The Death of the Code Review: What the Data Actually Says — Laurie Voss, Arize AI
AI agents boosted code output 741% but software releases grew only 30% due to verification bottleneck
“Developers who enabled autonomous agents wrote 741% more code, but only released 30% more software.”
Economist-led research tracking 100,000+ GitHub developers found that AI agents created a severe mismatch: code generation is nearly 8x faster, but software release rates increased only 30% because human verification remains the gating constraint. Laurie Voss (Arize AI, ex-npm) frames this as the defining engineering problem of the AI coding era, supported by real-world examples like Stripe/Anthropic migrating a million-line Ruby codebase in a single day. The implication is that teams must rethink or automate the review and trust-verification layer, not just the coding layer.