Writing the code was never the hard part of shipping a pull request. The hard part is everything after: waiting on review comments, rerunning a flaky CI check, rebasing when the base branch moves. GitHub Copilot's new Agent Merge feature, shipped in VS Code 1.136 as a public preview, is built specifically to close that gap.
Key Takeaways
- GitHub Copilot's new Agent Merge, shipped in VS Code 1.136 as a public preview, automatically clears everything blocking a pull request: unresolved review comments, failed CI checks, and merge conflicts.
- It runs as a repeating supervision loop, not a one time pass, rerunning workflows and re-addressing new comments until the pull request is genuinely ready to merge.
- Agent Merge is opt-in per session via the chat.agentMerge.enabled setting, not a blanket policy, so it only touches pull requests a developer explicitly enables it on.
- For product managers, this signals agents shifting from writing code to owning the full PR lifecycle, which pushes the human role toward reviewing outcomes rather than checking every line.
Learn this hands-on
Go from idea to a live product with real users. Build and ship your first SaaS with Claude Code in a live cohort, with auth, payments and deployment done by the end. Join the Ship Your First SaaS with Claude Code cohort.
What Agent Merge Actually Does
Agent Merge is not a new way to write code. It is a supervision loop that runs after code has already been written and a pull request is open. Once you enable it on an active agent session, it keeps monitoring that pull request for anything blocking it from merging, and works through each blocker automatically.
Concretely, it does three things:
- Addresses review feedback. It responds to unresolved review threads and changes-requested reviews, including comments left by human maintainers and by Copilot's own automated pull request reviewer.
- Fixes failed checks. It looks at failing required CI checks and pushes fixes to get them passing again.
- Resolves merge conflicts. When a branch falls behind its base branch, Agent Merge updates it and resolves any resulting conflicts.
The important part is that this is not a single pass. GitHub Copilot repeats the cycle, rerunning workflows as needed, until the pull request clears every blocker and is genuinely ready to merge. If a fix triggers a new review comment or a new failing check, the agent goes back in and handles that too.
How to Turn It On
Agent Merge ships behind a setting, chat.agentMerge.enabled, and GitHub is explicit that this is a preview: behavior may still change before general availability. Once the setting is on, you activate it per session, not globally for every pull request you open. You can start it from the Agents window by running "Enable Agent Merge for Active Session," or by clicking the dedicated Agent Merge button that appears in the editor's title bar for an active session.
That per-session activation matters. It means Agent Merge is something you opt a specific pull request into, rather than a blanket policy that lets an agent touch every open PR in a repository unsupervised.
Part of a Bigger Shift in the PR Lifecycle
Agent Merge did not land alone. It shipped alongside other VS Code 1.136 changes that push agent sessions deeper into everyday development work, including experimental support for running Copilot and Claude agent sessions across every folder in a multi-root workspace, and a reorganized Agents window that groups related chat sessions into a hierarchy and flags which ones need attention. This mirrors how Claude Code structures agents and subagents to hand off specialized work within a single session.
Taken together, GitHub frames this as part of a broader shift: the agent session, not the single prompt-and-response exchange, is becoming the core unit of engineering work inside the editor. Writing the diff was step one. Getting it merged, without a human babysitting every review comment and flaky check, is the part GitHub is now trying to automate.
This is also where the comparison to other agent-driven dev tools gets interesting. If you're weighing how Copilot's approach to agent sessions holds up against Claude Code's parallel agentic workflows, Agent Merge is a useful data point: it shows GitHub optimizing for autonomy across the full PR lifecycle, not just code generation. And if the review-feedback loop is what interests you most, it's worth seeing how that compares to a dedicated review agent in our breakdown of Claude Code Review vs Cursor Bugbot. For a broader look at how GitHub's agent approach stacks up against another CLI-first competitor, see our Claude Code vs Codex CLI comparison.
What This Means If You're Building With AI Agents
For product managers and builders using AI coding agents day to day, Agent Merge is a signal worth paying attention to even if you never enable the setting yourself. If you want a structured way to build that daily fluency, our course on mastering Claude Code to ship code faster and build AI agents covers the workflows this kind of agent tooling is pushing everyone toward. It shows where the tooling is heading: agents that don't stop at "code written" but keep working until "PR merged." That shifts the human role further from writing and fixing code toward reviewing outcomes and deciding what gets built next, which is exactly the kind of judgment call PMs are already used to making.
If you're evaluating GitHub Copilot for your team, it helps to know what you are actually paying for that autonomy: see our breakdown of what a product manager actually pays for Claude Code vs Cursor for a sense of how the economics compare. The practical takeaway is this: Agent Merge is opt-in, per-session, and still a preview. Try it on a low-stakes pull request first, watch how it handles a real review comment or a real failed check, and decide from there whether you trust it on something that matters.
