Prototyping Android apps in Google AI Studio (then moving to real code)
Google AI Studio can spin up a Jetpack Compose app from a prompt. I tried it for a side project and wrote down what stuck and what still belongs in Android Studio.
Most of my mobile work is Expo and React Native. Android native is the lane I dip into when a client needs Kotlin, or when I want to understand what the platform is doing under the hood.
Google AI Studio added a path to generate Android apps from a prompt: Kotlin, Jetpack Compose, Material 3, single-activity. That sounded like a fun weekend experiment, not a replacement for how I ship production apps.
What AI Studio is good at
You describe an app. The agent returns a Gradle project you can preview in a browser emulator, sideload to a phone over WebUSB, or push to Play internal testing. For a throwaway prototype ("show me a flashcard app with dark mode"), it's fast.
The output is readable Compose. ViewModels, string resources, theming. Not perfect, but not gibberish either. Good enough to click through flows before you commit to a build.
If you're learning Compose, seeing a full project appear is oddly useful. You can diff what you would have written against what the model guessed.
Where it stops being enough
AI Studio Android apps are client-side only. No Firebase wiring in that flow, no secrets management, no backend. Fine for a demo. Not fine for anything that needs auth, sync, or a real data layer.
Single-activity, single-module, Compose-only. No NDK, no Wear OS, no TV. Those limits are stated upfront; I hit them within an hour of poking around.
For anything I'd put my name on, I still open Android Studio. AI Studio is the sketchpad. Gradle, lint, dependency updates, App Check, Firebase AI Logic: that's the production path Google documents for Gemini on Android in 2026 anyway.
The handoff I actually use
Prompt in AI Studio → play with UI and navigation → clone the ideas into a proper repo.
When I need Gemini in a real app, I use Firebase AI Logic (firebase-ai), not the old deprecated generative AI client SDK. API key from AI Studio or Firebase console, GenerativeBackend.googleAI(), a sensible model like Flash for latency, App Check before you ship.
AI Studio helped me answer "does this screen layout make sense?" faster than scaffolding by hand. It did not replace knowing where the APK gets signed.
Worth your time?
If you build Android occasionally and want to explore Compose or Gemini features without a blank-project ritual, yes. Spend an evening.
If you need a production app next month, start in Android Studio and treat AI Studio as a reference generator, not the source of truth.
I explore tools like this because my day job is shipping for clients, not because every new Google banner feature becomes my stack. This one earned a bookmark.