Link: Promoting AI agents via DHH’s newsletter
For writing code in production code bases, DHH has started using OpenCode (a terminal interface for coding agents) with the newest available models (Claude Opus 4.5, Codex 5, Gemini 3, MiniMax M2.1, GLM-4.7) to produce its own contributions that he reviews and guides in addition to answering his questions or checking his work. He uses coding agents to fix bugs, finish features and get drafts of new initatives, but isn’t close to having agents write 90%+ of his code. He doesn’t use the in-editor AI autocomplete experience (GitHub Copilot and Cursor) becuase he wants to finish his own thoughts when he’s coding.
At the end of last year, AI agents really came alive for me. Partly because the models got better, but more so because we gave them the tools to take their capacity beyond pure reasoning. Now coding agents are controlling the terminal, running tests to validate their work, searching the web for documentation, and using web services with skills we taught them in plain English. Reality is fast catching the hype!
This is all very evident if you’ve tried to employ any of the new models — especially Claude Opus 4.5, Codex 5, Gemini 3, and even the Chinese open-weight models like MiniMax M2.1 and GLM-4.7 — in one of the modern terminal harnesses that give them access to all these autonomous powers. The code being produced by this new breed of AI is leagues ahead of where their predecessors were at the beginning of 2025.
I’ve thoroughly enjoyed putting them all to work in OpenCode, which is a terminal interface for coding agents that allows you to seamlessly switch between all of the models, capture your sessions for sharing, and simply looks astounding when theme-matched with the rest of Omarchy (where we’re making it a default in the next version!).