GitHub's new reviews: check the change before you accept it

Write the result you want to check

Before accepting an AI change to your app, write one sentence about what people should find easier. Then try that task in the changed app. A useful first exercise might be checking whether someone can tell that their request was received.

GitHub, where builders store and review their app's code, announced deeper Copilot reviews and new ways to fix blocked changes on October 28. The updates can support a review routine you understand, whether you are learning to build or commissioning a tool.

The central question stays practical: does this change do the job you agreed, including when something goes wrong?

Read the announcement: GitHub: Copilot code review Max and approvals.

Choose the review effort for the job

Copilot's new Max review setting is in public preview, meaning it is available to try and may change. It asks several AI models to review a proposed change independently, then combines and checks their findings. GitHub says this takes longer and uses more AI credits, its units for charging usage, than lighter reviews.

Paid Copilot plans have access to the new review features, subject to account controls. Ask the person managing your project which review setting is appropriate and what it will cost. The most detailed setting does not need to become the default for every edit.

Copilot can also approve proposed changes. This feature is now generally available but stays off until an administrator enables it. Administrators control whether its approvals count toward the required checks and which files qualify. A deeper review setting and permission to approve are separate choices.

Let fixes come back for a clear decision

Another update brings Agent Merge to GitHub's proposed-change pages. Paid Copilot users can select Watch and fix to address conflicting changes or failed automatic checks run by GitHub Actions. You can allow either kind of fix or both. Organization settings must permit its use.

This can help a change move forward while someone works on another task. It does not turn on automatic acceptance. GitHub treats that as a separate setting, and the project still needs to meet its review requirements.

If you try it, follow the proposed fix and check the app again afterwards. GitHub says the AI usage is charged to the person who enables Agent Merge for that change, so agree who owns that spending.

Practise a review you can explain

Use a practice project with made-up information. Ask the AI to explain what changed and why, then compare its answer with what you can see. If an explanation is unclear, that is a useful question to bring to a guided building session.

1. Choose one visible improvement and write the result a person should get.
2. Agree who reviews the change and which AI spending limit applies.
3. Try the normal task, missing information and a failed action.
4. Check any follow-up fix before the responsible person accepts the change.

Make understanding part of progress

Our view is that knowing how to check a change becomes more useful as AI takes on more of the building work. Keep a short note of what you tested and what still needs an answer. That gives your next practice session, or your conversation with a builder, a clear starting point.