One pull request, fifty review rounds: what Anthropic changed in its AI reviewer
The Claude Code team's account of its own review pipeline is mostly about one failure, an agent and a reviewer that never stop, and the cheap fixes that cut it.
6 min read