跪拜 Guibai
← Back to the summary

Android Studio Rabbit 2 Opens the IDE to 40+ AI Agents, Drops Gemini-Only Lock-In

Foreword

[Image]

October 28, Android Studio Rabbit 2 stable release: You can freely choose your AI Agent, no longer just Gemini.


📋 Table of Contents


🤔 What is Rabbit 2?

The latest stable version of Android Studio: an Android IDE open to 40+ third-party AI Agents.

[Image]

Key differences from Rabbit 1

Dimension Rabbit 1 Rabbit 2
AI Agent Gemini only 40+ ACP-compatible Agents
Agent choice No choice Claude / Codex / Junie / Antigravity, etc.
Skills Manager None ✅ 30+ skills, invoked with @
AI Change Review None ✅ /review command
LeakCanary integration None ✅ Built-in + "Fix with AI"
XR emulator Beta Stable
Gemini Enterprise ✅ ✅

Rabbit 1's AI capabilities were already strong, but it had a fundamental limitation: you could only use Google's AI.

Rabbit 2 removes that limitation: The IDE is no longer tied to a single AI vendor; it becomes a workbench for Agents. You pick the Agent, the IDE provides the tools.


🔌 BYOA: How do 40+ Agents get integrated?

BYOA stands for Bring Your Own Agent: you pick a coding Agent of your own choice inside the IDE.

[Image]

Which Agents are supported?

[Image]

Four are officially named:

Agent Provider Features
Claude Agent Anthropic Long-context understanding, code refactoring
Codex OpenAI Code generation, multi-language support
Junie JetBrains Deep IDE integration (JetBrains' own)
Antigravity Google Android-specific optimization, built-in Android knowledge

In total, 40+ ACP-compatible Agents are available.

ACP = Agent Communication Protocol, an open protocol defined by Google. Any Agent implementing this protocol can integrate with Android Studio.

How to use it?

In Android Studio's Agent Panel, you can freely select and switch the Agent you want.

[Image]


🛠️ What parts of the IDE can Agents access?

[Image]

In Android Studio, Agents can directly operate on the IDE's core tools:

Tool What Agents can do
Project structure Read project directories, module dependencies, build configs
Build system Trigger builds, read build errors
Symbol resolution Jump to definitions, find references, understand types
Compose preview Render @Preview, visually verify UI changes
Emulator Install APKs, operate the emulator, take screenshots
Logcat Read runtime logs, filter errors

What does this mean?

Traditional AI coding tools (like the ChatGPT web version) require you to paste your code before they can understand context.

But Agents inside Android Studio actually see your project structure, build artifacts, and runtime results. They are already sitting at your workstation.

The ceiling of an Agent's capability is not determined by the model itself, but by how many IDE tools it can reach. Rabbit 2 hands Agents a complete toolbox.


📦 Skills Manager: Installing "Skill Packs" for Agents

[Image]

Agents have tools, but they may not know "what Android's best practices are."

Skills address this — equipping Agents with Android domain expertise.

What are Skills?

[Image]

Skills are programmatic guidance packs officially provided by Google: Google's decade-plus of Android best practices, packaged as instruction sets Agents can execute directly.

Currently there are 30+ skills covering 20+ domains:

Domain Example skills
UI migration XML View → Jetpack Compose migration
Build upgrade Upgrade AGP 9
Layout adaptation Adaptive layouts (tablets/foldables)
Camera integration CameraX integration
Performance optimization Compose performance anti-pattern checks
Security review Intent security, permission security
Policy compliance Google Play policy pre-submission review
Firebase Firebase service integration

How to use?

[Image]

For example, if you want to migrate an XML layout to Compose:

@android-xml-to-compose help me migrate activity_main.xml to Compose

The Agent will automatically determine the migration scope, draft a plan, and convert the layout files, View Binding code, and logic to Compose.


🔍 Extra Features

There are also some very practical extra features: AI Change Review + LeakCanary

[Image]

AI Change Review

Rabbit 2 has built-in AI-driven code review:

Agent writes code → AI reviews code → Agent fixes issues. The whole chain is closed loop inside the IDE.

LeakCanary integration

[Image]

LeakCanary is the most popular memory-leak detection library for Android. Rabbit 2 integrates it directly into the performance profiler.

Key changes:

[Image]

Memory-leak hunting has turned from "manual investigation" into "one-click Agent fix." This is the most concrete example of Rabbit 2 baking Agents into the toolchain.


🧑‍💻 What does this mean for Android developers?

[Image]

Which Agent should you choose?

Your situation Recommendation
Enterprise projects with compliance requirements Gemini Enterprise — data privacy, audit logs, admin policies
Personal projects, want strongest code generation Claude Agent or Codex — try for two weeks, see which fits your coding style
Heavy JetBrains user Junie — deepest integration with the IntelliJ ecosystem
Pure Android development, not cross-platform Antigravity — built-in Android knowledge, Compose preview, device management
Not sure which one Start with Antigravity (free and Android-specific), switch when you hit limits

Choosing an Agent isn't about picking the "strongest model" — it's about picking the "partner best suited to your workflow."

The significance of Rabbit 2 isn't that any single Agent is powerful; it's that Android Studio has transformed from an "AI tool" into an "Agent platform." This shift matters more than any single feature.


Finally

[Image]

To summarize Android Studio Rabbit 2 in one sentence:

The IDE is no longer tied to a single AI vendor. You pick the Agent, the IDE provides the toolbox.

Which Agent would you choose as your first partner? Discuss in the comments!

Comments

Top 2 from juejin.cn, machine-translated. The original thread is authoritative.

NullPointExceptio197

Let's try the update.

红富士的星期五

Will a relay/proxy service key work?