Simon Willison Publishes Annotated Transcript Of Claude Code Team Interview
The post points readers to a longer conversation with Anthropic's Claude Code team, and a few claims from it are already getting passed around on X.
Entities: Simon Willison, Ethan Mollick
Simon Willison says he has published an annotated transcript of a fireside chat with Cat Wu and Thariq Shihipar from Anthropic's Claude Code team, recorded at aiDotEngineer and also available as a YouTube video. In follow-up posts on X, Willison highlighted two details from the conversation: that "Claude Tag," which he describes as Claude Code via Slack, is already landing 65% of the team's product-engineering pull requests, and that Claude Code's own system prompt recently shrank by 80%.
Combined views
166.2K
8 posts, first seen 3h ago
Reactions from ranked influencers
8 posts
I interviewed @trq212 and @_catwu from the Claude Code team at @aiDotEngineer a couple of weeks ago - the video is now out, so I've published an annotated transcript of our conversation https://simonwillison.net/2026/Jul/21/cat-and-thariq/
We talked about Claude Code, Claude Tag, Fable, coding agent security, evals, tool design, and how Anthropic use these tools themselves Claude Tag (Claude Code via Slack) is already landing 65% of the product engineering PRs for the Claude Code team!
Here's my full write-up of our conversation - or jump straight to 21:24 in the YouTube video here: https://www.youtube.com/watch?v=uU5Gv2h8-9g&t=1284s
I interviewed @trq212 and @_catwu from the Claude Code team at @aiDotEngineer a couple of weeks ago - the video is now out, so I've published an annotated transcript of our conversation https://simonwillison.net/2026/Jul/21/cat-and-thariq/
A big thing about Fable (and Sol, though a little bit less) is that you need to be really careful about treating it the way you would a less capable model. Skills that worked well for Opus often make Fable outputs worse, and too many lists of negative instructions can do the same
It really is a step change in how you approach getting work out of a system, requiring as much thinking as moving from GPT-4o to GPT-5/o3. And, of course, we don't have a lot of accumulated knowledge or clear studies about how to prompt these more powerful systems yet.