Here is what each change does and what to check before turning it on.
Adaptive: Copilot picks the review depth
Until now, someone had to choose an effort level for each pull request, either in settings or at request time. GitHub's example is that a documentation tweak may only need Lite, while a change to authentication logic benefits from Max.
Adaptive hands that choice to Copilot. It weighs each pull request's complexity and risk, then reviews at Lite, Balanced, or Max. Each review shows which effort level Adaptive picked.
GitHub's documentation describes the three levels this way:
- Lite: a cost-efficient review aimed at glaring bugs, security vulnerabilities, and style inconsistencies.
- Balanced: deeper analysis of complex logic, security-sensitive code, and cross-service changes, using a higher-reasoning model.
- Max: the most intensive review, using multiple high-reasoning models.
How to configure it
- Select Adaptive for a single pull request from the Reviewers bar.
- Or set it as your default effort in personal, repository, organization, or enterprise settings.
- Balanced remains the default. Setting Adaptive as the default is an explicit choice.
The admin cap and the cost question
Adaptive never picks a level above the maximum set for your enterprise, organization, or repository. An administrator who capped reviews at Balanced will not see Adaptive quietly escalate to Max.
Cost is the real consideration. GitHub's docs say higher effort consumes more AI credits. As a rough average, a review costs about $0.25 on Lite, $1.50 on Balanced, and $10 on Max. Actual costs vary with factors such as pull request size.
Adaptive could save money if most of your pull requests are routine. It could also raise spending if your risk-heavy changes now get Max reviews instead of Balanced. The sensible move is to watch the effort levels Adaptive chooses over a few weeks before making it a default across an organization. GitHub has not published how Adaptive scores complexity and risk, so treat its judgment as something to audit.
Cross-repository context
Changes often break things outside the repository you are editing. With this preview, Copilot code review can read related public repositories during a review. It can find code elsewhere that depends on your change. It can also pull in documentation, custom instructions, and skills stored in other repositories.
The scope is narrow. The announcement says this is in public preview for public repositories. Don't assume it reads your private repositories.
GitHub's documentation, as summarized in the research notes for this story, adds two details:
- The search covers the default branch of related public repositories.
- Repository administrators can turn off the "Search public repositories" setting, which is on by default.
Organizations and enterprises that provide Copilot must also enable the policy for opt-in to preview features.
Secret detection in review comments
Copilot code review can now flag exposed credentials in pull request review comments, so you can remove them before merging. It uses two approaches:
- Known formats: GitHub collaborates with 150+ technical partners through GitHub's secret scanning partnership program to build high-confidence detectors. These cover 500+ secret types and use deterministic methods such as regular expressions.
- AI/ML-detected secrets: this covers credentials with no recognizable format. It uses LLMs and a GitHub purpose-built model for secrets.
To turn it on, go to your repository or organization settings and select Copilot, then Code review.
The announcement does not say this replaces GitHub secret scanning, push protection, or a proper secrets manager. It also makes no claim of complete coverage. The AI-detected category is probabilistic, so expect some noise and some misses. Think of it as an extra net at review time. If a credential has already been committed, it still needs to be rotated, not just deleted from the diff.
Review outcomes in Repository insights
You can now see how Copilot code review feedback is addressed and used in each repository. Open the Insights tab of any repository you have read access to, then click Repository insights. The announcement does not list the specific metrics, so check the linked "Track Copilot code review outcomes by repository" guide before building reports around it.
Azure Repos: now generally available
Copilot code review for Azure Repos has left public preview. The announcement says it is now generally available. This is the part most relevant to Microsoft-centric shops that haven't moved to GitHub.
The history so far:
- Microsoft's Azure DevOps blog first announced a limited public preview.
- Sprint 279 notes opened the preview to all Azure DevOps customers, with no early-access signup.
- A later update added Lite and Balanced effort, bulk comment resolution, and audit events.
What the GA announcement lists:
- Enablement: administrators turn it on at organization, project, and repository level.
- Automatic reviews: a branch policy requests a review on pull requests that target a branch. Manual requests come from the Reviewers list.
- Effort defaults: Lite or Balanced for a project or repository, with a different level selectable per review.
- Customization: custom instructions and agent skills.
- Follow-up: apply and commit suggestions while resolving the related comments.
- Auditing: audit log events record when the feature is turned on or off and when its settings change.
- Billing: usage is billed in GitHub AI credits to the Azure subscription linked to the Azure DevOps organization, with Azure Cost Management available for tracking and budget alerts.
Note that Adaptive is not listed among the Azure Repos effort defaults. The GA announcement mentions only Lite or Balanced there.
Operational details to verify
Microsoft Learn's Azure Repos page describes some requirements. The page I reviewed still carried a preview-era notice and a "last updated" date of October 19, 2026, so confirm against your own tenant:
- Reviews run through an Azure Pipelines agent pool, by default your organization's default pool. Self-hosted pools and Managed DevOps Pools were not supported at the time of that page.
- Git repositories only. TFVC isn't supported.
- Copilot always leaves a Comment review. It never approves or requests changes, so it doesn't satisfy required-reviewer policies or block merging.
- Listed limits include 100 or fewer changed files and a 10 GB repository size cap. Concurrency limits are 5 reviews per organization and 2 per user.
- One credit equals $0.01. InfoQ noted that charges appear under the product GitHub Copilot for AzDO, 48 hours after a review completes.
- Microsoft's guidance is to enable it for one or two repositories first, compare effort levels, and watch daily usage.
You can create an Azure budget alert filtered to that product. Budgets only notify you. They don't stop reviews.
What to do this week
- Check your caps. Find the maximum review effort set at enterprise, organization, and repository level. Adaptive works inside that ceiling.
- Pilot Adaptive on a few repositories. Record which levels it picks and compare against your spend.
- Decide on cross-repository search. Repository administrators who don't want it can switch off the public repository search setting.
- Turn on secret detection in a test repository. Judge the false-positive rate before rolling it out widely. Keep secret scanning and push protection in place.
- Azure DevOps teams: set up budget alerts on the GitHub Copilot for AzDO product before enabling branch policies broadly. Automatic reviews on every pull request add up.
Bottom line
None of this makes AI review a replacement for human reviewers, and GitHub doesn't claim it does. Adaptive is the most interesting change because it trades manual control for automation, within an administrator-set limit. Its value depends on how well Copilot judges risk, which you can only learn by watching its choices. The Azure Repos GA matters most for organizations still on Azure DevOps, who now have a supported, billed, audited review option.
References
- Copilot code review: Adaptive effort, secret detection, and more GitHub Changelog · 2026-11-17T18:02:05+00:00
- Copilot Code Review Reaches Azure Repos, Billed Per Review with Reporting Two Days behind - InfoQ infoq.com
- Using GitHub Copilot code review - GitHub Docs docs.github.com