Google claims 75% of its new code is AI‑generated, Anthropic says roughly 80% of merged production code comes from Claude and cites an eight‑fold quarterly output increase, while OpenAI repeats a similar 80% figure. Vendors tout these volume numbers, but they merely translate lines of code into a flashier metric. These claims ignore whether AI improves speed, incidents or customer value.

Earlier studies measured outcomes: GitHub’s Copilot showed developers finishing tasks 55% faster. Later research from GitClear observed rising code churn as AI use deepened, and METR’s 2026 study initially found experienced open‑source engineers 19% slower with AI, only to later concede measurement challenges and suggest a speedup. Nevertheless, the lack of outcome data makes it hard for executives to justify AI investments beyond headline percentages.

The shift toward vanity metrics influences hiring, budgeting and layoffs; Jack Dorsey’s Block cut 40% of staff citing AI‑driven efficiency, and Atlassian trimmed 10% for similar reasons despite strong revenue. Without clear productivity evidence, organizations risk basing decisions on inflated adoption numbers rather than reliable DORA or reliability metrics, and long‑term innovation to sustain growth.