Android I/O 2026 Part II: Gemini 3.5, XR Glasses, Android 17, and the Agentic Dev Era

This is Part II of our Google I/O 2026 coverage. If you missed it, Part I covers The Android Show: I/O Edition — Gemini Intelligence, AppFunctions, Googlebook hardware, Jetpack Compose 1.11, and more.

The Google I/O 2026 keynote landed on May 19 at the Shoreline Amphitheatre, and Google delivered on virtually every promise made at The Android Show the week before — then went further. Where Part 1 was about the Android platform evolving into an intelligence system, the keynote was about the infrastructure that makes that intelligence real: new models, a persistent personal agent, agentic developer tooling, and the first public hardware reveal of Android XR glasses.

For developers, the signal is clear: AI-assisted development is no longer optional. Google has structured its entire toolchain — from the Gemini API to Android Studio to a brand-new CLI — around the assumption that agents will be doing a significant portion of the heavy lifting in your workflow.

Here is everything that matters.

Gemini 3.5 Flash: The New Developer Default

The biggest model announcement was Gemini 3.5 Flash — Google's new workhorse model that they describe as their strongest agentic and coding model to date. The headline claims: 4x faster than competing frontier models, lower cost (roughly half the price per token compared to alternatives), and benchmark improvements across coding, reasoning, and multimodal tasks.

Google is processing over three trillion tokens per day with this model internally — including using it to co-develop Antigravity 2.0 itself. That scale gives you a sense of the confidence Google has in its reliability.

For Android developers specifically, Gemini 3.5 Flash is the engine powering the new agentic dev tools (covered below), AI Studio's Android vibe coding, and Managed Agents in the Gemini API. If you're integrating Gemini into your own backend — whether via the Gemini API directly or through Firebase AI Logic — this is the model you'll want to evaluate first.

Independent benchmark confirmation is still pending at time of writing, but if Google's claims hold up, the cost and speed profile makes this very compelling for any team running AI-heavy workloads.

Gemini Omni: Conversational Video Editing

Also announced: Gemini Omni, a new multimodal model family starting with Omni Flash — designed for native video creation and editing through conversation.

The demo showed uploading a photo or video clip and describing changes in plain language: lighting adjustments, scene replacements, style transforms. This goes beyond text-to-video generation into genuine interactive editing. Mixed inputs — text, images, video, audio — are handled natively.

Omni Flash is available today for Google AI Plus, Pro, and Ultra subscribers. It's also accessible via YouTube Shorts and YouTube Create for those building content-creation features. If you're building media-related apps on Android, this capability in the Gemini API is worth evaluating for in-app editing workflows.

Gemini Spark: The Persistent Personal Agent

The biggest consumer-facing announcement was Gemini Spark — a personal AI agent that runs continuously in the cloud, independent of whether your device is active.

Spark can monitor Gmail, Calendar, Docs, and other Workspace services. It tracks ongoing tasks, monitors for information you've asked it to watch (flight prices, document changes, meeting prep), and proactively surfaces updates. This is fundamentally different from a reactive chat interface — Spark operates as an always-on background process that takes action on your behalf.

From a developer perspective: Spark represents where AppFunctions (announced in Part 1) will eventually plug in. As Spark expands to third-party apps, the developers who have registered for the AppFunctions Early Access Program will be best positioned to offer their app's capabilities to Spark's task automation layer.

Spark is included in the new AI Ultra subscription at $100/month, which also bundles all Gemini models, 30TB Google One storage, Gemini in Workspace, and priority processing.

Android XR Glasses: The Hardware Reveal

Google and Samsung took the stage together to unveil Intelligent Eyewear — lightweight smart glasses running Android XR with Gemini natively on-device. This was the hardware preview Google teased at the end of The Android Show.

Unlike bulky mixed-reality headsets, these are designed to look like regular prescription glasses. They include a built-in camera for visual AI input, a display for surfacing information, and Gemini processing integrated at the hardware level.

Use cases demonstrated: real-time language translation by looking at text, contextual information about landmarks, and step-by-step task guidance through the display. Samsung is handling the hardware manufacturing; Google provides the Android XR OS and Gemini AI layer.

Target launch: fall 2026. The device ships with Android XR Developer Preview 4 APIs already available (released during I/O week as promised in the Android Show announcement).

