Lenny's Podcast: How Anthropic's product team ships in days, not months

What the video covers

Cat Wu, Head of Product for Claude Code and Cowork at Anthropic, appears on Lenny’s Podcast to describe how Anthropic’s product team actually works and why its development cadence has changed so drastically. The episode was published in April 2026. Wu explains how the team reduced its cycle time from six months per release to one month — and in some cases one week or one day. The conversation covers what made that compression possible, what it requires of product managers, and what she expects the PM role to look like as agent capabilities continue to develop.

Who it’s for

Product managers and heads of product at organizations that have started shipping AI features or building AI products, and are finding that their existing planning and release processes don’t fit the pace the work demands. Also useful for PMs earlier in their career who want a concrete model of what AI-native product work looks like at one of the most closely watched AI organizations in the industry.

Key takeaways

1. Cycle compression changes the PM role fundamentally, not incrementally. When a team ships daily instead of quarterly, the PM job shifts from coordinating a roadmap to enabling the team to make good decisions in real time. Wu describes her work as maintaining shared context and clearing obstacles as work happens — a fundamentally different function from traditional product planning, not just a faster version of it.

2. Access to unreleased models creates a product feedback advantage that is hard to replicate externally. One factor behind Anthropic’s shorter timelines is that the product team gets early access to new model versions before public release. The gap between what the model can do today and what it will be able to do in three months is something they can observe directly, which changes how they scope features and set expectations. Teams building on external APIs do not have this.

3. The future PM role involves managing fleets of AI agents simultaneously. Wu describes a direction in which PMs run fifty or a hundred simultaneous agent workflows. This is not a description of her team’s current daily work — it is a framing she uses to think about what infrastructure and tooling decisions matter now versus later. For product managers thinking about what to build or invest in, it is a useful frame for distinguishing short-term and long-term decisions.

4. Shared metrics, not shared plans, are the mechanism for alignment in fast-moving teams. When a plan changes weekly or daily, it cannot be the source of team coordination. Wu emphasizes that clearly defined outcome targets are what keeps a fast-moving team coherent — not a synchronized roadmap, which becomes stale too quickly to be useful at that cadence.

Worth watching if…

Your organization has started talking about accelerating its release cadence or building AI-native products, but you are not sure what that actually requires of the product function. Wu gives a ground-level account of what changes and what stays the same when the development cycle shortens from months to days — and where traditional PM instincts still apply.