Choose how the app is built
| Route | Language | Good to know |
|---|---|---|
| Kotlin with Jetpack Compose | Kotlin | Google's recommended way to build native Android apps; Android only |
| Flutter | Dart | One codebase for Android and iPhone, with its own interface toolkit |
| React Native, usually with Expo | JavaScript or TypeScript | One codebase for Android and iPhone; familiar if you know web development |
Pick Kotlin if the app is Android only or needs deep access to the device. Pick Flutter or React Native if you want an iPhone version from the same code. AI models write all three well, because all three are widely used and documented.
What the AI does well
- Creating the project and explaining its folders.
- Building screens, navigation, lists and forms from a description or a screenshot.
- Storing data on the device and calling a server.
- Reading long build errors and fixing them. Android builds use a tool called Gradle, and its error messages are long; paste them in full.
- Writing tests and the text for the store listing.
It is weaker at things it cannot observe: how the app behaves on a small, slow phone, battery use, and settings that live in the Google Play Console website.
Android Studio and the emulator
Android Studio is Google's free development tool. Installing it gives you the Android SDK, the kit of libraries and command-line tools every Android build needs, even if you write your code in another editor.
It also includes the emulator, a pretend Android phone on your computer. You can create several virtual devices with different screen sizes and Android versions, which matters because Android phones vary a lot.
Testing on a real phone is simple. Turn on developer options and USB debugging on the phone, connect it, and the build tools can install your app directly. Test on at least one real device, ideally an older one, before you publish.
Signing
Every Android app must be signed: stamped with a private key that proves updates come from the same author. The key lives in a file called a keystore.
- Debug builds are signed automatically with a test key, so you can ignore signing while you develop.
- A release build needs your own key. Create it once and back it up somewhere safe.
- Never put the keystore or its password in your code or in a public repository. Tell the AI this rule at the start and add the file to your ignore list.
- Google Play offers a service that manages the final signing key for you. Check the Play Console documentation for how it currently works.
Publishing
To publish on Google Play you need a Google Play developer account. You upload a release file, fill in the listing, answer the content and privacy questions, and send it for review. Google sets testing requirements for some accounts before an app can go public, and the rules change, so read the current Play Console guidance early.
You can also share the app file directly with people for testing, without the store.
A sensible step-by-step
- 1. Write a short brief: purpose, main screens, data.
- 2. Pick the framework and install Android Studio.
- 3. Ask the AI to create the project and get an empty app running in the emulator first.
- 4. Build one screen at a time and run the app after each.
- 5. Add data, sign-in and permissions. Ask only for the permissions the app needs.
- 6. Test on a real phone and on a small and a large virtual device.
- 7. Create your release key and make a signed release build.
- 8. Prepare the listing, run a test release, then publish.
Save a checkpoint after each working step, as described in what is vibe coding. For the Apple side, see how to build an iPhone app with AI.
Doing this in VibeCode.surf
VibeCode.surf is a general desktop AI code editor, in beta. It has no Android emulator of its own and no Play Store publishing. It handles the coding side: agents write Kotlin, Dart or TypeScript, keep a plan in PLAN.md and project notes that every agent reads, and run the command-line tools you have installed, such as Gradle, Flutter or Expo commands, after you approve each one. You start long-running things like a development server yourself in the built-in terminal, and a Fix with agent button sends a terminal error to the active agent. The emulator still comes from Android Studio.
If your framework can also run the app as a web page on localhost, agents can check layouts at phone and tablet size in their browser. That is a layout check, not a replacement for the emulator or a device. Agents are confined to the project folder, so keep your keystore outside it. The app builder guide has more, and the model guide helps you pick a model. There is no debugger with breakpoints and no language-server autocomplete yet, so keep Android Studio for those. macOS is the tested platform; the Windows and Linux builds are new.