GitHub Copilot lands in JetBrains Air and Xcode alongside new agent oversight controls

GitHub spent a single week pushing Copilot into two new places where developers write code: JetBrains' Air environment and the coding assistant built into Xcode. According to a post on dev.to summarizing the announcements, the expansions arrived together with a batch of features in the Copilot app aimed at supervising agent work — session dashboards, review controls and merge automation.

Copilot arrives in JetBrains Air

As the dev.to write-up describes it, Copilot in Air supports tracking multiple sessions at once, with tasks running either locally in the IDE or in the cloud through Air Teams, billed against a user's existing Copilot subscription. The notable part, the post argues, is that GitHub now treats several agents running simultaneously as a normal working state that deserves a dashboard rather than an unusual case.

The post also raises several caveats. The announcement reportedly says nothing about resource limits, performance, or how work moves between local and cloud execution, and the conditions around plan eligibility and Air Teams are left unstated. It also draws a distinction between watching sessions and coordinating them: nothing in the announcement suggests agents share context, divide a project between them, or reconcile conflicting edits.

A native but preview-stage Xcode integration

On the Apple side, Copilot is now available inside Xcode's built-in coding assistant as a public preview, starting with the Xcode 27.3 beta. That means no separate extension or companion tool for Apple developers. The dev.to post cautions, however, that there is no general-availability date, no stated support commitment and no published list of limitations. It also warns against assuming feature parity: the session-management capabilities are documented for JetBrains only, and each integration appears to have its own scope and gaps.

The Copilot app gains review and session controls

The less headline-friendly changes may matter more in practice. Per the dev.to summary:

• Agent Merge can be configured from a pull request and now remains active while required checks are still pending, merging once everything passes. Required checks still gate the merge; the change is that the automation no longer turns itself off during the wait.
• Reviewer status, additional review requests and the ability to quote selected comment text to discuss it with Copilot are all available in the app. Discussing a reviewer's comment remains distinct from automatically implementing what was asked, and the post notes GitHub does not claim otherwise.
• Session housekeeping tools include archiving pinned sessions whose pull requests merged or closed, pinning and unpinning in bulk, splitting nested chats without losing the thread, and more than ten saved custom views in My Work.
• When a conversation hits its context limit, users can compact it or inspect what is consuming the space. The limit itself remains, and compaction offers no guarantee that every detail survives.

Autonomous approval stays local

One decision stands out for its restraint. Personal instructions that let Copilot approve certain tool requests without asking are opt-in and restricted to local sessions. The dev.to author reads this as GitHub deliberately drawing a line: convenience and autonomy where the developer can observe everything, explicit approval everywhere else. In a market where many agent vendors are pushing to reduce how often they ask permission, that boundary is itself a product decision worth noting.

Why it matters

The question of which editor an AI assistant lives in is close to settled — with Air and Xcode covered, Copilot now spans the major IDEs and Apple's toolchain natively. What separates offerings now is supervision: dashboards for parallel sessions, structured review conversations, merge automation that respects required checks, and clear limits on autonomous approval. The unglamorous features — archiving sessions, saved views, context compaction — are the ones that determine whether agent-assisted development scales or collapses into unmanageable tabs. At the same time, the open questions are real: Xcode support is a preview with no GA date, plan eligibility for cloud execution is undocumented, and capabilities differ between integrations. Teams considering adoption should verify the specifics of each environment rather than assume the feature set travels with the Copilot brand.