Microsoft's Study Proves Claude Code Boosts PR Output by 24%

Key Takeaways

• Claude Code users merge 24% more PRs per Microsoft's study.
• Drive adoption via peer visibility, not mandates.
• Retention correlates with coding activity, not demographics.

What Changed — The First Large-Scale Study of Claude Code in the Enterprise

A peer-reviewed study published on arXiv (July 1, 2026) by Microsoft researchers Emerson Murphy-Hill, Jenna Butler, and Alexandra Savelieva analyzed the rollout of Claude Code and GitHub Copilot CLI across tens of thousands of Microsoft engineers. This is the first rigorous, large-scale empirical study of command-line AI coding agents in a real enterprise setting.

The headline finding: adopters merged roughly 24% more pull requests than they would have otherwise. The lift persisted across the four-month study window, suggesting the productivity gain is not a novelty effect.

But the paper's deeper insights are about who adopts these tools, why they stick, and what organizations get wrong when rolling them out.

What It Means For You — Concrete Impact on Daily Claude Code Usage

1. Your coworkers matter more than your title

The study found that first use spread primarily through social networks. An engineer with one coworker who used Copilot CLI was significantly more likely to try it themselves. The effect compounded with more peer exposure.

For you: If you're the first on your team to use Claude Code, you're not just helping yourself — you're the adoption catalyst. Show your teammates what you're doing. Run claude in shared terminals during pair programming. Share your CLAUDE.md config. The study proves visibility drives adoption more than any mandate.

2. Retention is about activity, not demographics

Engineers who were already active — measured by baseline pull-request creation — were more likely to continue using the tools. Demographics (tenure, role, team size) had minimal predictive power.

For you: Claude Code sticks when you use it for real work, not just experiments. Set up your CLAUDE.md to match your project's conventions so the tool produces mergable code from day one. The faster you ship your first PR with Claude Code, the more likely you'll keep using it.

3. Prior IDE Copilot users adopted faster

Engineers who already used GitHub Copilot in their IDE were more likely to try and retain the CLI agents. Familiarity with AI-assisted coding lowered the barrier.

For you: If you're already using Copilot or Claude in your IDE, the CLI version is a natural extension. Start with claude code on a small refactor task you'd normally do manually. The study suggests this "graduation path" works.

Try It Now — Actionable Steps Based on the Research

For individual developers:

1. Set up your CLAUDE.md today — The study shows retention correlates with active use. Make your first session productive by defining your project's conventions upfront.

# CLAUDE.md
## Code Style
- Use TypeScript strict mode
- Prefer named exports over default exports
- All functions must have JSDoc comments

![Figure 2: Change in odds of initial use \tikzbaseline=-0.6exinitialblue (0,0) circle (0.7ex); and retention \tikzba](https://arxiv.org/html/2607.01418v1/x2.png)


## Testing
- Run `npm test` before every PR
- Aim for 80%+ branch coverage

![Figure 1: Change in odds of initial use \tikzbaseline=-0.6exinitialblue (0,0) circle (0.7ex); and
retention \tikzba](https://arxiv.org/html/2607.01418v1/x1.png)


## Commands
- Build: `npm run build`
- Lint: `npm run lint`

1. Run your next PR through Claude Code — The 24% lift came from real PRs. Pick a task you'd normally do and try:

claude code "Implement user authentication for the /login endpoint. Follow our CLAUDE.md conventions."

1. Share your experience visibly — Post in your team's Slack channel or show during standup. The study proves peer exposure is the #1 adoption driver.

For engineering leaders:

1. Don't mandate — demonstrate — Top-down rollout "orders" underperform peer-driven adoption. Instead, identify your early adopters and give them visibility.

2. Track PR velocity, not token spend — The study uses merged PRs as a proxy for output. Monitor this metric across your team before and after rollout. The 24% lift is a benchmark to aim for.

3. Budget for social adoption — Since peer visibility matters, invest in demos, hackathons, and pair programming sessions. Token costs can run "millions of dollars annually" at scale, so getting adoption right saves money.

The Bottom Line

The Microsoft study validates what Claude Code users have anecdotally reported: you ship more, faster. But it also reveals a crucial insight: the tool is only as effective as the social and workflow context around it. Set up your environment, show your peers, and start with real work — not toy examples.

Paper reference: Murphy-Hill, E., Butler, J., & Savelieva, A. (2026). Adoption and Impact of Command-Line AI Coding Agents: A Study of Microsoft's Early 2026 Rollout of Claude Code and GitHub Copilot CLI. arXiv:2607.01418.

Source: arxiv.org