Enterprise developers, led by a senior AMD AI director, are reporting quality regressions in Claude Code following a February update, particularly for complex multi-file, systems-level engineering tasks. A quantitative analysis of thousands of sessions shows shallower reasoning, with the model skipping code-reading steps before making edits. Analysts attribute the issue to Anthropic's compute capacity constraints as usage scales, noting that rate limits compound the frustration. While no mass exodus is occurring, trust is quietly eroding among power users. Analysts predict all frontier AI providers will face similar trade-offs, and suggest premium tiers with greater reasoning capacity as a potential solution. Anthropic has yet to respond to the GitHub issue.
Sort: