BRK278 Breakout Tue 17 Nov 16:30 PST
Inside Copilot code review: From context to approval
David Apirian, Ria Gopu
Summary
From the recording
GitHub's Copilot code review is agentic and reads repository context beyond the diff. New in public preview are cross-repo context, the Dreaming feature (learning from past reviewer comments), secret scanning and an adaptive tier that picks light, balanced or max effort per pull request. Speakers gave average costs of about 25 cents (light), $1.50 (balanced) and $10 (max) per review. Copilot code review is GA in Azure Repos. Real approvals are off by default and can be enabled per enterprise, organisation or repository, including by file path. Caveat: costs are speaker averages, and model mixes change every one to two weeks. Cross-repo context currently reads only public repositories by default; per-repo configuration is coming soon.
Session description
Microsoft's catalogue description
Merge your code with confidence after Copilot reviews it. Learn how Copilot pulls context from the right places to conduct informed reviews, and how it can be tailored to your team's coding standards. We'll show how we built 3 review effort levels, each backed by an ensemble of agents and tuned to different cost and quality preferences. Finally, learn how we built confidence in Copilot's ability to approve pull requests. You'll leave knowing how to make Copilot code review work for you.