Adoption and Impact of Command-Line AI Coding Agents: Microsoft's 2026 Claude Code Rollout Study

My take: Tens of thousands of Microsoft engineers, four months of real telemetry, and one result that is hard to ignore: developers who adopted Claude Code merged 24% more pull requests than they would have otherwise. This is one of the first large-scale academic studies measuring the real-world productivity impact of AI coding agents.

This is not a lab benchmark or a marketing claim. It is field evidence, and it should matter to any engineering team still evaluating whether AI coding agents are worth the investment.

One more data point worth noting: Claude Code generated 2.3 times more file edits per session than Copilot CLI, suggesting that agent depth matters as much as the model name. Access is not enough — the tool has to do the actual work.

Has your team started measuring the real impact of AI tools on development speed? If not, this study is a good starting point for knowing which metrics to track.