Anthropic admitted that its own engineering missteps caused a monthlong decline in Claude Code performance, following weeks of pushing back on user complaints and pointing to latency and token use changes before publishing a detailed postmortem. The company’s April 23 postmortem detailed three separate issues. On March 4, Anthropic changed Claude Code’s default reasoning effort from high to medium to reduce long latency, later calling that the wrong tradeoff and reverting it on April 7. On March 26, a caching change meant to clear older thinking from idle sessions had a bug that made clearing happen on every turn, making the model look forgetful and repetitive while using more of a customer’s limit; that fix shipped April 10. On April 16, a system prompt instruction told Claude to keep text between tool calls to 25 words or fewer, with final responses capped at 100 words unless more detail was required; Anthropic said it hurt coding quality and removed it on April 20. All three issues were marked resolved in Claude Code version 2.1.116, and the API was not affected. The postmortem also said Claude Agent SDK and Claude Cowork were touched by broader product-layer issues. Fortune reported a senior AI executive at AMD called Claude "unusable for complex engineering tasks" during the worst stretch.