Why Anthropic Replaced Engineering Managers With Five Lifecycle Roles

In this 50-second commentary, Gabe Larsen reports that Anthropic threw out the traditional engineering org chart on its Claude Code team. Instead of engineering managers and a PM hierarchy, the team runs on five roles tied to where a product sits in its lifecycle: prototype, build, grow, maintain, improve.

The video matters because it frames the reorg as a consequence of AI-driven productivity, not a management fad. Larsen's argument is that when individual engineers can ship at 10 to 30 pull requests a day, the coordination layers that org charts were built to provide stop paying for themselves.

The reorg: five roles, no managers

Larsen opens by saying Anthropic did something "that should probably scare every middle manager." The Claude Code team dropped engineering managers and the PM hierarchy entirely, replacing them with five roles based on product lifecycle stage: prototype, build, grow, maintain, and improve.

The video is a single talking-head shot throughout, with green-highlighted captions carrying the key phrases. No source document or org chart is shown on screen.

The math that changed

Larsen's explanation for the reorg is throughput. He says the average software engineer ships a few pull requests a week, while AI-native builders can ship 10 to 30 a day. When one person can take an idea from concept to production, he argues, you no longer need layers of people coordinating handoffs.

Org charts as an information problem

The closing argument reframes org charts as a workaround for slow information flow. Larsen says org charts existed for years because information moved slowly between people, and AI has made that problem almost disappear. He ends with a prediction: the biggest change from AI will probably not be the software, it will be the company itself.

Key Takeaways

• Anthropic's Claude Code team replaced engineering managers and PM hierarchy with five lifecycle-based roles: prototype, build, grow, maintain, improve.
• Larsen claims the average engineer ships a few pull requests a week, while AI-native builders can ship 10 to 30 a day.
• When one person can go from concept to production, coordination layers built around handoffs lose their purpose.
• Org charts were a response to slow information flow; Larsen argues AI has largely removed that constraint.
• His closing prediction: AI's biggest impact will be on company structure, not the software itself.