GitHub Copilot Code Evaluation Hits 60M Evaluations as AI Handles 20% of Pull Requests

GitHub’s AI-powered code evaluation device has crossed 60 million evaluations, with utilization rising 10X since its April 2025 launch. The function now handles multiple in 5 code evaluations on the platform—a major shift in how enterprise software program will get shipped.

The milestone comes as dad or mum firm Microsoft trades close to a $3 trillion market cap, with AI tooling more and more central to its developer ecosystem technique.

What Modified Beneath the Hood

GitHub rebuilt Copilot code evaluation on what it calls an “agentic structure”—basically giving the AI reminiscence and the flexibility to discover repositories quite than reviewing code in isolation. The shift produced an 8.1% improve in constructive developer suggestions, in line with the corporate.

The sensible variations: The system now catches points because it reads quite than ready till the tip of a evaluation. It maintains context throughout a number of pull requests. And it reads linked points and PRs to flag gaps the place code seems wonderful alone however would not match undertaking necessities.

“Copilot code evaluation handles pull request evaluations and summaries, permitting groups to deal with extra advanced duties,” mentioned Suvarna Rane, Software program Growth Supervisor at Normal Motors.

The High quality vs. Velocity Tradeoff

GitHub made a deliberate alternative: slower evaluations that floor actual points beat immediate suggestions that provides noise. One current mannequin improve improved constructive suggestions charges by 6% whereas rising evaluation latency by 16%. The group considers that a suitable commerce.

The numbers inform the story of restraint over quantity. Copilot stays silent in 29% of evaluations the place it finds nothing actionable. When it does remark, it averages 5.1 ideas per evaluation—clustered into logical teams quite than scattered throughout the pull request timeline.

Enterprise Adoption Patterns

Over 12,000 organizations now run Copilot code evaluation mechanically on each pull request. Monetary providers firm WEX gives a case examine: two-thirds of its builders use Copilot, together with essentially the most lively contributors. After making AI-assisted evaluations the default throughout repositories, WEX reviews delivery roughly 30% extra code.

The function integrates with deterministic instruments like CodeQL and ESLint for safety and high quality checks—which means AI ideas layer on high of present automated safeguards quite than changing them. Importantly, Copilot’s evaluations do not rely towards required human approvals for merging, preserving the human oversight loop.

What’s Subsequent

GitHub signaled two improvement priorities: deeper personalization to be taught team-specific preferences, and two-way conversations that allow builders refine fixes earlier than merging. The objective is transferring from a one-shot reviewer to one thing nearer to an interactive pair programmer.

Copilot code evaluation requires a paid subscription—Professional, Professional+, Enterprise, or Enterprise tier. Organizations already utilizing the function can allow computerized evaluations on each pull request by way of repository or group settings.

Picture supply: Shutterstock