Learning Android Development

Google IO 26: Gemini, Big Rocks, Opportunities for Android

In the AI Age, Google continue to thrive in its mobile ecosystem

Google IO 26 Products Image by Sundar https://x.com/sundarpichai/status/2056809925574287490

Understood it’s almost 1.5 month ago since Google IO 26 had happened, I’ve jot down a few tings worth learning about, partly to share, partly serve as my future memory.

This year, it took place on May 19 and 20. The main theme highlights AI models Gemini and Antigravity, AI tools for developers. The event also featured extensive Android content, which this article focuses on, in particular the four areas below.

1. Quick high-level introduction to Gemini models
2. Android Developer Ecosystem introduced at this I/O
3. Android 17 Watch: Our and potential Big Rocks to tackle for some apps
4. Android 17 Goodies (Opportunities): UI, Performance, AI items, and more

The Gemini Models

Since 2022, AI has remained the main theme of Google I/O, with a focus on its Gemini model. Gemini is believed to be the most mentioned keyword again, as shown by the bento-style images presented at this year’s event to showcase its AI products.

While this article focuses on Android, it would not do Google I/O justice without a brief overview of the new Gemini model.

To aid comprehension, below is a table summarizing the model, including past prominent models and the one officially shared during I/O.

Existing Gemini models

• Gemini Nano — An Android local model for offline use, free of charge.
• Veo — A video generation model used within Omni (details to be shared later).
• Nano Banana — An image generation model used within Omni (details to be shared later).
• Gemini 3.1 Pro — The top Gemini model for logic reasoning before Google I/O 26.

New models in Google I/O 26

• Gemma 4 — An open model family executable locally for free. Will be as Gemini Nano 4.
• Gemini 3.5 Flash — The next-generation Gemini model for the free tier.
• Gemini Omni — A multi-model multimedia blend of Veo, Nano Banana, and Lyria.
• Gemini Spark — An individual agent like Open Claw, available only in the US.
• Gemini 3.5 Pro — The most powerful model, releasing later after Google I/O 26

The Android Developer Ecosystem

More ways to develop Android App.

Traditionally, Android apps are developed using Android Studio. However, at Google I/O 26, Google introduced a new method to enable AI developers and Vibe coders to build for Android OS.

1. Google AI Studio, traditionally for AI app development on Google Cloud, now offers a feature to convert AI apps into Android apps. To manage projects locally, you can transfer your Google AI Studio project to Antigravity as shared below. Reviewer feels this might only apply to simple app.
2. Antigravity is an AI-powered, agent-first IDE by Google for Agentic and Vibe coding. It now supports Android app development, which is more feasible using Android CLI as described below. For finer control, you can still transfer projects to Android Studio.
3. Android CLI, launched in early 2026, assists Agentic Coding in generating Android apps. Unlike conventional web development, Android apps require extensive setup; Android CLI simplifies this process. It accesses Android Skills and Knowledge Bases to support agentic coding and can perform automated Android UI tests to aid the coding flow.

Android CLI access to Android Skill

Making Android Studio more agentic code friendly

With agentic coding trending, Android Studio now supports it better and leverages Agent Skills. Other enhancements in Android Studio Panda include:

1. Uploading app to PlayStore. Previously available only via CLI, this feature is now accessible in the UI to enable non-developers to complete the end-to-end flow in Android Studio.
2. Access to OEM devices for testing. Beyond the Pixel emulator, Android Studio now supports testing on 3rd party OEM devices like Samsung.
3. Migrate from iOS or React Native projects. An early pre-announced feature enabling migration from iOS and React Native apps to Android. This will be complex to implement.
4. Migrate Android projects to Kotlin Multi-Platform. An early preview allowing Android apps to migrate to Kotlin Multi-Platform, enabling deployment on iOS.

The Android 17 Watch Outs

If there’s one area Android developers should look out for in Android 14, it should be, what’s the pitfall our app will have face in our migration to Android 17?

Here are the six most relevant to our apps, with the first four as top priorities.

1. Compose First. Google now treats Compose as the primary framework, placing traditional Android View and XML in maintenance mode. They may deprecate it next year and retire it the year after.
2. Ignored APIs and attributes in tablet. Moving into Android XR, many tablet-specific attributes that restrict layout resizing will be ignored. If unaddressed, our app will look wonky on tablets. The team should start adopting compose Adaptive layout for our Apps, enabling future Android XR support.
3. Navigation 3 Compose. Our apps, developed before Jetpack Compose became standard, use Fragment-based navigation. With Compose First and Android XR, Navigation 3 now works seamlessly for Android XR.
4. Memory Limits. Android 17 introduces stricter memory controls, especially for apps with leaks. Google I/O stated they may terminate apps with memory issues, showing a “memory limiter” message.

