Android Studio Rabbit 2 AI agents are no longer a preview experiment. On October 28, 2026, at the Android Dev Summit, Google made Bring Your Own Agent stable in Android Studio Rabbit 2, so developers can run more than 40 coding agents that speak the open Agent Client Protocol (ACP) directly inside the IDE.
That list includes Anthropic’s Claude Agent, OpenAI’s Codex, and Google’s own Antigravity. You install them from the marketplace, switch between them as you work, and each one inherits the IDE’s understanding of your project.
What Google actually shipped
The headline is choice. Until now, Android development pushed you toward Google’s own assistant. Now you pick the agent that fits the task, and the agent gets real tools: project structure and build configuration, symbol resolution, Compose Preview, the emulator, and Logcat.
Google also gave agents deeper access than a chat window ever had. Agents can investigate bugs, test changes, and verify their own fixes, using devices, performance profilers, and production crash reports. In plain terms, the agent stops guessing and starts checking its own work.
Three supporting pieces matter just as much:
- Change Review. A new agentic code review experience that reads pending agent changes, reports potential issues, and walks you through long diffs. Invoke it with the /review command and have your agent fix what it finds.
- Android CLI upgrades. Terminal-based agents gain device testing, UI validation, performance profiling with Perfetto traces, and app-quality analysis that mirrors Google Play’s own checks before upload.
- More than 30 agent skills across 20+ areas. These cover permission security, Glance widget layouts, Compose performance, and other Android-specific jobs that generic agents usually fumble.
The evidence: what is stable, what is still early
| Detail | What the source says |
|---|---|
| Announced | October 28, 2026, at Android Dev Summit |
| Stable now | Bring Your Own Agent in Android Studio Rabbit 2; 40+ ACP-compatible agents |
| Agents named | Claude Agent (Anthropic), Codex (OpenAI), Antigravity (Google) |
| Code review | Change Review: reports issues in pending agent changes, with AI-guided walkthrough |
| CLI additions | Device streaming, Compose previews without an emulator, Perfetto profiling, app-quality analysis |
| Agent skills | 30+ skills across 20+ areas, including permissions, widgets, and Compose performance |
| Still in alpha | ML Kit Agentic API powered by Gemini Nano 4; agent-to-UI interfaces that render native Jetpack Compose |
| Still in beta | AppFunctions, which let apps expose functions for authorized AI agents to discover and execute |
ACP, the Agent Client Protocol, is the open standard letting different coding agents plug into the same IDE. AppFunctions are capabilities an app exposes so an assistant can run them with permission.
How to run a second agent in your Android project
1. Open Rabbit 2 — Update to the stable Android Studio Rabbit 2 release.
2. Install your agent — Go to Settings > Tools > AI > Agents and pick Claude Agent, Codex, or Antigravity from the marketplace.
3. Give it a real job — A bug with a crash report, not a toy example, so you can judge its fixes.
4. Run Change Review — Read the issue report before you merge anything the agent wrote.
5. Ship only reviewed code — Your name is on the commit, not the agent’s.
How to profit from it
Android freelancers sell speed and reliability. This release gives you both, if you package it right:
- Sell an agent-assisted build sprint. Quote fixed-price Android builds with shorter timelines: the agent handles boilerplate, tests, and crash triage while you handle architecture and review.
- Offer a crash-triage retainer. Agents with profiler and crash-report access are built for monthly work: find the crash, fix it, prove it with a test.
- Productize migration work. The Canary preview’s XML-to-Compose migration assistant is a service in itself: convert legacy layouts, verify with Compose Preview, and charge per screen.
- Learn one agent deeply. Clients hire proven workflows, not long tool lists.
What to watch before you switch
First, the Agentic API and agent-to-UI pieces are alpha, and AppFunctions is beta. Build client work on the stable BYOA feature, and treat the rest as experiments until Google moves them up.
Second, no pricing for the agents, CLI, or Agentic API was given in the announcement coverage. Your agent subscription remains your cost.
Finally, agent-written code still needs a human reviewer. Google’s own answer to messy agent diffs is Change Review, which tells you everything about where the risk sits.
Sources: Google’s Android Developers Blog announcement as reported by The Decoder, FoneArena, and XenoSpectrum, linked below. The announcement itself was not fetched directly from developers.googleblog.com in this run, so figures are reported from that coverage.
Frequently Asked Questions
What is Bring Your Own Agent in Android Studio?
A stable feature in Android Studio Rabbit 2 that lets developers install and run more than 40 ACP-compatible coding agents, including Claude Agent, Codex, and Antigravity, inside the IDE.
How do I install a third-party agent?
Open Settings > Tools > AI > Agents in Android Studio Rabbit 2 and install agents from the built-in marketplace. You can install more than one and switch between them.
What can these agents access?
Project structure and build configuration, symbol resolution, Compose Preview, the emulator, Logcat, devices, profilers, and production crash reports.
What is Change Review?
A new agentic code review experience that analyzes pending agent changes, generates a report of potential issues, and offers an AI-guided walkthrough, with a /review command to invoke it.
What is still in preview?
The ML Kit Agentic API is in alpha, AppFunctions is in beta, and the XML-to-Compose migration assistant is in Canary.
Sources: The Decoder — Google opens Android development to third-party AI agents · FoneArena — Android Dev Summit 2026 updates
Keep Reading
Want more tools that actually pay for themselves? Browse the AI Tools section for the latest launches, pricing breakdowns, and profit playbooks — and check the official source linked above before you buy or deploy anything.