GitLab Automates Full Merge Request Lifecycle with Developer Flow

In GitLab 19.0, the Developer Flow AI agent is no longer limited to writing code — it now manages the entire merge request lifecycle: from handling reviewer feedback and resolving merge conflicts to the final merge. This is the next step after last year's launch of Developer Flow, which back then could only turn an issue into a ready-made MR.

What changed in GitLab 19.0

According to the GitLab Blog, until now AI only accelerated code writing, while all the work between opening a merge request and merging it remained manual: assigning reviewers, round after round of handling feedback, untangling conflicts, rebasing before the merge — every step required the developer's attention. The bottleneck simply shifted, and the tools didn't adapt to it.

Developer Flow was launched last year and solved only one task — turning an issue into a merge request, automating the routine preparatory work between setting the task and the MR appearing for review. But once the MR was created, further iterative work on it remained entirely manual.

The new version of Developer Flow closes exactly this gap: the same AI agent now handles reviewer feedback, resolves conflicts in long-lived branches, works through unfamiliar code, and splits bloated MRs into parts — paired with autonomous merge conflict resolution and one-click rebase-and-merge.

How the agent connects to work on an MR

Developer Flow can be launched at any stage of work on a task. There are three entry points: the Generate MR button right on the issue, assigning the Duo Developer service account directly to an issue or MR, or a new trigger via an @-mention in any discussion thread on an issue or merge request.

The key difference from the previous version is that the agent picks up the conversation and continues working on the same MR, rather than creating a new one that would then have to be reviewed manually from scratch. Combined with autonomous merge conflict resolution and the one-click rebase-and-merge feature, this reduces the amount of manual work across the entire stretch between opening an MR and its final merge.

A new category of AI tools for development

GitLab describes Developer Flow as part of a new, still-forming category of AI tools for code. The first wave of such tools accelerated writing the next line of code (autocomplete), the second gave developers a chat window for questions. What's emerging now is fundamentally different: agents that participate in the work throughout its entire span, rather than at a single fixed moment.

As GitLab puts it, developers can now stay "above the loop" — delegating tasks and checking the result while the agent handles the actual execution: reviews, conflicts, refactoring oversized MRs, and studying unfamiliar parts of the codebase.

What is Developer Flow in GitLab 19.0?

Developer Flow is GitLab's AI agent that, in version 19.0, expanded its area of responsibility from creating a merge request to its entire lifecycle: handling reviewer feedback, resolving merge conflicts on long-lived branches, studying unfamiliar code, and splitting overly large MRs into parts. It works together with autonomous conflict resolution and one-click rebase-and-merge.

How does the new version of

Developer Flow differ from last year's?

Last year's launch of Developer Flow automated only one stage — turning an issue into a merge request, that is, the routine preparation between setting the task and the MR appearing for review. Further work on an already-created MR remained entirely manual. In GitLab 19.0, the same agent continues working on the same MR even after its creation: it handles comments, resolves conflicts, and takes part in polishing it up to the merge.

How do you launch Developer Flow on a merge request?

There are three ways: click the Generate MR button on the issue, assign the Duo Developer service account directly to the issue or merge request, or mention the agent via an @-trigger in any discussion thread on the issue or MR. In all cases, the agent picks up the discussion context and continues working on the already-existing MR, rather than creating a new one.