Copilot code review now shows what each review spent. Decide review effort per repo before the bill decides it
Automatic AI review on every pull request felt free when nobody could see what it cost. GitHub now shows the credits a Copilot review consumed, lets admins set review effort through the API, and runs balanced reviews past three models. Somebody should choose those settings on purpose.

GitHub's November 6, 2026 update lets organizations manage Copilot code review settings, including default review effort, through the REST and GraphQL APIs, shows the AI credits each review session consumed, and lets the automatic-review rule skip actors on a bypass list. Pick one typical pull request, compare what the available effort levels cost and catch, then set defaults per repository, exempt bots that don't need review, and name one owner for the settings.
What GitHub changed on November 6
GitHub's changelog on November 6 bundled several Copilot code review changes that, together, turn AI review from a switch into something you can actually manage.
Settings now work through the REST and GraphQL APIs, from enterprise policy down to individual repositories. Enterprise admins can turn code review on or off for every organization, or leave it to each organization. Organization admins can manage whether it's enabled, the default review effort and the runner configuration. Repository admins can set review effort and decide whether reviews use custom instructions and MCP tools. Settings the enterprise manages can't be overridden below it.
The update also makes cost visible. Copilot code review agent sessions now show the AI credits consumed in a review. Balanced-effort reviews now use three reviewer models, with another model combining their findings. The rule that requests Copilot review automatically now honours the bypass list, so exempt actors don't get an automatic review. And the API now tells you which effort a review used, and which skills, MCP servers and custom instructions Copilot reported as sources for a comment.
It's available on Copilot Pro, Pro+, Max, Business and Enterprise.
Why this is a buying decision now
A review setting nobody chose is still a spending decision.
When AI review was a single toggle, the only question was whether to turn it on. Most teams did, everywhere, at whatever the default was. Now each review has a visible cost, effort is a dial, and the dial can be set differently in every repository. That's good. It also means that leaving everything on the default is a choice, made by nobody, that repeats on every pull request.
It matters more as coding agents open more of your pull requests. If an agent opens twenty small changes a day in a busy repo and each gets an automatic review at a high effort, you're paying one AI to check another, all day, on changes a person may not even read. And because failed or timed-out reviews are now retried once automatically, a flaky setup may run more reviews than its pull request count suggests, so look at the sessions, not just the count. None of that is a reason to switch review off. It's a reason to set it.
How the work moves
Choosing a default in an afternoon
- Pick
One recent, ordinary pull request from your busiest repository, with a known bug a human reviewer caught.
- Run
Request a Copilot review at each effort level available to you.
- Read
Note the AI credits each session shows, and whether it found the known bug.
- Choose
Set the cheapest effort that caught what mattered as the repo default.
- Write down
Record the choice, the date and the owner, so the next person doesn't redo it.
A review setting nobody chose is still a spending decision.
Compare effort levels on one ordinary pull request
You don't need a study. Take one recent, ordinary pull request from your busiest repository. Ideally pick one where a human reviewer caught a real problem, so you have something to check against.
Request a Copilot review at each effort level available to you. The organization and repository settings responses list the levels you can use. For each run, note two things: the AI credits the session shows, and whether the review found the problem the human found. Read the comments too. A review that finds the bug and adds fifteen nitpicks costs your team attention as well as credits.
Then choose the cheapest effort that caught what mattered, and make it that repository's default. Repeat for two or three repositories of different kinds, and you have a sensible starting policy in an afternoon. Our earlier advice to ask what the coding agent removed is a good extra test while you're there.
Set defaults by risk, and exempt the bots
Some repositories deserve a careful second reader on every change: anything touching payments, authentication, customer data or infrastructure. Others don't: docs, internal scripts, prototypes. Set effort accordingly, rather than one level for the whole organization.
Then use the bypass list. Dependency update bots, formatting jobs and release automation often open pull requests that don't need an AI review at all. Exempting them from automatic review is now supported. GitHub also restored the ability for Actions, Dependabot and your own apps to request reviews on bot-authored pull requests in some licensing setups, as long as your organization pays, so decide deliberately which bots should and shouldn't get one.
If you manage settings through the API, note one limit from the changelog: the enterprise and organization settings endpoints don't support GitHub Apps or fine-grained personal access tokens. The repository endpoints do, with Administration read or write permission. Plan your automation around that, and keep the token that changes organization settings with the person who owns them.
Decision guide
Which repos get more effort
| Consideration | Higher effort by default | Lower effort, or no automatic review |
|---|---|---|
| What breaks if it's wrong | Payments, auth, customer data, infrastructure | Docs, internal tools, prototypes |
| Who opens the PRs | People and coding agents writing new logic | Dependency bots and formatting jobs |
| Human review | Thin, or one reviewer for many changes | Already careful and well staffed |
| Change pace | Few, large, risky changes | Many small, routine ones |
Read the new overview with your reviewers
The same update changes what reviewers see. Copilot's overview comment is more compact: findings stay at the top, the effort used and any surveys move into a footer, and skipped or content-excluded files share one "Files not reviewed" section with the reasons linked. On later reviews of the same pull request, findings unrelated to your recent changes are collapsed and then hidden.
That's a sensible design, and it has one consequence worth saying out loud to your team. The "Files not reviewed" list is now the place to check what the AI didn't look at. If a reviewer assumes Copilot covered the migration file and it was excluded, the compact overview won't shout about it. Make reading that section part of how your team approves.
There's also a feedback loop. A thumbs-down on an overview now lets you pick a reason, such as an approval assessment you disagree with or findings that are incorrect or nitpicky. If your team finds the reviews noisy at a given effort, using that is cheaper than living with it.
Name an owner and look again monthly
Put one person in charge of review settings across the organization. Their job is small: keep the per-repo defaults written down, check the bypass list when new bots are added, and look at credit use once a month. If a repo's review spend jumps, it's usually because a new agent or bot started opening pull requests there.
Credits change how teams treat a tool. Our Notion AI credits review made the same point about deciding who pays for each agent before the invoice arrives. And keep human review where it counts. Copilot's review is a second reader, not a replacement for the person who understands why the change exists.
If you'd like help setting up review policy, agent-friendly pull request workflows or the automation around them, that's part of our web application engineering work. Talk to TWIMCO.
Related work: Sutton|Place
We build alongside the people doing the work.
Bring us the problem you’re thinking about.