# Coding agents raised code output, not finished work | QANATONE > A Harvard working paper on 700+ software firms finds AI coding agents raised code output about 30% but did not significantly increase… Kaynak: https://www.qanatone.com/en/haber/harvard-kodlama-ajanlari-calismasi/ --- [News](https://www.qanatone.com/en/haber/) # Coding agents raised code output, not finished work 12 Oct 2026 3 min read News · Science A Harvard working paper on 700+ software firms finds AI coding agents raised code output about 30% but did not significantly increase completed work, as the bottleneck moved to code review. ## What happened A working paper by Harvard's Fiona Chen and James Stratton studied how AI coding agents affect real software teams. They analyzed about 300 million work events across more than 700 firms from 2021 to March 2026, combining Jellyfish engineering telemetry with GitHub activity. It has not yet been peer reviewed. ## Findings After agents arrived, firms produced on average 30% more lines of code, 20% more commits and 23% more pull requests. Completed issues and epics did not rise in a statistically significant way. ## Where the gain went Time from opening to merging a pull request rose 49%, the share of pull requests with changes requested nearly doubled, comments per pull request rose 35% and the share of staff assigned to review rose 14%. By March 2026, 80% of firms used AI review tools, but only 23% of review comments came from AI. ## Key terms A commit records a code change, a pull request asks to merge changes into the main code, and code review is when another developer checks them for bugs and security issues before release. ## Why it matters Speeding up one step does not speed up the whole process; the bottleneck moves downstream. The same applies to any AI-accelerated work, from proposals to content: more output means little without capacity to check it and turn it into results. ## Earlier studies Earlier research mostly used controlled experiments that timed a single task with and without AI and often found 20% to 50% speedups. The Harvard paper instead uses years of everyday data from real companies, where review, testing and coordination absorb much of the lab speedup, explaining the gap between the two kinds of results. ## What it means for your business Find the next bottleneck before adopting an AI tool, measure finished outcomes rather than volume, do not cut review steps as the load grows, and when commissioning software, judge by delivery dates rather than coding speed. ## More articles [News · Models ### Meta opens its first frontier model, Muse Wonder, via API 12 Oct 2026 · 3 min](https://www.qanatone.com/en/haber/meta-muse-wonder-ilk-sinir-model/)[News · Models ### Microsoft on track to pay Anthropic $2B a year for Copilot 12 Oct 2026 · 3 min](https://www.qanatone.com/en/haber/microsoft-anthropic-copilot-odemesi/)[News · Infrastructure ### Microsoft's Maia 300 chip will run its AI services from 2027 12 Oct 2026 · 3 min](https://www.qanatone.com/en/haber/microsoft-maia-300-cikarim-cipi/)[News · Funding ### Amazon in talks to buy AI startup Decart for about $7 billion 11 Oct 2026 · 3 min](https://www.qanatone.com/en/haber/amazon-decart-7-milyar-dolar-gorusme/)[News · Funding ### Spend management startup Ramp valued at $60 billion 9 Oct 2026 · 3 min](https://www.qanatone.com/en/haber/ramp-60-milyar-dolar-degerleme/)[News · Policy ### White House unveils $6 billion science and AI package 8 Oct 2026 · 3 min](https://www.qanatone.com/en/haber/beyaz-saray-6-milyar-dolar-bilim-genesis/)[News · Search ### Google's September 2026 spam update has finished rolling out 8 Oct 2026 · 3 min](https://www.qanatone.com/en/haber/google-eylul-2026-spam-guncellemesi-tamamlandi/)[News · Infrastructure ### AMD lifts its Taiwan supply-chain investment 6 Oct 2026 · 3 min](https://www.qanatone.com/en/haber/amd-tayvan-kapasite-yatirimi/)