- The study examines the adoption, retention, and impact of command-line AI coding agents like Claude Code and Copilot CLI at Microsoft in early 2026.
- Findings indicate that initial adoption spread mainly through social networks among engineers.
- Retention was linked more to engineers' coding activity levels rather than demographic factors.
- Adopters merged approximately 24% more pull requests compared to their baseline, using merged PRs as a proxy for output.
- The productivity lift persisted over a four-month observation period.
- The results suggest CLI coding agents are not uniformly adopted or mere novelties, and peer influence is crucial for rollout strategies.
- Organizational token costs can be high, making accurate assessment of adoption and impact essential to avoid wasted investment.