Intel / podcast

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

Source: lennysnewsletter.com
Author: Lenny Rachitsky
Date: Published Apr 23, 2026

Context

Product teams need concrete descriptions of good model behavior to improve it. Cat Wu, who leads product for Claude Code and Cowork at Anthropic, describes combining informed user feedback with targeted evaluations. A teammate might notice that a model tests too little or writes questionable memories. That observation becomes a hypothesis to check against broader data. Small sets of carefully chosen tasks then make the desired behavior explicit. This connects a user's impression to something the team can investigate and measure.

Terms in this piece · Glossary

eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.

Key quotes

“We want to remove every single barrier to shipping things.”
Cat Wu

“So an extreme example is, if Cloud Code failed but Anthropic succeeded, I would be extremely happy.”
Cat Wu

“As code becomes much cheaper to write, the thing that becomes more valuable is deciding what to write.”
Cat Wu

“And if an automation doesn't work 100% of the time, it's not really an automation.”
Cat Wu

“I think the hard thing is figuring out for the current model, how do you elicit the maximum capability?”
Cat Wu

Chapters

00:00Introduction to Cat Wu
01:29Working with Boris Cherny
04:29What Anthropic looks for when hiring PMs
06:18How to help your teams move fast
08:58How PRDs and roadmaps have evolved at Anthropic
10:28The Mythos model and Anthropic's shipping velocity
11:54What happened with the Claude Code source code leak
12:53Integrating with OpenClaw
14:19How the PM team is structured at Anthropic
15:42How engineer and PM roles are merging
17:54Why product taste is the most valuable skill
20:10Where human brains will continue to be useful
22:23How to stay sane in constant chaos
24:16What gets sacrificed when you ship so fast
27:47The /powerup command
28:32Why Anthropic has been so successful
32:28When to use Claude Code vs. Desktop vs. Cowork
35:58Tips for getting started with Cowork
38:44Demo: Using Cowork to build slide decks overnight
41:48Cat's PM tech stack and internal tools
46:47Which teams use the most tokens
51:15The emerging skills PMs need for AI companies
55:00Why building evals is underappreciated
58:44Why Claude's character and personality matter so much
1:00:44How new models force product changes
1:05:11The vision for Claude Code and Cowork
1:07:22Advice for thriving in an AI-driven world
1:09:18Why 95% automation isn't good enough
1:11:58Build apps you use every day, not prototypes
1:13:41The divide between AI skeptics and believers
1:15:19Lightning round

Read the source www.lennysnewsletter.com