Microsoft study: AI coding agents boost pull requests by 24%

A large-scale Microsoft study published July 1, 2026 found that developers who used command-line AI coding agents—Anthropic’s Claude Code and GitHub Copilot CLI—merged roughly 24% more pull requests per engineer per day than they otherwise would have over a four-month period. The researchers reported a likely range of +14.5% to +33.7%, and the gain did not fade during the study window. A robustness test (shifting the rollout start date) showed no similar jump, confirming the boost was tied to real tool use. Critically, the lift was not automatic: engineers who used the tools five or more days per week saw a greater than 50% increase, while those using them three days per week saw only around 15%. This suggests that actual usage frequency, not just license adoption, drives results.

Within Microsoft’s environment, Copilot CLI users experienced about 2.2 times the pull-request lift of Claude Code users during comparable weeks, but the researchers cautioned against treating this as a general ranking. A separate July 2 enterprise study tracked 802 developers and 196,212 pull requests at a mid-sized company where a CTO mandated a “2x” throughput increase using merged PRs as the metric. By April 2026, per-developer throughput had reached 2.09 times the baseline (from 21.2 to 44.3 merged PRs per month). However, gains were concentrated in newer repositories; legacy codebases saw little improvement, and seniority did not explain the gap. This underscores that AI coding agents work best in clean, well-documented codebases, not complex legacy environments with heavy dependencies and review overhead.