✓ ShippedByAgents
Independent Research / benchmark

CMU and Stanford study of a company's "2x" AI mandate: merged PRs per engineer doubled, and automated review overtook human review

Updated on Set published to 2026-07-02 from the arXiv submission date on abs/2607.01904.

Researchers followed a mid-sized, AI-forward company that set out in mid-2025 to double merged pull requests per engineer. Across 802 developers and 196,212 pull requests from January 2024 to April 2026, per-person throughput reached 2.09 times the pre-mandate baseline by April 2026. Their analysis ties the gain to AI adoption and to how much people used the tools, with the mandate acting as a catalyst. Review load per reviewer roughly doubled and automated review overtook human review, while merge and revert rates held steady.

The numbers (as reported)

Sample802 developers, 196,212 pull requests (Jan 2024 to Apr 2026)
Throughput2.09x pre-mandate merged PRs per engineer by April 2026
Review loadPer-reviewer load roughly doubled
Quality signalsMerge and revert rates held steady

Facts

Company / subject
Carnegie Mellon University and Stanford University
Tools
Unstated
Codebase
Not applicable
Work type
Research / benchmark
Stack
—
Published
2026-07-02
Last checked
2026-11-16
Updated
2026-11-16

What the source claims Independent

Reported by an independent source, such as a study or the press.

  • Merged PRs per engineer reached about 2x the pre-mandate baseline, but only ten months after the mandate 2.09x
    “Per-capita PR throughput approached 2× only in April 2026, ten months after the announcement”
  • PR throughput rose about 12% when a developer started using AI, then kept growing with use +12%
    “PR throughput rises by about 12% when a developer starts using AI and then keeps growing with accumulated use”

Caveats

One anonymous company. Adoption was not randomly assigned, so the authors avoid exact causal attribution. The abstract does not name the AI tools.

More stories by tool: Unstated

Related topics: Research and benchmarksIndependent sources

Source

arXiv 2607.01904

One email a week: what actually shipped

New sourced stories, the numbers behind them, and what didn't work. No hype.

We'll only use this to send the newsletter. See the privacy page.