GitHub co-pilot medium depth reviews (public preview)

GitHub co-pilot code review now offers a "medium" depth review mode in public preview that routes pull requests to a higher-reasoning model for longer analysis. The speaker says medium depth is intended for the repositories and services you "rely on the most" and lists three concrete target scenarios: complex logic, "securities and step code", and "cross service changes". To enable it per repository: open Settings → select co-pilot → Code review → select medium. Organization admins can set a default review level for all repos that haven't chosen their own.

Medium reviews use a "latest frontier model" and generally consume more AI credits than low, enabling more in-depth analysis. The speaker gives a tangible example: an accounting bug caused by integer-math order of operations—multiplying before dividing was required to avoid the budget showing 0% used until it hit 100%. As rollout evidence, the speaker reports improved common-address rates and a "resolution rate of over 70% of findings." The speaker emphasizes that medium is best for high-value, high-risk code where longer, cross-file or cross-service reasoning is required.

Actionable points retained verbatim: enable via Settings → co-pilot → code review → select medium; note that "medium generally consumes more AI credits than low"; and that medium depth reviews and GitHub co-pilot code review are "available in public preview now." The feature is positioned as a trade-off: higher cost in AI credits for deeper, higher-recall analysis and better catching of subtle issues in critical repositories.