AI repos lost half their human code review since 2024, analysis of 2000+ projects finds
ronbodkin · x · 2026-10-09
In a thread replying to ghadfield, ronbodkin proposes a systematic review mechanism for AI-generated code contributions:
- His team analyzed over 2,000 leading open source AI repositories and found human review comments per line of code fell by roughly half since early 2024, as reviewers are stretched thin.
- Core argument: the real threat is systematic loss of understanding, not occasional lapses. If one in ten contributions would fail, 22 random audits a quarter catch at least one 90% of the time.
- Mechanism: on a failing grade, releases depending on the contribution are paused until understanding is rebuilt, even for internal use; repeated failures suspend a unit's release authorization — a clear gate without arbitrary automation limits.
- The responsible team must explain what the contribution does, how it works, and its dependencies (not model internals), in a blameless thesis-defense-style meeting, with the level of understanding graded.
More from coding & agent
- Dev jokes Hone's launch 'took my job and best friend' as AI agents run ops — graceisford · 2026-10-09
- LlamaIndex explains why Markdown is its default parsing output (HTML for merged headers) — llama_index · 2026-10-09
- 3 engineers, $1M and 1.5T tokens: Compound V2 rebuilds Office suite for AI agents — peterjliu · 2026-10-09
- While waiting for his AI coding limit to reset, a dev went back to hand-writing Go — haydendevs · 2026-10-09
- Orchestrator rebuilds its agent terminal: 13ms typing lag, 5x faster output — julianweisser · 2026-10-09
- Dev releases free 'AI Software Factory' ebook on agentic SDLC — Pavan_Belagatti · 2026-10-09