An opinionated guide to which AI to use to do stuff
Simon Willison's blog post reflects on the changing landscape of AI tool recommendations, specifically tracking Ethan Mollick's guide. A year ago, the guide focused on chat-based interactions with models like o3, Claude 4 Opus, and Gemini 2.5 Pro, along with Deep Research. Now, the emphasis is on agentic systems capable of performing hours of human work in one go. Notably, Gemini has been removed from the list because Google still lacks an established entry in the Codex/ChatGPT Work/Cowork category, and Gemini Spark has yet to prove itself. Willison provides a detailed explanation of the confusing naming conventions for agent modes in ChatGPT and Claude. In ChatGPT, the two agent modes are Work and Codex; in Claude, they are Cowork and Code. These names do not map onto each other in an intuitive way. He clarifies that giving AI access to your computer requires downloading the ChatGPT or Claude apps and selecting a mode. The modes operate differently depending on the platform: ChatGPT Work on a mobile device is distinct from ChatGPT Work in the desktop app, where it functions as a less intimidating skin on top of Codex. A key difference is that flipping ChatGPT mobile from Chat to Work mode allows its Code Interpreter container to access the internet, which is otherwise restricted.
Developers must navigate confusing agent mode naming across platforms to effectively leverage AI for complex tasks.