Simon Willison — 2026-07-27#

Highlight#

Today’s most notable takeaway is Simon’s commentary on the messy industry transition from standard chat-based LLMs to agentic systems that can directly operate our computers. His breakdown of the spectacularly confusing naming conventions across ChatGPT and Claude’s desktop agent modes is a highly practical read for developers trying to navigate these new local execution environments.

Posts#

An opinionated guide to which AI to use to do stuff Simon reviews Ethan Mollick’s updated AI guide, noting a massive shift over the past year from standard chat interfaces to agentic systems capable of executing hours of real work. He highlights how Google’s Gemini has fallen off the recommended list because tools like Gemini Spark remain unproven, while ChatGPT and Claude are pushing forward with powerful—yet confusingly named—desktop control modes like “Codex” and “Cowork”. For those using these tools daily, Simon shares a brilliant tip: switching the ChatGPT mobile app to “Work” mode removes the internet restrictions from its Code Interpreter container.