Microsoft study tracks command-line AI coding-agent adoption

A Microsoft arXiv study analyzes adoption and impact of command-line AI coding agents across tens of thousands of Microsoft engineers. The authors find that early adoption spread through social networks, retention was tied to actual coding activity, and adopters merged roughly 24% more pull requests than they otherwise would have. The authors also caution that merged PRs are a proxy, not a direct measure of value. This is a backlog correction for the papers lane because it directly addresses workplace AI productivity with large internal data.

Key details: The paper was submitted to arXiv on July 1, 2026, It studies tens of thousands of Microsoft engineers, The authors report adopters merged roughly 24% more pull requests than expected, Merged pull requests are treated as a proxy rather than a complete measure of value.

Why it matters: Large internal adoption data is more useful than anecdotal coding-agent claims, especially when the paper states the limits of its productivity proxy.