Adoption of coding agents at Microsoft spread through peers

Adoption and Impact of Command-Line AI Coding Agents: A Study of Microsoft’s Early 2026 Rollout of Claude Code and GitHub Copilot CLI

Organizations rolling out agentic command line tools like Anthropic’s Claude Code and GitHub’s Copilot CLI need to know who will try them, who will keep using them, and whether the tools produce enough output to justify their cost. At organizational scale, token spend can run into millions of dollars annually, so misreading adoption, retention, or impact can make a rollout expensive without changing engineering velocity. Studying tens of thousands of engineers at Microsoft over its early-2026 rollout, we find that first use spread primarily through social networks, retention was associated more with engineers’ coding activity than with demographics, and adopters merged roughly 24% more pull requests than they would have otherwise. We use merged pull requests as our proxy for output — acknowledging that a merged PR is not the same as the value it delivers — and the lift persists across our four-month window. These results suggest that CLI coding agents are neither uniformly adopted nor mere novelty effects and that organizations should treat visible peer use as central to rollout strategy.

arxiv.org

Three Microsoft researchers studied the company's early 2026 rollout of command-line coding agents, Claude Code and the Copilot command-line interface (CLI), across tens of thousands of engineers from January through April. It is the largest enterprise rollout study I have read, and two findings stand out.

Adoption was social. The strongest predictor of an engineer trying the tool was whether the people around them already had. When more than a quarter of an engineer's skip-level peers were using it, the odds they would try it were more than three times higher. A manager who used it raised the odds by 82%. The authors' recommendation is blunt: treat visible peer use as central to the rollout, because that is what actually spreads it.

Adopters merged about 24% more pull requests (PRs) than a synthetic control predicted, and the gain held across the four months rather than fading after the novelty. It scaled with use: about 15% more for people using the tool three days a week, around 50% for five or more. The authors are careful to say that a merged PR is not the same as the value it delivers, and I would underline that.

What I take from this as a manager is the social finding. A tool rollout is not a license purchase and an announcement. It is a question of who the team watches, and whether those people are visibly using the thing, and whether what they are doing with it is good.

When a new tool arrived on your team, who did people actually learn it from?

Photo source: https://photos.robertstowe.com/san-diego