Harvard paper: AI coding agents boost code 30% but PRs take 49% longer to merge
blaizedsouza · x · 2026-10-12
A Harvard working paper (Fiona Chen & James Stratton) uses 300M work events — GitHub activity, Jira issues, calendar data — across 718 firms with staggered difference-in-differences to study AI coding assistants and agents.
- Teams wrote 30% more lines of code after adopting agents, but PRs took 49% longer to merge
- Share of PRs sent back for changes nearly doubled; review comments rose 35%, with more engineers in reviews
- No statistically significant increase in completed Jira issues or larger projects
- Productivity gains don't fully pass through to output: code review becomes the bottleneck
More from AGI Musings
- Karpathy: only ~5,000 people have seen what AI can really do — sourdub · 2026-10-13
- Existential-risk debate misses the elephant: AI's toll on young people — BachFrancis · 2026-10-13
- mayfer: turn-based RL can't close AGI's gap — realtime perception is required — mayfer · 2026-10-13
- Outsiders wander into repligate's thread baffled by why anyone clings to older AI models — repligate · 2026-10-13
- Gamers are getting AGI-pilled like programmers did, as Opus 5.5 mods Halo via decompilation — JeffLadish · 2026-10-13
- David Patterson: all cars will be autonomous by 2030, saving 40,000 American lives a year — davidpattersonx · 2026-10-13