Simon Willison Publishes 8,600-Word Transcript of a Claude Code Team Fireside Chat — Agent Security, Evals, and How Anthropic Dogfoods Its Own Tools

simonwillison.net·high signal

On July 21, 2026 Willison posted a full write-up of a fireside chat with Cat and Thariq from Anthropic's Claude Code team, recorded at the AI Engineer World's Fair, covering Claude Code, Claude Tag, Fable, coding-agent security, eval design, tool design, and Anthropic's internal usage patterns. At 8,609 words it is the longest primary-source account of Claude Code's design philosophy published to date, and it is a transcript rather than secondary coverage. For builders it is the closest thing to a design doc for how the team that ships the agent thinks about tool surface area and eval discipline.