Anthropic confirmed that Claude Code had gotten worse after weeks of user complaints, but denied intentionally "nerfing" or degrading the model. In a Thursday blog post, the company identified three issues at the product level—a change to Claude Code's default thinking level, a cache-optimization tweak that introduced a bug, and a system prompt meant to make the tool less verbose—that contributed to a worse user experience. Anthropic stated the underlying model was not affected and that as of April 20, those issues were fixed, with steps taken including having a larger share of staff use the public build, improving its code review tool, and adding tighter controls to system prompt changes. The company thanked users for feedback and reset usage limits for all subscribers as of Thursday. The post followed weeks of complaints, including from AMD's senior director of AI Stella Laurenzo, who wrote that "Claude has regressed to the point it cannot be trusted to perform complex engineering." Anthropic also faced scrutiny on Tuesday for experimenting with removing Claude Code from the Pro plan.