Android Studio: choose your AI and check the app

Start with a change you can see

If you want to build an Android app, you can now bring a familiar AI coding assistant into the same place you build and test it. Try a small change first, such as making a form easier to complete. Decide what should improve before asking AI to work.

On October 28, Google released Android Studio Rabbit 2 in its stable channel, the regular release intended for daily work. Android Studio is a computer program for building Android apps. It now supports a choice of coding agents, AI assistants that can work on project files and use development tools.

Claude Agent, OpenAI Codex and Google Antigravity are among the choices. Google says more than 40 compatible agents can connect. This feature first appeared in an early test release on September 24; the material change this week is its arrival in the stable release.

Read the announcement: Android Studio: Rabbit 2 stable release announcement.

Give the assistant a way to check its work

Instead of only suggesting code in a separate chat, a connected assistant can use Android Studio to understand the project, inspect build problems and control the emulator. An emulator is a virtual phone on your computer, where you can try the app.

That gives you a useful learning habit: ask the assistant to show how it checked a change, then repeat the task yourself. A successful build means the software could be put together. You still need to see whether someone can complete the intended task.

Google also described updates to Android command-line tools, which let coding assistants run build and testing tasks through typed commands. Official Android plugins for Claude Code and Codex provide another route for people who already work that way. You can choose the route that fits your experience rather than learning every tool at once.

Check the account and cost before connecting

In the latest stable Android Studio, choose an assistant in the agent window and sign in or supply the access key its provider requires. More choices appear in Settings, under Tools, AI, then Agents.

The connection supports personal or company subscriptions and access keys, depending on the assistant. It does not include a new free allowance for every outside service. Check the chosen provider's access, usage limits and billing before starting. Android Studio's existing built-in agent remains available at no extra charge, according to the release notes.

For a work project, also agree which assistant may access the code and what actions it may take. Start with a practice project containing invented information. Keep release decisions and account permissions clear as you learn.

Practise one complete phone task

Here is a suggested exercise for learning or a guided build. It is a practice plan, not a test we have run on the new release.

1. Choose one screen and write what a person should be able to finish on it.
2. Ask the assistant for a short plan, including how it will check the change.
3. Try the result on the virtual phone, then on a real Android phone when available.
4. Check a blank field, a wrong entry and the back button as well as the normal path.
5. Ask for an explanation of the changed files before accepting the work.

Keep learning how to judge the result

Our perspective is that more choice of assistant is useful when it helps you understand and check your app. Switching assistants will not decide what customers need or whether the finished experience is clear.

You can build those skills now: describe a useful outcome, make a small change and observe someone using it. If you are learning with a trainer or building alongside a partner, bring one phone task and the questions it raises. The next step becomes easier to see.