The below two are less concern, and should be looked into too.

1. Static final fields. To optimize Android 17 with the R8 optimizer, it disallows modifying static final fields, even via reflection. We discourage reflection use, as its future reliability is uncertain. Any reflection usage, including in third-party libraries, will pose issues in Android 17.
2. Stricter control of ClearText traffic. Android has long disallowed ClearText traffic in production. For internal debugging, we sometimes allow it using usesCleartextTraffic=”true”. In Android 17, explicit Network Security Config is required to permit it.

The Android 17 Goodies

Besides the major updates, Android 17 offers goodies where some of them, possibly we can leverage for our Apps. They are categorized into Compose related, performance, AI, and others.

Compose Related

1. Skill for View to Compose. Google is now maintaining Android View and provides Agent Skill to help us transition from this legacy view to Compose.
2. Compose 1.10/11 Improvements. Google is making Compose a first-class citizen by addressing its limitations in versions 1.10 and 1.11, including various animation features like transition animation.
3. Style Component Customisation. Google introduced a “Styles” API to further customize Compose’s basic components, enhancing their flexibility.
4. Media Query. This lets code query device state — keyboard presence, screen size, pointer position — to adjust layout.
5. FlexBox. Inspired by web CSS, it provides a proven, resizable layout mechanism.
6. Grid. Supports better real estate use on larger Android XR devices.

4–6 is to prepare Compose for Android XR’s flexible layout management, we enhanced it with these APIs.

Performance Related

1. Compose Performance Improved. As part of making Compose the primary framework, Compose performance has improved, reducing screen jankiness.
2. Better Garbage Collection. To improve memory management, Android 17 will include better internal garbage collection.
3. Performance Analyser. To help users better understand how their apps are performing, an enhanced performance analyser is now available.
4. Better Compile Optimisation. Improved R8 out-of-the-box optimisation allows apps to gain better performance in terms of app size and startup time.
5. R8 Configuration Analyser. To give users better control over how R8 optimises code, Google provides a configurable analyser.
6. WebGPU. This technology is taken from the web and applied to native Android UI development, improving lower-level UI performance more directly.

AI Related

1. Gemini Nano. Google expanded Gemini Nano to 140M devices and will boost its performance with Gemma 4.
2. Auto Filling In. Google Android can now autofill user information in forms.
3. Connecting with Other Apps. Google aims to link apps for a seamless user experience, e.g., a photo app working with Gmail during agentic flow.
4. App Function. Similar to iOS AppIntent, this exposes our app’s internal functions to Google’s Gemini, acting as our MCP on Android. It lets Gemini interact with our app and expose content to users.
5. A2UI Compose Renderer. This agent-to-UI feature lets the app dynamically generate layouts based on AI inputs, enabling custom UIs (e.g., forms) tailored to user interactions.

Misc…

1. 3D Emoji Icon. To appeal to the newer generation, emoji is essential, making communication more expressive than words.
2. Bubbles. Like a widget on the Android homescreen, it lets users access our app faster than searching the icon. It existed in older Android versions but was removed. It returns to enable more agentic, accessible app use.
3. Accessibility Scanner. It scans app accessibility to identify gaps for improvement.
4. Image Description Talkback. Previously, each image needed a description for Talkback to assist visually impaired users. This feature provides better insight even without descriptions, useful for dynamic images.
5. Eye Dropper. To get the color of a pixel, apps used to require full image access. Now, they no longer need broad permission just for the little detail.
6. Contact Picker. To access a single contact, apps used to request full contact access. Now, they no longer require broad permission for that.
7. Picture in CameraX ViewFinder. While using the camera, users can flip to view app content, like chatting, without leaving the camera screen.

It’s not just the phone…

Often, when we think of Android, we think of Samsung and other OEM devices, direct counterparts to iPhone and iPad. But the Android ecosystem extends far beyond phones and tablets.

At Google I/O 26, they introduced GoogleBook, likely an extension of ChromeBook and PixelBook, showing Google’s continued interest in desktop OS. Android XR supports glasses experiences with audible voice, augmented, and immersive features. Android also powers Automotive, Wearables, and large screens like TVs, making it one of the world’s largest OS platforms.

Making our app available on Android opens vast opportunities to reach many users beyond just mobile. With Agentic AI, Android XR, and accelerated Android enhancements driven by AI, we are given much more to explore in this landscape.