Android Studio Rabbit 2: bring your own AI coding agent

Google just stopped forcing Android developers to use its own AI. Android Studio Rabbit 2, now stable, lets you plug in more than 40 AI coding agents, including Claude Agent and OpenAI's Codex, through the open Agent Client Protocol (ACP). It also ships Change Review, a built-in pass that reviews code an agent just wrote. If your mobile app is built by a small team or a contractor, the AI tooling question just got simpler: pick the agent, keep the IDE.

What actually happened

Google announced the release at Android Dev Summit on October 28, in an Android Developers Blog post:

• Bring Your Own Agent: any ACP-compliant agent works. Google names Claude Agent, Codex, Junie and Google Antigravity. Setup is Settings → Tools → AI → Agents, authenticate, prompt.
• Real tool access: agents can drive the Android Emulator, the build system and Logcat, and can load Android skills for bigger jobs.
• Skills manager: install and manage agent skills. Preloaded ones cover Firebase, XML Views to Compose migration, AGP 9 upgrades and adaptive layouts.
• AI Change Review: launch it from Agent Changes, Pending Commits or the Git log. It suggests fixes, then builds a walkthrough that groups the diff by theme instead of by filename.
• Agentic LeakCanary: the built-in agent explains memory leaks and proposes code changes.

The post does not list pricing. Each agent's costs come from its own provider. A companion post covers the rest of the tooling, including an android preview command in Android CLI that renders Compose previews without an emulator.

Why bring-your-own-agent matters for your business

Two things change.

Your agent choice stops being an IDE choice. If your web team already pays for Claude or Codex, the Android work can run on the same subscription, the same prompts and the same skills. One vendor relationship, one bill, one set of habits. And if pricing or quality shifts next quarter, you swap the agent, not the toolchain.

Review becomes a first-class step. Agents write code faster than people read it. Change Review grouping a diff into logical themes is the right idea: it makes a human sign-off on agent output realistic instead of ceremonial. Use it as a gate, not a suggestion. Google's speed claims, like agents finishing UI tasks 77% faster with android preview, come from its own internal testing, so measure on your codebase before you plan around them.

Sources: Android Developers Blog: Your favorite agents are now integrated in Android Studio, Android Developers Blog: What's new in Android developer tools at Android Dev Summit.