Harvard paper: code review, not code writing, is the AI coding bottleneck
11/20/2026 — 11/21, 08:20·2 sources (all)·2 reports (all)
Key takeaways
- A working paper by two Harvard PhD candidates, using data from 718 companies and 725,938 employees, found AI coding agents raised lines of c…
Background & analysis
On November 20, 2026, a report described a working paper by two Harvard doctoral candidates that measures what AI coding agents actually deliver in real organizations. Drawing on a sample of 718 companies and 725938 employees, the study found that after adoption, the average monthly lines of code written per person rose from 5017 to 6511, an increase of roughly 30%, while the number of commits grew by 20% and Pull Requests by 23%. However, when the authors cross-checked activity against Jira data, the work was not being finished in step with the added output. They therefore concluded that the current bottleneck sits in the Code Review stage.
On November 21, 2026, DEV Community's AI section covered the same research and named the two researchers, Fiona Chen and James Stratton. That report said the pair examined data from 655 companies and more than 720,000 workers, and likewise found that AI tools help people write considerably more code while the volume of finished software being shipped barely changes, with the cause pointed at code review. The report also stressed that companies handed developers coding agents while keeping their old working processes, and argued this is the key reason the expected efficiency gains did not materialize.
The two reports do not agree on the sample size: one cites 718 companies and 725938 employees, the other 655 companies and more than 720,000 workers. No source available so far explains the discrepancy, so the figures remain an open discrepancy rather than a settled figure.
AI-generated from 2 reports · updated 1 day ago
Latest turnA Harvard study covering 655 companies and more than 720,000 workers found that AI coding tools sharply increase the volume of code written but barely move how much software actually ships, pointing to code review as the bottleneck. Companies handed developers coding agents while keeping their old workflows, which the authors treat as the reason the speedup never showed up.
- Heat index
- 76
- All sources
- 2
- All reports
- 2
- First seen
- 2 days ago
Reports on this story headlines open the original
A Harvard study covering 655 companies and more than 720,000 workers found that AI coding tools sharply increase the volume of code written but barely move how much software actually ships, pointing to code review as the bottleneck. Companies handed developers coding agents while keeping their old workflows, which the authors treat as the reason the speedup never showed up.
DEV Community · AIAI score 53
A working paper by two Harvard PhD candidates, drawing on 718 companies and 725,938 employees, found AI coding agents raised monthly lines of code per person from 5,017 to 6,511, with commits up 20% and pull requests up 23%. But Jira data showed the underlying work was not completed at a matching rate, leading the authors to name code review as the current bottleneck.
开源中国AI score 62
Other stories people are talking about
- Heat index 1143RisingAnthropic Reaches $1.6 Trillion Market Value in First Trading Week After IPO21 sources
- Heat index 1011Anthropic Mathematician Says Claude Formalized the Generalized Riemann Hypothesis21 sources
- Heat index 622Anthropic IPO total raise reaches $93.5 billion after greenshoe exercise20 sources
- Heat index 596RisingThinking Machines Lab Open-Sources Inkling 2 Mini with 32GB Model Card Recommendation10 sources
- Heat index 490RisingSEC probes Nvidia accounting practices as shares post seven-week slide9 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion

