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

Microsoft researchers published a study on the early 2026 rollout of Anthropic's Claude Code and GitHub's Copilot CLI across tens of thousands of engineers. They tracked adoption, retention, and impact, using merged pull requests as a proxy for output. The study found that first use spread primarily through social networks—engineers were more likely to try the tools if colleagues nearby had already adopted them. Retention correlated more strongly with engineers' baseline coding activity than with demographics like tenure or role. Most strikingly, adopters merged roughly 24% more pull requests than they would have without the tools, and the effect persisted across the four-month study window.

The findings have major implications for organizations deploying AI coding agents. At scale, token costs can run into millions of dollars annually, so understanding who will actually use the tools and whether they deliver measurable output is critical. The 24% lift in PRs suggests the tools are not just a novelty, but they also aren't uniformly adopted—engineers who code more already are more likely to stick with them. The researchers recommend treating visible peer use as a central rollout strategy, as social networks drive initial adoption more than top-down mandates. For engineering leaders, this is the first large-scale evidence that command-line AI coding agents can meaningfully increase throughput, but careful attention to adoption dynamics is needed to justify the investment.