An update on recent Claude Code quality reports Anthropic

There were grumbles about Claude Code getting less smart and here Anthropic shares what they changed that caused this and what they did to fix it. It is an interesting read because the three changes span a change in a default, an actual bug, and a reasoning change. When I read this it feels like the kind of thing that agentic engineering teams are going to need to be really good at. It also strikes me as a difference between products and agents. If you build a product and ship it, minus bugs it will behave deterministically. The last item on Claude's issue was a change in behavior that impacted coding performance. That isn't a feature change. This is what "performance management" for agents looks like.