The interesting part of this release is not that Copilot can write code. It is where the work now starts. Since 23 July, the trigger is no longer a developer opening an editor — it is a ticket being assigned in Linear.
How the loop works
Assign a Linear issue to Copilot and, according to GitHub’s changelog, the agent reads the issue contents, works in an ephemeral development environment, streams progress updates back into the Linear activity timeline, and opens a draft pull request that it then requests review on.
The GA release adds the controls the preview was missing. You can now choose which model handles the job, point it at a custom agent, set both the base branch and the working branch, and steer a session mid-flight by mentioning Copilot in the issue comments with new instructions.
What it costs and what it needs
The integration is available on Copilot Pro, Pro+, Business and Enterprise. Setting it up takes two things that are easy to underestimate:
- The GitHub Copilot for Linear app, installed from the GitHub Marketplace
- Organisation owner permission in GitHub and workspace admin permission in Linear — one person rarely holds both
The part worth thinking about before you enable it
A draft pull request that appears without a human having opened an editor is a genuinely different review situation. Nobody on the team watched the reasoning happen, so the diff arrives with no context beyond the issue text and whatever the agent streamed into the timeline.
That is manageable, but it is a process change rather than a tooling change. Teams that already require a named reviewer on every pull request will absorb it easily. Teams where PRs get approved on trust because "Nam wrote it and Nam is careful" will find that heuristic stops working, because Nam did not write it.
The steering controls help here. Being able to redirect a session from the issue comments means a reviewer can correct course before the diff is finished rather than rejecting it afterwards — closer to pairing than to reviewing.
Where this sits against the alternatives
This is a different shape of automation from what n8n does. n8n connects systems and moves data between them on triggers you define; the Copilot agent takes an open-ended engineering task and produces a change for a human to approve. They overlap only in that both remove a manual step.
It is also distinct from Cursor, which remains the stronger tool once a person is actively editing — it scores 9.0 against Copilot’s 8.6 largely on multi-file implementation work. The Linear agent is aimed at the tickets nobody has started yet: the small, well-specified, tedious ones that sit in the backlog because they are not worth interrupting anyone for.
That is a real category of work. Whether an agent should be trusted with it is a question each team answers by looking at its own review discipline, not by reading a changelog.