source: simon willison: a fireside chat with cat and thariq from the claude code team
level: technical
cat wu and thariq shihipar from anthropic's claude code team shared how their daily work has shifted with coding agents. a year ago, they closely monitored every permission prompt. now, with models like fable, they delegate routine implementation and focus on creative work and product decisions. thariq noted that outputs are so high quality that he pushes for the best work faster, even using fable to edit its own launch video.
they challenged old software engineering rules. cat said the value of product sense has risen because the time from idea to implementation dropped from months to a week. thariq argued that rewrites are now good, contrary to past wisdom. he explained that a codebase acts as a spec, and agents can distill or recreate it. they rewrote parts of claude code in rust and ship it internally.
claude tag, a slack integration, lands 65% of the team's product prs. it works proactively, monitoring channels and opening prs without manual prompts. it also supports team memory and multiplayer sessions, letting non-engineers query metrics or get feature recordings. the team uses internal user retention to decide which features ship, and they were surprised by the popularity of remote control, which lets users manage claude code sessions from a phone.
why it matters: these practices show how ai coding agents shift engineering toward higher-level work, team collaboration, and automated quality checks, changing how software teams operate.
source: simon willison: a fireside chat with cat and thariq from the claude code team