Lawrence Jenger
March 5, 2026 20:52
GitHub’s AI code reviewers currently handle 1 in 5 pull requests worldwide. The new agent architecture increased developer satisfaction by 8.1%.
GitHub’s AI-powered code review tools have over 60 million reviews and usage has increased 10x since its launch in April 2025. This feature now handles more than 1 in 5 code reviews on the platform. This is a major change in the way enterprise software is shipped.
The milestone comes as parent company Microsoft trades with a market capitalization of nearly $3 trillion and AI tools are central to its developer ecosystem strategy.
What has changed under the hood?
GitHub has restructured Copilot’s code reviews based on what it calls an “agent architecture.” This essentially gave the AI the ability to explore memory and repositories rather than reviewing code individually. The company says the move has resulted in an 8.1% increase in positive feedback from developers.
Actual difference: The system now catches issues as they are read, instead of waiting for the review to finish. Maintain context across multiple pull requests. It also reads linked issues and PRs and flags gaps that look OK in code but don’t match the project requirements.
“Copilot Code Review handles the review and summarization of pull requests, freeing up teams to focus on more complex tasks,” said Suvarna Rane, software development manager at General Motors.
Quality vs. speed trade-off
GitHub made a deliberate choice. A slow review that brings real issues to the surface is better than immediate feedback that adds noise. Recent model upgrades have increased positive feedback rates by 6% and review wait times by 16%. The team believes it is an acceptable trade.
The numbers speak more about restraint than quantity. Copilot remains silent on 29% of reviews where nothing actionable was found. When commenting, there are an average of 5.1 suggestions per review, organized into logical groupings rather than spread out across the pull request timeline.
Enterprise adoption patterns
Today, over 12,000 organizations automatically run Copilot code reviews on every pull request. Financial services company WEX provides a case study where two-thirds of its developers use Copilot, including its most active contributors. After making AI-assisted reviews the repository-wide default, WEX reported an approximately 30% increase in code shipments.
This feature is integrated with deterministic tools such as CodeQL and ESLint for security and quality checks. This means that AI suggestions are layered on top of existing automated safety measures, rather than replacing them. Importantly, Copilot reviews do not count toward the human approval required for the merge, maintaining a human oversight loop.
what’s next
GitHub outlined two development priorities: deeper personalization to learn team-specific settings and two-way conversations that allow developers to adjust fixes before merging. The goal is to move from being a one-shot reviewer to being more like an interactive pair programmer.
Copilot Code Review requires a paid subscription at the Pro, Pro+, Business, or Enterprise level. Organizations already using this feature can enable automatic review of all pull requests through their repository or organization settings.
Image source: Shutterstock
