Android Studio Opens to Claude, Codex and 40+ AI Coding Agents

Google has opened Android Studio to outside AI coding agents. At its Android Developer Summit on Wednesday, the company said its official app-building software now works with more than 40 agents, among them Anthropic's Claude, OpenAI's Codex and Google's own Antigravity.

The feature, called Bring Your Own Agent (BYOA), ships in Android Studio Rabbit 2, which Google labels a stable release. Developers can download it now.

What agents can do inside the IDE

Android Studio gives each connected agent access to the project structure, builds and the Emulator, the virtual phone used for testing. An agent can run the app on the Emulator and read Logcat, the running log of what the app is doing, to verify its own work rather than return code that only looks correct.

Agents connect through the Agent Client Protocol (ACP), an open standard for linking AI tools to code editors. Because any agent that supports ACP can plug in without a custom integration, the list can grow quickly. Developers open Settings > Tools > AI > Agents, browse a marketplace and can install several agents side by side.

Google also introduced Change Review, an AI-assisted reviewer that scans code changes, including those written by agents, and can walk developers through longer ones. Google acknowledges that large agent-driven changes are hard to review and understand.

Faster debugging, with caveats

Google's Android CLI, a command-line toolkit for building and testing apps, now has official plugins for Claude Code and Codex. In an internal test, Claude Opus 5.5 hunted memory leaks across 45 user journeys. With the CLI it finished in 46 minutes, versus 88 minutes without it. Google credits the gain to analyzing memory snapshots on the computer instead of the phone. The result comes from one run per setup, so it is a demonstration rather than a benchmark.

For users, the potential payoff is fewer slowdowns and crashes in apps, since memory leaks are a common cause. But developers still have to act on what these tools find, so any improvement will arrive gradually.