Claude Code shows the operating loop

Simon Willison published an annotated interview with Anthropic's Cat Wu and Thariq Shihipar covering Claude Code, Claude Tag, Fable, security, evals, tool design, and internal usage patterns. Why it matters: The model is only one part of the product. The interview is useful because it exposes the habits around prompts, permissions, evals, and tool boundaries that decide whether a coding agent becomes reviewable work.

Try this: Extract one operating rule from the transcript before the next coding-agent task: what needs permission, what evidence proves the change, and when a human should interrupt the loop.