Dave Winer (@dave on Micro.blog) has been extolling the virtues of AI agents to help with coding for months now. So when he writes:
I would like to have one day when Claude Code doesn’t fuck everything up by doing things it is explicitly prohibited from doing. I find things that were changed in critical parts of the program, that were indefensible.
I can’t help but wonder if some sort of tide is beginning to turn.
That said, at my work we’re about to start a trial of using AI agents, with Codex rather than Claude. So something to be cautious about.