Four Developer Tools Host Other Vendors' Coding Agents, but Their Pages Differ on Who Lists Agents, Which Account Is Used and Where Usage Shows

Overview

Apple, Google, JetBrains and GitHub each now document a way to run another vendor’s coding agent inside their own developer tool: Xcode, Android Studio, Air and GitHub itself. The vendors’ pages agree that the developer may choose, and differ on three practical points: who lists the agents, which account an agent runs on, and where usage, policy and audit records are described. This Analysis compares the four using their own pages as read on November 10, 2026, and reports from 9to5Mac, The Decoder, The Register and InfoWorld. Parts of this are preview or beta software, as noted below. Sentences that are the Herald’s own reading, not a source’s statement, say so.

Background: four hosts, four dates

GitHub’s February 4, 2026 post said “Claude by Anthropic and OpenAI Codex are now available in public preview on GitHub and VS Code with a Copilot Pro+ or Copilot Enterprise subscription”, and that each coding agent session “consumes one premium request”.

On March 10, The Register reported that JetBrains Air supported OpenAI Codex, Anthropic Claude Agent, Google Gemini CLI and JetBrains Junie, and that it “can use the Agent Client Protocol (ACP), which Zed and JetBrains sponsor, as a vendor-neutral protocol for agent-editor communication”. The Herald previously reported on JetBrains’ March announcements, which included IntelliJ IDEA 2026.1 support for external agents through ACP.

On October 28, Google’s Android Developers Blog said Android Studio Rabbit 2 was “now stable and ready for you to use in production”, with a Bring Your Own Agent (BYOA) feature and a choice of “over 40 coding agents”. The Decoder reported the same day that developers can install and run Anthropic’s Claude Agent, OpenAI’s Codex and Google Antigravity through the “built-in marketplace”.

On November 3, two further steps arrived. 9to5Mac reported that Apple released Xcode 27.3 beta 1 with GitHub Copilot in the built-in coding assistant, after earlier additions of ChatGPT in Xcode 26, followed by Claude, Codex and Claude Agent in Xcode 26.3 and Google Gemini in Xcode 27; the piece is built around a one-line Apple release note. Also on November 3 (GitHub’s entries are dated in US Pacific time), a GitHub changelog entry announced Copilot in Air over ACP, which JetBrains describes as running “the official GitHub Copilot CLI over the Agent Client Protocol (ACP)”. The Herald has covered the Xcode and Android Studio steps as separate reports.

Current situation: three questions, four answers

Host	Who lists the agents	Which account the agent runs on	Usage, policy and audit, as the vendor pages describe them
Xcode	Apple’s page says to “turn on the agents that you want to use” in Intelligence settings, and that “You can also use agents and tools from other providers that don’t appear under Agents”; an ACP agent can be added by hand	Per Apple: “If you have an account, sign in”; for ChatGPT, “a free ChatGPT account, or a paid account with higher limits”. For Copilot, GitHub says all Copilot plans are supported	Per Apple, a device-management key, CodingAssistantAllowExternalIntegrations set to false, turns off the coding assistant on managed devices; the page does not describe a usage view, spending limit or audit log
Android Studio	Google offers “over 40 coding agents”; The Decoder describes a built-in marketplace	Google’s setup steps end with authenticating the chosen agent, and it frames the choice in terms of “the cost efficiency of a particular agent for your project” and of needing “a specific provider to align with an enterprise agreement”	Google’s post lists “Compliance & Audit Logging” and “Admin Policies” under Gemini Enterprise, a separate section; it does not say whether they cover agents added through BYOA, and it describes no usage view or spending limit
Air	JetBrains says Copilot is in the agent picker, in “the same task list as Claude, Codex, and Junie”	GitHub: “use your existing Copilot subscription” and “Tasks use your Copilot plan rather than JetBrains AI credits”. JetBrains: “GitHub, not Air, handles your Copilot account”	GitHub: “Air’s governance and analytics features do not include Copilot usage”, and administrators “can disable GitHub Copilot through a governance policy, but cannot set Copilot spending limits within Air”
GitHub	The documentation lists Anthropic Claude and OpenAI Codex under “The following third-party agents are supported on GitHub”; agents must be enabled in account settings, organization policies or an enterprise’s “Agents” AI controls	Per the documentation, agents “can be powered by your existing Copilot subscription” and consume “GitHub Actions minutes and AI credits”	Per the documentation: “Actions taken by these GitHub Apps will be visible in your audit log”, and “These policies do not apply to local agents in Visual Studio Code”

Two cautions apply. GitHub labels its third-party agents a public preview, and Copilot in Xcode is, per GitHub, a public preview in a beta build. And one figure differs between two GitHub pages: the February 4 post says a session consumes one premium request, while the documentation page read on November 10 says each session consumes AI credits “based on the model used and the number of tokens processed”. Separately, InfoWorld, in an October 30 report on announcements at GitHub Universe, said GitHub is changing how enterprises purchase Copilot by “shifting to an AI credit-based consumption model”.

Implications

The Herald’s reading. The protocol and the pickers made the choice of agent portable between hosts, but the pages do not make the record of what an agent used or did portable. Where the host is also the account holder, as with GitHub’s partner agents, GitHub’s documentation describes AI credits, audit-log entries and enterprise controls. Where it is not, the pages are thinner: GitHub says Air’s governance and analytics features do not include Copilot usage, and JetBrains tells users “you have to keep track of it on GitHub”; Apple’s page and Google’s post describe no usage view or spending limit for an added agent, and Google places audit logging under Gemini Enterprise. None of the cited sources states this contrast as a rule, and four hosts are a small sample.

What it means for developers, on that reading. A team choosing where to run an agent has to ask the question per pair of host and agent: which account is billed, which dashboard shows the usage, and whether a policy set in one place reaches the other. GitHub’s own documentation shows the same boundary inside one vendor, because its agent policies “do not apply to local agents in Visual Studio Code”.

An outside view. Amit Kumar Jena of Kanerika, quoted by InfoWorld on GitHub’s Copilot overhaul, said CIOs should “monitor agent activity through audit logs” and “set credit limits for individual projects”. His remarks concern Copilot’s own app and Projects feature, not bring-your-own agents. The Herald notes only that spending limits are one of the controls GitHub says Air does not provide for Copilot.

What We Don’t Know

• None of the cited independent outlets (9to5Mac, The Decoder, The Register and InfoWorld) reports a test of how any of the four hosts bills, limits or logs an agent the developer adds. The behaviour described here rests on vendor pages as read on November 10, 2026, and 9to5Mac and The Decoder largely restate vendor announcements.
• Apple’s page, as read, does not say how usage or spending for a third-party agent is shown, and Google’s post does not say whether Gemini Enterprise audit logging and admin policies reach BYOA agents.
• GitHub’s changelog entry for Air was modified on November 6 after its November 3 publication; the text quoted is as read on November 10.
• None of the cited sources says whether the shift to AI-credit billing that InfoWorld reported is the reason the billing unit for partner-agent sessions differs between GitHub’s February post and its documentation.