Anthropic's ci system overloaded by ai-driven code submissions

Internal data from Anthropic reveals that AI models like Claude are significantly impacting the company's development processes. According to the data, Claude contributes to 80% of the company's code, with engineers delivering eight times the average code volume compared to the period between 2021 and 2025. Tasks such as code reviews and merge approvals are also largely automated by Claude. The AI's coding behavior, however, differs from human patterns. It submits PRs with extreme granularity, leading to a tenfold increase in test cases and a 25-fold surge in CI run tasks within six months. Anthropic attempted to address the issue with three patches, but they failed. Eventually, the company rebuilt its service using a new distributed stateless architecture, which resolved the CI overload problem. This incident highlights the broader impact of AI on software engineering systems, as pipelines designed for human workflows struggle to handle AI-driven coding rhythms.