Workshop Overview: 3 Acts

AI tooling in Android Studio has evolved from simple autocompletion into an interactive co-developer across the entire engineering lifecycle.

• Act 1: Daily In-Editor AI & Guardrails — AGENTS.md, .aiexclude, in-editor diffs via context menu, Java-to-Kotlin math transforms, and XML-to-Compose.
• Act 2: Design to Code & Agent Mode — Wireframe multimodal generation, Transform UI in Previews, GitHub MCP Server, and multi-module self-healing Gradle builds.
• Act 3: Quality, Verification & CLI — In-memory test fakes with Turbine, Logcat crash diagnostics, Studio Journeys E2E tests, and the headless Android CLI.

Modern Android Challenges

Today's Android codebases demand rigorous architectural patterns and rapid modernisation:

• Multi-Module Overhead: Coordinating changes across :feature, :core:ui, :core:data, and :core:model.
• Legacy Debt: Migrating older Java utility classes and XML layouts to Jetpack Compose Material 3.
• Reactive State: Moving from legacy LiveData streams to type-safe Kotlin StateFlow.
• Deterministic Testing: Verifying asynchronous UI and data streams without fragile mockito overhead.

Introducing ICanHazStream

me.kartikarora.icanhazstream

Our companion application is a multi-module streaming availability finder. The starter app is immediately functional and interactive (Explore screen with trending movies, movie detail Where to Watch navigation, and Watchlist).

Modern Tech Stack & Standards

Built on the latest Android & Kotlin toolchain:

• Android Studio Rabbit 2 (2026.2.2 Canary 3): Studio Labs, Agent Mode & bundled JDK 25.
• Kotlin 2.4.20 & Gradle 9.8.1: Type-safe build-logic convention plugins.
• Jetpack Compose BOM 2026.09: Material 3 design system with Albert Sans typography.
• JetBrains Ktor Client 3.6.0: Clean HTTP networking with kotlinx.serialization.
• Testing with Turbine 1.2.1: Streamlined testing of StateFlow with in-memory test fakes.

Guardrails Architecture

Grounding Gemini with explicit architecture rules while safeguarding secrets:

• AGENTS.md: Mandates MVVM, Clean Architecture, and reusing pre-built :core:ui components.
• .aiexclude: Completely blocks local.properties, keystores, and API credentials from AI context indexing.

Grounding with Brand Theme Skill

Ground Gemini in our pre-built Compose design system so it produces consistent, branded composables instead of raw unstyled components:

# Install the brand AI skill into Android Studio project
npx skills install https://distribute.kartikarora.me/ai/kartikarora-compose-theme.skill
                

• Teaches Gemini about MovieCard, ProviderBadge, and RatingChip in :core:ui.
• Automatically applies Albert Sans & JetBrains Mono design tokens.

In-Editor Prompts & Live Diffs

Refactoring LiveData to StateFlow with Gemini in-editor live diffs:

• Highlight legacy MutableLiveData streams in TrendingMoviesViewModel.kt.
• Right-click in the editor and choose Gemini > Generate Code with Gemini (or Refactor with Gemini).
• Review side-by-side diffs before committing changes into your file.

// In-editor prompt result:
private val _movies = MutableStateFlow<List<Movie>>(emptyList())
val movies: StateFlow<List<Movie>> = _movies.asStateFlow()
                

Modernising Legacy Java & XML

Modernising movie cards from initial baseline/XML into full Compose MovieProviderCard with :core:ui design tokens:

Refining UI from Wireframe Mockup

Refining and generating the Where to Watch detail screen from wireframe mockup sketches:

Tweaking UI in Compose Previews

Iterate on user interfaces visually without re-running the application:

• Open WhereToWatchScreen.kt with @PreviewLightDark.
• Click Transform UI with Gemini in the Compose Preview toolbar.
• Prompt: “Add a subtle outline around the 4K Ultra HD badge and increase provider spacing to 12dp.”
• Changes reflect live in preview rendering instantly.

Connecting Context via GitHub MCP

Model Context Protocol connects Gemini to upstream GitHub repositories:

Scaffolding Features with Agent Mode

Scaffolding the Watchlist feature across modules (:feature:watchlist, :core:data, navigation graph) with Agent Mode:

Testing with In-Memory Fakes & Turbine

Test reactive streams deterministically without complex mock frameworks:

@Test
fun testChangingRegionUpdatesProviders() = runTest {
    val fakeRepo = FakeMovieRepository()
    val viewModel = WhereToWatchViewModel(fakeRepo)

    viewModel.uiState.test {
        assertThat(awaitItem()).isInstanceOf(UiState.Loading::class.java)
        
        fakeRepo.emitProviders(listOf(StreamingProvider("Netflix", Quality.UHD_4K)))
        val state = awaitItem() as UiState.Success
        assertThat(state.providers.first().name).isEqualTo("Netflix")
    }
}
                

Studio Journeys: Plain-English E2E Tests

Studio Journeys E2E test verifying live navigation from Explore → Dune: Part Two → Stream tab → Netflix 4K AU$16.99 → Watch:

Headless Automation: Android CLI

Execute Android Studio symbol indexing, preview renders, and lint fixes directly in terminal:

App Quality Insights & Gemini

Correlating production Crashlytics and Google Play Vitals with source lines for instant fixes:

Workflow Summary

The modern AI-assisted Android development lifecycle:

• Ground with Intent: Use AGENTS.md and theme skills so the model adheres to your architecture.
• Accelerate In-Editor: Right-click and choose Gemini from the context menu for instant, reviewable diffs and code transformations.
• Scaffold in Agent Mode: Let the agent coordinate multi-module changes and self-heal Gradle builds.
• Verify Rigorously: Assert state with Turbine, execute Studio Journeys, and automate with the Android CLI.

On-Device AI Options

Bring inference directly to user devices with zero server latency:

• Gemini Nano via AICore: System-managed on-device foundation model on Pixel and supported flagship Android devices.
• Google Play services ML Kit GenAI Prompt API: Zero-setup on-device text generation with automated model downloading and fallback.
• Offline Availability & Privacy: Process sensitive queries, movie summaries, and recommendation rankings entirely on-device.

Dig Deeper & Resources

Keep exploring with these workshop links and documentation:

• Codelab: codelabs.kartikarora.me/gemini-in-android-studio
• Starter Repo: github.com/kartikarora/icanhazstream
• Gemini in Studio Docs: developer.android.com/studio/gemini/overview

Codelab

codelabs.kartikarora.me

Happy building with Gemini in Android Studio!