David Heinemeier Hansson

AI coding agents have evolved significantly, moving beyond autocomplete to autonomous tools that can control terminals, run tests, and search documentation. Modern models like Claude Opus 4.5 and Gemini 3, when used with terminal interfaces like OpenCode, can produce production-grade code contributions. While not replacing programmers entirely, these agents enable supervised collaboration where developers review outcomes rather than write every line. The shift represents a paradigm change from pair programming to team-based workflows, though claims of 90%+ AI-generated code remain unrealistic for quality professional work.

Promoting AI agents

And still not worth the RAM shortages

ai coding is getting wild, not just autocomplete anymore. claude 4.5 and gemini 3 can actually run stuff and contribute code. still need humans to check it but kinda feels like coding with a super chill teammate

AI code quality is still not up to my standards