Why AI-Heavy Teams Are Ditching Code Reviews

Ankit Jain, founder of Aviator, makes the case on Latent Space that code reviews can't survive the volume of AI-generated code. He cites Faros.ai data across 10,000+ developers showing that high-AI-adoption teams merge 98% more pull requests - but review time jumps 91%. The math breaks down quickly. His alternative is spec-driven development: humans review plans and acceptance criteria upstream, then let layered verification handle the rest - competing agents, adversarial testing agents, deterministic guardrails like type checks and contracts. Worth noting that line-by-line code review only became standard practice around 2012-2014 anyway. The shift from reviewing diffs to reviewing specs is already visible in teams running Claude Code and Codex at scale.