September 7, 2026: The VS Code Agent Merge preview arrived in Visual Studio Code 1.136, released September 2. It gives an agent a repeated task: address pull-request review feedback, repair failing checks and merge conflicts, and rerun workflows until the PR is ready to merge. Microsoft documents the feature in its 1.136 release notes.
What changed
GitHub’s September 4 Copilot weekly update also highlights Agent Merge as a public preview. The focus is the follow-through after a PR exists: an agent can keep working through feedback and CI failures as part of preparing it for merge.
To try the feature, enable chat.agentMerge.enabled. The release notes say it can currently be enabled for a session only from the Agents window, using Enable Agent Merge for Active Session or the title-bar Agent Merge button. Its preview status is relevant when planning a team rollout.
Why this matters for QA engineers
QATechTools analysis: Repeated CI repair could reduce the manual handoffs between a failed build, a patch and another test run. The useful evaluation is whether the agent fixes the underlying defect while preserving the test’s intent. A green check alone cannot establish that.
For an initial pilot, use a small PR with a known regression and a failing assertion. Review whether the final patch corrects the behavior, retains meaningful assertions and passes the relevant regression suite. Separately, try a merge conflict involving test fixtures and check that the resolution preserves both branches’ intended coverage.
Record the original failure, each repair attempt and the final test evidence. Watch for changes that skip tests, weaken assertions or broaden retries without justification. These are suggested QA evaluation criteria, not outcomes claimed by Microsoft or GitHub.
The immediate opportunity is to assess how reliably an agent can carry a test failure through to a reviewable fix. Teams should judge the preview by the resulting code and evidence, alongside their existing review and merge requirements.