How Anthropic’s product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code)
TL;DR
Cat Wu, Head of Product at Anthropic, reveals how Claude Code ships features in days instead of months by redefining PMs as velocity enablers rather than coordinators, leveraging research previews to eliminate shipping barriers, and prioritizing product taste over process as code becomes increasingly cheap to write.
🧠 The Evolving PM Role in AI 3 insights
Product taste becomes the primary skill
As AI makes code cheaper to write, the PM's value shifts from managing development to deciding what to build and eliciting maximum capability from current models.
From roadmap alignment to velocity optimization
Modern AI PMs must abandon multi-quarter roadmap coordination in favor of shortening the idea-to-shipping cycle to one week or even one day.
Convergence of product and engineering
Successful AI teams hire engineers with strong product taste or PMs who code, allowing individuals to ship end-to-end features with minimal cross-functional overhead.
⚡ Operational Velocity & Process 3 insights
Research preview as default
Anthropic ships nearly all features branded as 'research previews' to reduce long-term support commitments and remove psychological barriers to daily releases.
Principles over documents
The team relies on shared principles and weekly metrics readouts rather than lengthy PRDs, enabling autonomous decisions without PM bottlenecks.
Evergreen launch pipelines
A dedicated Slack room allows marketing, docs, and DevRel to turn around announcements within 24 hours of engineering completion, creating frictionless cross-functional flow.
🏗️ Organizational Structure 3 insights
Complementary leadership split
Cat Wu manages cross-functional alignment and the path to 3-6 month vision, while tech lead Boris defines the long-term AGI-level product vision.
PM team organization
Anthropic's ~30-40 PMs are organized across research (model launches), developer platform (APIs), Claude Code (core product), enterprise (security/compliance), and growth.
Low-process culture
The explicit goal is removing every single barrier to shipping, empowering any engineer to take an idea from concept to production in under a week.
💼 Strategic Decisions 2 insights
OpenClaw policy change rationale
Anthropic restricted third-party OpenClaw usage because the $200 subscription was designed for individual use, not subsidized third-party compute, forcing a prioritization of first-party products and API customers.
Source code leak response
A recent leak of Claude Code's source code resulted from human error during PR review; the company hardened safeguards and treated it as a process failure rather than an individual one.
Bottom Line
To compete in AI product development, eliminate coordination-heavy roadmaps in favor of clear goals and team principles that empower engineers to ship research previews daily, while hiring for product taste to maintain quality without sacrificing speed.
More from Lenny's Podcast
View all
Why half of product managers are in trouble | Nikhyl Singhal (Meta, Google)
Product management is undergoing a radical shift from 'information movers' to AI-empowered builders focused on judgment, creating a bifurcated market where AI-first talent thrives while mechanical roles face obsolescence in the coming 12-24 months.
Hard truths about building in the AI era | Keith Rabois (Khosla Ventures)
Keith Rabois shares contrarian frameworks for the AI era, arguing that product managers must evolve into CEO-like strategists, that high-performance cultures must sacrifice psychological safety for results, and that founders should ruthlessly prioritize hiring "barrels" (independent drivers) over "ammunition" (executors) to avoid scaling burn without output.
Anthropic’s $1B to $19B growth run: how Claude became the fastest-growing AI product in history
Anthropic grew from $1B to $19B ARR in just 14 months by prioritizing model excellence over traditional marketing, with the growth team spending 70% of their time firefighting 'success disasters' while using Claude itself to automate experimentation through their 'CASH' initiative.
An AI state of the union: We’ve passed the inflection point & dark factories are coming
AI coding agents crossed a reliability threshold in November 2025, enabling engineers to produce 10,000 lines of working code daily without typing, while the industry moves toward 'dark factories' where AI handles all coding and testing without human review, raising urgent questions about safety and institutional overconfidence.