Simon Willison · 2026-07-21 · notable
Simon Willison — Fireside chat with Anthropic's Claude Code team on tools and safety
Simon Willison published an 8,000-word transcript of his AI Engineer World's Fair chat with Anthropic's Cat Wu and Thariq Shihipar on Claude Code, Claude Tag, credential injection, and dropping few-shot system prompts.
Simon Willison turns his AI Engineer World's Fair chat with Anthropic's Claude Code team into a searchable, 8,000-word annotated transcript.
What is it?
Simon Willison's July 21 post is a full annotated transcript of the fireside chat he ran with Anthropic's Cat Wu (Product) and Thariq Shihipar (Engineering) at the AI Engineer World's Fair. The write-up covers Claude Code, Claude Tag, coding-agent security, and how the team decides what ships.
How does it work?
Willison timestamps the video, bolds the key claims, and inserts editorial commentary between quotes. Notable disclosures: Claude Tag now lands about 65% of Anthropic's own product-engineering PRs; the team cut example-heavy few-shot system prompts by ~80% because Fable 5 and Opus 4.8 do worse with them; Auto mode uses a Sonnet classifier to manage dynamic permissions; credential injection lets agents call authenticated APIs without ever seeing the API key.
Why does it matter?
It is one of the clearest first-party pictures of how Anthropic is actually building with Claude Code internally, at a moment when every large team is trying to figure out where to trust an agent and where to keep a human in the loop. Willison's annotations let you jump to any specific claim, so the post doubles as reference material rather than only a talk write-up.
Who is it for?
Anyone building with Claude Code or a similar coding agent; teams wiring credentials and permissions into agentic workflows.
Try it
https://simonwillison.net/2026/Jul/21/cat-and-thariq/