The coding agent is moving into the toolchain instead of living beside it. Android Studio Rabbit 2 makes that explicit: developers can connect the agent they prefer and give it access to the IDE's project graph, build diagnostics, terminal, and emulator controls.
In its Android Dev Summit announcement, Google says Bring Your Own Agent supports Claude Agent, OpenAI Codex, Google Antigravity, and more than 40 ACP-compliant agents. The same update describes Android CLI features for device streaming, Compose previews, profiling, analysis, and specialized skills.
The IDE becomes the context layer
That is a meaningful change in the agent contract. The model is no longer limited to the files a developer pasted into chat. It can receive build failures, device state, previews, and project relationships from the environment where the work is actually judged.
Choosing an agent is becoming an IDE decision; choosing its scope is still a team decision.
The builder takeaway is to treat agent access like a toolchain permission, not a cosmetic preference. Start with a bounded project, review what the agent can call, and keep tests and device checks in the loop. More context can make an agent faster, but it also makes mistakes more consequential.
Android says Rabbit 2 adds Bring Your Own Agent through ACP, supports agents such as Codex, Claude Agent, and Antigravity, and passes IDE-native project and device context to them.
Android Developers' Blog · Agentic and Flexible ↗
