AI Writes More Code Than Ever, But Human Code Review Is Not Keeping Up

Tech giants including Google and Microsoft report that AI now generates a significant and growing share of new code, with some estimates suggesting up to 75% at Google as of 2026. While engineers still formally approve this code, researchers warn that human presence in a review process no longer guarantees genuine comprehension of what is being approved. Code generation has become far cheaper and faster, but the cognitive capacity required to review code has not scaled accordingly, creating a widening bottleneck. Industry data from DORA and GitClear show that higher AI adoption correlates with increased delivery instability, a near-collapse in refactoring activity, and a sharp rise in duplicated and error-masking code. The core concern is that syntactically correct, well-commented AI-generated code can still be logically wrong in ways that only become visible with deeper contextual knowledge reviewers may not apply under time pressure.