Within 48 hours last week, GitHub shipped Copilot into two new places simultaneously: Xcode 27.3’s built-in coding assistant and JetBrains Air. No new model, no benchmark press release — just distribution. That framing matters. The real story is not what Copilot can do; it is where your Copilot session now lives, and what that means for your team’s budget.
Copilot Is Now Inside Xcode
Apple’s Xcode already had ChatGPT, Claude, and Gemini. Copilot was the last major AI coding assistant that required a separate extension. That gap closed November 4 when GitHub announced Copilot is now available in Xcode’s built-in coding assistant, currently in public preview with Xcode 27.3 beta. The stable release follows Xcode 27.3’s general availability, expected in December.
Setup is minimal: download Xcode 27.3 beta, then navigate to Xcode > Settings > Intelligence and add Copilot. All plan tiers — Free through Enterprise — are supported. The macOS requirement is Tahoe 26.6 or later.
Two paths now exist for Copilot in Xcode. The new built-in route lives inside Apple’s native assistant infrastructure: fewer moving parts, but a slower update cadence tied to Xcode releases. The existing GitHub Copilot for Xcode extension remains available and will likely receive updates faster during the preview period. The practical recommendation: trial the built-in on one machine, keep the extension on others until the preview graduates to stable.
JetBrains Air: Local Tasks and Cloud Tasks
The JetBrains story is architecturally different. Copilot is now one of five selectable agents in JetBrains Air — alongside Junie, OpenCode, Pi, and Cline — all running on the Agent Client Protocol (ACP). The protocol abstracts the agent from the editor, which is what makes multi-agent selection possible in a single interface.
There are two execution modes. Local tasks run Copilot inside your existing JetBrains IDE using your own environment. Cloud tasks spin up an isolated Docker container on a fresh virtual machine, run the task, then terminate — no state carries over between runs. Cloud mode requires a JetBrains business account with Air Teams enabled by an admin; it is not on by default.
The Billing Reality Nobody Talked About
This is where most coverage stopped paying attention. Cloud tasks in JetBrains Air bill against your Copilot plan, not your JetBrains AI credits. That matters for three reasons.
First, the credit denominations are not equivalent. GitHub charges $0.01 per credit; JetBrains charges $1.00 per credit. If your finance team is comparing credit balances across the two platforms, that is a 100x difference waiting to cause a surprise invoice.
Second, Air’s analytics dashboard does not include Copilot usage. Your organization’s consumption of Copilot through Air will not appear in Air’s governance reports — you have to monitor from the GitHub billing dashboard. Admins can disable Copilot through an Air governance policy, but cannot set spending caps inside Air itself. Those controls exist only on the GitHub side.
Third, the Copilot Fusion discount — currently 30% off model costs — expires November 30. Teams evaluating cloud task costs have about three weeks to run those experiments at the lower rate.
What Your Team Should Do Now
- Swift and iOS teams: Download Xcode 27.3 beta on one machine and test the built-in integration. Keep the extension active elsewhere. Document behavior differences on your actual codebase, not toy examples.
- JetBrains shops: Check whether Air Teams is enabled in your organization. Cloud tasks will not work until an admin enables the Copilot agent — this prerequisite is easy to miss during evaluations.
- Finance and platform leads: Set budget alerts on the GitHub side — thresholds at 75%, 90%, and 100% — before cloud tasks start generating charges. Air will not surface Copilot spending figures for you.
- Everyone: Note the November 30 Fusion discount deadline. After that date, standard model rates apply. Run your cost-per-task benchmarks now.
The Bigger Picture
Copilot is now native in Xcode, JetBrains, VS Code, Visual Studio, Neovim, and Eclipse. GitHub did not ship a new model this week — it built ubiquity. The question for most developer teams is no longer whether to use Copilot. It is which agent gets which task, and what that costs when the sandbox invoices arrive.
