Anthropic: AI Writes 80% of Code, but Verification Costs Soar 25x

Anthropic revealed that Claude now generates around 80% of its internal code, but the broader impact on its engineering pipeline tells a more complex story. Over six months, the company's test suite grew tenfold while CI jobs increased 25 times, signaling that verification — not code generation — has become the new bottleneck. The author argues that AI compresses the development constraint so rapidly that downstream systems like code review, CI capacity, and deployment controls struggle to keep pace. Anthropic addresses this partly through a deterministic test-impact analysis service that selects relevant tests per change, rather than running full regression suites every time. The piece concludes that as AI agents iterate autonomously, feedback latency becomes a throughput issue, reshaping how quality assurance must operate at scale.