Claude Is Writing Claude. And That Should Terrify You a Little.

Reflect w/ Ed Fassio — where AI tells the stories that matter. Anthropic just published internal data that changes the conversation. As of May 2026, more than 80% of the code merged into Anthropic's own codebase was written by Claude. Engineers are shipping 8 times as much code per day as they were in 2024. One Anthropic employee hasn't written a single line of code himself in five months. The person building the AI. The Anthropic Institute is calling this the early stages of recursive self-improvement — an AI system actively accelerating the development of its own successor. The task duration that AI can reliably handle has been doubling every four months. Tasks taking days could come into range this year. Tasks taking weeks by 2027. Julius and Hale break down what the data actually says, what the governance gap looks like when review scales linearly and output scales exponentially, and why the company most focused on AI safety just handed the keys to the AI. Your Move: The 80% number isn't a projection. It's happening now. Build the review infrastructure before you build the generation infrastructure. — Reflect w/ Ed Fassio | reflectpodcast.com