AI-generated code
Source code written by a model rather than a person; its share of commits is rising and its quality profile differs measurably from human code.
Estimates of the share vary by surface — 50% of characters at Google via completion, ~95% of some YC startups' codebases. Measured differences from human code: more duplication (GitClear), higher defect rates in some studies (Uplevel +41% bugs), and a 44% security-flaw rate that has not improved in four years of models (Veracode 2026). Provenance tracking of AI-generated code is emerging as a compliance requirement.