Internal data from Anthropic shows that the AI model Claude is responsible for writing 80% of the company’s code. Engineers are now delivering code volumes eight times higher than the average recorded from 2021 to 2025. Claude also takes on extensive responsibilities for code reviews and merge approvals. However, AI’s programming approach differs significantly from that of human developers. Claude tends to submit highly granular code change requests (PRs) continuously, 24/7. This pattern has led to a tenfold increase in test cases. As a result, the continuous integration (CI) system became overwhelmed within just six months, with task volumes surging by 25 times. Anthropic tried several mitigation strategies, such as adding processing cores, implementing sharding, and enforcing daily forced restarts, but none of these measures proved effective. Eventually, the company followed Claude’s suggestion to completely overhaul the service architecture using a distributed, stateless design. This change successfully resolved the CI overload. This incident highlights that the impact of AI programming has expanded beyond individual developers, now affecting the capacity of entire software engineering systems.