Copilot Reaches Xcode and JetBrains Air as Claude Code, OpenCode and Codex Ship Updates

Executive Summary

• GitHub Copilot is now in public preview inside Xcode's built-in coding assistant (Xcode 27.3 beta) and is available in JetBrains Air through the Agent Client Protocol.
• Copilot usage API now reports bring-your-own-key (BYOK) and enterprise custom model activity.
• Claude Code 2.1.314 to 2.1.318 adds write-boundary permissions, gateway budgets and policies, and a warning before sending your account email over the network. Fast mode for Opus 4.8 is deprecated.
• OpenCode 2.2.3 to 2.2.5 adds a built-in Claude Agent provider that reuses your Claude Code login, opt-in artifacts and scheduled prompts.
• OpenAI Codex 0.166.0 is the latest stable release, and 0.168.0 alpha builds are already in flight.

GitHub Copilot goes where developers already work

Copilot in Xcode's coding assistant

On November 4, GitHub announced that GitHub Copilot is now available in Xcode's coding assistant. It is a public preview that starts with the Xcode 27.3 beta and needs macOS Tahoe 26.6 or later. All Copilot plans are supported.

You set it up under Xcode > Settings > Intelligence. The older GitHub Copilot for Xcode extension still exists, so you can pick whichever fits your workflow.

Copilot in JetBrains Air

A day earlier, GitHub shipped Copilot in JetBrains Air. It uses the Agent Client Protocol (ACP) to let you delegate tasks, steer them and review the results.

• Local tasks: install Air, pick GitHub Copilot and connect your account.
• Cloud tasks: need an organization with Air Teams enabled. Each task runs in an isolated Docker container on a fresh VM, behind a proxy, with no state carried over.
• Billing: cloud tasks use your Copilot plan, not JetBrains AI credits.
• Limits: Air's governance and analytics don't cover Copilot usage, and admins can't set Copilot spending limits inside Air.

BYOK and custom models in the usage API

The Copilot usage API update adds used_local_byok, totals_by_local_byok and an is_custom_enterprise_model flag per model. Local BYOK numbers come from client telemetry, so they undercount when telemetry is off. Enterprise custom model counts are observed server-side and are not affected.

For context, last week's Copilot changelog also covered GitHub Agentic Workflows going generally available, Copilot Fusion for routing across models, and GPT-6.1 Terra arriving in Copilot.

Claude Code 2.1.314 to 2.1.318

The Claude Code changelog shows five notable items:

• permissions.blockWritesOutsideWorkingDirectories (2.1.314) makes Write, Edit and NotebookEdit refuse paths outside your working directories in every permission mode.
• Gateway admins get pooled monthly budgets and a layered policy mode (2.1.316), plus a cost report in the admin API (2.1.317).
• CLAUDE_CODE_SUBAGENT_EFFORT sets the effort level subagents run at (2.1.316).
• Claude Code now asks before a web or shell tool sends your account email address over the network (2.1.317).
• claude auth status shows your sign-in expiry, and claude -p warns 10 days ahead (2.1.318).

One deadline matters: fast mode for Opus 4.8 is deprecated as of October 30, 2026, and removed on November 30, 2026. Plan your switch now.

{
  "permissions": {
    "blockWritesOutsideWorkingDirectories": true
  }
}

I'm turning that setting on for my own projects. It is a cheap guardrail against an agent writing outside the repo.

OpenCode 2.2.3 to 2.2.5

The OpenCode releases page lists three releases on October 30:

• A built-in Claude Agent provider that uses your Claude Code login, with guided setup in the provider picker.
• Opt-in artifacts that let agents publish and serve HTML or static files.
• Scheduled prompts under opencode serve that fire only once across multiple server processes.
• compaction.buffer accepts a percentage of the context window, such as 20%.
• Breaking changes: TUI plugins need OpenTUI 0.8 or later, and the managed policy integration.use now covers only integrations. New mcp.use, skill.use and plugin.use actions control the rest.

If you run OpenCode with managed policies, check the policy split before you upgrade.

OpenAI Codex 0.166.0

The Codex releases page marks 0.166.0 (October 29) as the latest stable build. It adds keyboard-driven transcript selection, Markdown rendering fixes in the TUI, codex queue --resume and a codex doctor --network-only diagnostic mode. Alpha builds of 0.168.0 appeared on November 4 and 5, but they carry no release notes yet.

What I take from today

Two patterns stand out. Copilot is spreading into other vendors' editors and agent hosts through open protocols like ACP. And the CLI agents are converging on the same governance features: write boundaries, budgets, per-tool policies and diagnostics. If you manage a team, those controls matter more than the next model bump.

Sources

• GitHub Copilot in Xcode's coding assistant
• GitHub Copilot in JetBrains Air
• BYOK and enterprise custom models in the Copilot usage API
• GitHub Changelog: Copilot
• Claude Code CHANGELOG
• OpenCode releases
• OpenAI Codex releases