly need technical depth. Human common sense, stakeholder management, and EQ remain important because models still struggle with launch complexity and tacit organizational context. Evals are underappreciated and should be used to define what good looks like and track progress, even if only a small set of high-quality evals exists. New model releases should simplify products by removing old scaffolding, not just add features, because stronger models naturally handle tasks that earlier models could not. Anthropic’s mission to bring safe AGI to humanity creates alignment and focus, enabling hard tradeoffs across product lines. Tools should be built for real daily use cases, not just demos; otherwise the leverage from AI never materializes.

Data Points: Anthropic PM headcount: around 30 to 40 PMs - Kat estimates the company’s PM organization size. Feature timelines before AI: 6 to 12 months - Typical product planning horizon before AI accelerated development. Feature timelines now: 1 month, 1 week, or even 1 day - Kat says some features now ship on dramatically shorter cycles. PM team weekly cadence: every week - The team does weekly metrics readouts with the entire team. Launch speed target: less than a week - Idea-to-launch for many features in the Cloud Code process. Research preview commitment: 1 to 2 weeks - Features are often shipped first as research previews to lower commitment. Anthropic team-lunch feedback group: about 5 people - Kat says a small trusted group gives especially useful model feedback. AI model jumps increase usage: every time there is a model jump - Token cost per engineer/knowledge worker rises when models improve. Custom deck generation time: hours reduced to seconds/few seconds - Co-work can create tailored decks far faster than manual work. Tokens vs salary: still much lower than average engineer salary - Kat notes token spend is rising but remains below salary costs for most workers. Product org principle: almost no process - Anthropic emphasizes removing barriers to shipping. Decision hierarchy: safe AGI for all of humanity - Mission is positioned above individual product/org goals.

Pivotal Quotes: "It is very hard to be the right amount of AGI pilled." — Kat Wu: She describes the challenge of building for current models without overfitting to a future superintelligent version. "As code becomes much cheaper to write, the thing that becomes more valuable is deciding what to write." — Kat Wu: Her core thesis on why product taste is becoming the key PM skill. "We want to remove every single barrier to shipping things." — Kat Wu: She explains Anthropic’s low-process, high-agency operating style.

Implications: PMs must become faster, more technical, and more taste-driven. For AI teams, the winners will be those who tightly couple model capabilities, evals, and user feedback while shipping early and simplifying as models improve.