GitHub Copilot code review now runs your builds and tests
The reviewer can now execute code and hit APIs, auto-resolves its own threads, and turns Lite into an agent ensemble that gets 47% more high-severity fixes.
Copy markdown
The review agent can run your build now
Copilot code review can now execute builds, run tests, and call APIs mid-review through the Copilot SDK toolkit, instead of reasoning from the diff alone. Findings it can prove by running your code means far fewer hallucinated nitpicks.
Lite reviews are now an agent ensemble
The cheap Lite effort level runs several agents together rather than one, lifting addressed comments 47% for high-severity findings, 31% for medium and 11% for low, while cutting review cost about 8%. The budget tier just got better for free.
Comments that resolve themselves
Copilot now auto-closes its own review threads once a later commit addresses the feedback, and writes a contextual commit message when you apply one of its suggested fixes. Less manual triage on every PR.
Elsewhere: Washington weighs a legal duty of care for AI
US Senate negotiators are reportedly drafting a bipartisan bill that would legally require frontier labs to mitigate known catastrophic risks and let regulators hold up 'unsafe' releases. If it lands, the frontier model drops you depend on could get slower and more gated.