Simon Willison hosted a fireside chat at the AI Engineer World's Fair with Cat Wu and Thariq Shihipar from Anthropic's Claude Code team. Key highlights include: Claude Tag (Anthropic's Slack-integrated collaborative coding agent) now lands 65% of product engineering PRs internally; the Claude Code system prompt was reduced by 80% for frontier models like Fable and Opus 4.8, as examples and 'do not do X' instructions are no longer best practice; rewrites are now considered good practice when paired with solid test suites; auto mode has been hardened with thousands of evals and red-teaming to defend against prompt injection and data exfiltration; and Anthropic uses internal user retention as the bar for deciding whether to ship features externally. The conversation also covers tool design philosophy (fewer, more distinct tools), credential injection for secure agent access, and how engineers should respond to AI-induced feelings of loss by being more ambitious.
