GitLab pitches $0.25 AI code review that runs on every merge request

Review queues as the new bottleneck

GitLab argues that AI agents have changed how much code a team can produce in a day, while review queues have not kept pace. Delivery stalls as changes wait for a closer look, and the usual fixes, such as reminders, reviewer rotations or assigning a whole team, buy little time before they burn people out, the company says.

The company also addresses client-side AI review, which helps authors catch issues while code is being written, before a merge request opens. GitLab says this only goes so far. Comments on a diff do not decide whether a change can merge or reach production, and every merge request still needs a consistent certification step tied to pipelines, security findings, code owners, approval rules and audits. In GitLab's view, review in the author's environment cannot standardize that work across projects or show leaders what an agent did before a change merges.

Its answer is to run automated review in the platform that already tracks each change from plan to production. Earlier this year, GitLab made Code Review Flow generally available on GitLab Duo Agent Platform.

How Code Review Flow works

Code Review Flow is native to GitLab merge requests. It runs as a CI/CD job on the customer's runners, whether GitLab-hosted or self-managed, so review executes in the same environment as the rest of delivery. GitLab says it works the same way on GitLab.com, GitLab Dedicated and GitLab Self-Managed. Air-gapped or regulated environments can keep inference under their control with self-hosted models, and teams can choose between GitLab-managed models and their own as the foundation for review.

The process follows the merge request:

• A merge request opens or becomes ready. Automatic review can start the flow, or an author can assign GitLab Duo.
• The flow analyzes the change with repository and merge request context, guided by custom review instructions and the models the group selected, and comments on the diff.
• Authors and their coding agents push fixes. A follow-up review verifies Duo's earlier findings and carries context forward.
• When a human must approve, Recommend Reviewers Flow proposes who should take the request, and the merge request is assigned from there.

GitLab highlights several capabilities. Automatic reviews can be enabled at the project, group or instance level, and settings cascade. Custom review instructions capture language conventions, security practices and path-specific rules, and a group can point the flow at a template project so standards travel with the work. Code owners and approval rules still apply, and Recommend Reviewers Flow suggests the smallest group that can satisfy them, based on expertise, workload, availability and timezone.

Roadmap and getting started

GitLab lists several planned additions for agentic code review:

• An extended analysis pass that pulls in calling code and tests for high-complexity merge requests, under the customer's model-selection controls.
• The ability for GitLab Duo to approve or request changes under settings the customer controls.
• Risk assessment on the merge request, to show which changes deserve more attention.
• Local review in Duo CLI, letting developers run project review guidance against working-tree changes before the merge request runs the organization's official check.

New users can get access to GitLab Duo Agent Platform through a free trial. GitLab Premium and Ultimate customers can turn on Duo Agent Platform and use the GitLab Credits included with their subscription. The post carries a forward-looking statements disclaimer, so roadmap items are subject to change.

What it means for teams

GitLab's pitch is that a predictable unit cost of $0.25 per review is low enough to leave automatic review on for every merge request. For teams already on GitLab, that means review, pipelines, security findings and approvals sit in one place, and the company says a "reviewed" change becomes one certified on the path to production. The 100x ROI figure is GitLab's own estimate, and the source gives no details on how it was calculated.