A Fireside Chat with Cat and Thariq from the Claude Code team

Simon Willison sat down with Cat Wu and Thariq Shihipar from Anthropic's Claude Code team at the AI Engineer World's Fair, and the transcript he posted afterward is full of the kind of internal detail that usually stays behind closed doors. The headline number: Claude Tag, Anthropic's new collaborative Slack integration for Claude, is now landing 65% of the product engineering team's pull requests. Not half. Not a third. Two-thirds of the code shipping through that team is going out via a chatbot living in Slack channels, monitoring bug reports and proactively opening fixes without anyone tagging it in.

What struck me most is how casually both Cat and Thariq describe the collapse of old workflows. Cat talked about how, two years ago, a product manager might spend six months aligning with cross-functional teams on a spec before a single line of code got written. Now that timeline has shrunk to as little as a week, and her advice to engineers is blunt: spend less energy on execution, more on developing product taste, because building things is no longer the bottleneck. Thariq went further, declaring that rewrites — long treated as an engineering sin, per the old Mythical Man-Month wisdom — are now good. Anthropic apparently rewrote Bun in Rust and has shipped it internally, treating the existing codebase as a spec to distill rather than a monument to preserve.

The system prompt engineering angle is quietly one of the more useful nuggets here. According to the team, stuffing examples into a system prompt is no longer best practice for models like Fable 5 or Opus 4.8, and Anthropic actually shrank the Claude Code system prompt by 80%. Lists of