Google AI Studio can now build Android apps with a database, sign-in and Play publishing
From one prompt, Google AI Studio's Build agent now writes a Kotlin app, provisions a Firebase backend and publishes it to your personal Google Play library.

Google says Google AI Studio can now build full-stack Android apps from a single prompt. The Build agent writes Kotlin code, sets up a Firebase backend and publishes the app to the developer’s personal Google Play library.
A Firebase backend set up by the agent
Until now, Android apps generated in AI Studio were limited to local session storage, according to Google. The Build agent now detects when a prompt calls for persistent storage, user sign-in or generative AI features, and after the user confirms, it provisions the matching services and wires them into the Kotlin project.

- Data: a Cloud Firestore database, the Firebase Android SDKs and Security Rules tailored to the app’s data model. Google says data syncs across devices and caches offline.
- Sign-in: Firebase Authentication with Google Sign-In, plus per-user rules so people can only read and write their own documents.
- Gemini calls: Firebase AI Logic with Firebase App Check, so that API keys are not embedded in the APK. Google says App Check uses Google Fraud Defense, formerly reCAPTCHA Enterprise, to confirm requests come from the genuine app.
Google’s example generates a workout repository class that calls gemini-4-hydrogen through Firebase.ai and saves the result to a Firestore collection. The company says the Build agent now runs on Gemini 4 Carbon and Hydrogen.
Emulator upgrades and a path to your own phone
The in-browser emulator starts faster and has an HD toggle that streams at 1200p. It also plays audio in real time, which Google pitches for testing voice companions built on the Gemini Live API, text-to-speech coaches and games. A screenshot button attaches the live emulator frame to a prompt so the agent can fix layout problems with the picture in view. Users can also ask for custom fonts, which the agent downloads and applies to the Jetpack Compose theme.

Behind the scenes, the agent now uses the official Android CLI to scaffold projects, look up current Android documentation and run Gradle builds. Google says this reduces build errors and supports more complex multi-screen apps.
For testing on a real device, the publish panel lets users set up a Play account in a few clicks. AI Studio builds and signs the Android App Bundle (.aab), and the app appears in the developer’s personal Google Play library, ready to install. According to Google, this removes the need for USB debugging, local keystores or a full developer console account. The post says everything is rolling out today and that testing on your own device from Play is at no cost. It does not name regions, usage limits or paid tiers for the Firebase services, so those details are unconfirmed.

What changes for people who prompt their way to apps
The practical shift is the gap between a demo and something that holds user data. Google’s own sketch-to-app feature in the Gemini app, which we covered earlier, turns a photo of a drawing into a working app. Third-party coverage cited in that article says those Canvas apps run client-side and lack built-in databases, sign-in or server logic. The AI Studio route targets exactly those missing pieces, and it outputs a native Kotlin project rather than a browser-based app.
The model underneath is also new. Google’s Gemini 4 Hydrogen announcement from November 12 describes it as a fast model for everyday coding and agentic workflows, with four configurable thinking levels. It is available in AI Studio and Android Studio, so developers who prefer an IDE have a second route to the same model family.
The security design is the part professional developers will look at first. Putting a Gemini key inside an APK is a well-known mistake, and the agent defaults to routing calls through Firebase AI Logic with App Check instead. Generated Security Rules are a different matter: Google presents them as tailored to the data model, but the post offers no audit data or error rates, so reviewing them before shipping to real users remains sensible. Distribution is also limited by design: the publishing path described is to a personal Play library, not a public store listing.
Everything above comes from Google’s own description. We have not tested the feature, and the company gives no figures on emulator latency or build success rates. Developers can try it by opening AI Studio Build, selecting the Android chip and entering a prompt such as the workout-coach or family-recipe examples Google suggests.
Sign Up for the Daily AI Pulse
One email a day. All the stories that matter.
By signing up, you agree to our Terms of Use and acknowledge the data practices in our Privacy Policy. You may unsubscribe at any time.
Alex CarterAI News & Big Tech CorrespondentAlex Carter writes for Aiholics, keeping readers updated on the fast-paced world of AI and Big Tech.