Android XR Ecosystem Partners

Beyond Samsung, Google confirmed partnerships with multiple hardware makers for the Android XR platform:

XREAL (Project Aura) — wired XR glasses with display, targeting the more immersive end of the spectrum
Gentle Monster and Warby Parker — screenless AI glasses competing more directly with Meta Ray-Ban glasses, handling audio-only interactions with Gemini

This multi-partner approach mirrors how Google handled Android on phones — the OS is Google's, the hardware is a broad ecosystem. For developers, this means one SDK (developer.android.com/develop/xr) targets all partners.

What to build now: Android XR Developer Preview 4 added Title Chips, Button Groups optimized for touchpad input, the ProjectedTestRule API for automated testing, and Jetpack Compose Glimmer for glanceable display UIs. The XR SDK's Projected APIs bridge phone app experiences into a user's field of view without requiring separate XR-specific code paths.

Android 17 "Cinnamon Bun": Stable in June 2026

Google gave a timeline: Android 17 stable release is targeted for June 2026. The OS (internally codenamed Cinnamon Bun) was previewed with deeper Gemini integration at the system level — building on the AppFunctions, Gemini Intelligence, and platform changes announced at The Android Show.

One UI 9 from Samsung is based on Android 17, and with Samsung's smart glasses products arriving this year, the connection between Android 17 and XR hardware is tight. If you're targeting the XR glasses launch this fall, Android 17 is your baseline.

Key things confirmed for Android 17:

Deeper Gemini system-level integration (AppFunctions, task automation)
Creator features: optimized Instagram experience, Adobe Premiere support
Chrome + Gemini (webpage summaries, AI image generation, contextual event help)
UI changes from Android 16 continuing into Android 17 (redesigned notification panel, quick settings, lock screen)

Android 17 also lays the groundwork for Aluminium OS — Google's longer-horizon vision for merging Android and ChromeOS into a unified platform for Googlebooks and beyond.

The Agentic Developer Stack: What's New

This is where I/O 2026 got most interesting from a pure developer tooling perspective. Google shipped a significant set of new tools targeting the reality that many developers now use AI agents rather than manually writing every line of code.

Android CLI 1.0 — Now Stable

Android CLI is now stable at version 1.0 and ready for all Android developers. This is a programmatic interface that lets AI agents — any agent, not just Google's own — interact directly with Android Studio capabilities.

With Android CLI, an agent can:

Download the Android SDK
Run your app on physical or virtual Android devices
Execute Gradle builds
Navigate project structures
Trigger UI tests

The key framing from Google: Android CLI works with Claude Code, OpenAI Codex, Antigravity, and any other agent of your choice. Google is not trying to lock you into their toolchain — they're making Android development accessible from whatever AI coding environment you prefer.

To update if you're already using it: android update

Open-Sourced Android Skills

Alongside the CLI, Google open-sourced a set of Android Skills — structured instructions that help LLMs understand how to correctly execute complex Android-specific workflows. Current skills cover:

Migrating code to Jetpack Compose
Migrating to Jetpack Navigation 3
Best practices for complex Jetpack API usage

These are essentially high-quality, curated prompting frameworks that encode Google's engineering best practices in a format any LLM can consume. If you're rolling your own AI-assisted dev setup, these are worth pulling into your agent's context.

Android Bench: LLM Leaderboard for Android Tasks

Google launched Android Bench — a leaderboard that evaluates LLMs specifically on Android development tasks. At launch, it includes open-weight models like Gemma 4 alongside proprietary models, so you can compare performance on the actual tasks that matter: code generation, API usage, migration complexity, and so on.

This is useful for teams evaluating which model to use in their CI/CD pipelines or AI coding integrations. It's also Google's mechanism for incentivizing model providers to improve on Android-specific capabilities.

Android Studio Migration Agent

Google previewed a Migration Agent inside Android Studio that can port apps from iOS, React Native, or web frameworks to native Android Kotlin. The agent analyzes your existing codebase, maps features to Android equivalents, converts assets (storyboards, SVGs), and outputs Jetpack Compose-based native code using recommended Jetpack libraries.

Google's claim: migrations that previously took weeks can now complete in hours. This is currently in preview — not GA — but worth tracking if you have any cross-platform or legacy codebases you've been hesitant to migrate.

Google AI Studio: Kotlin Support + One-Click Deploy

Google AI Studio now supports native Kotlin for vibe-coding Android apps. You can describe an app, have it generated in Kotlin, and deploy it directly to Cloud Run with one click — or push it to Firebase services. When you're ready to take it further, you can export the complete project state to Google Antigravity and continue iterating locally.

AI Studio also received a mobile app, so you can prototype on the go.

Antigravity 2.0: Google's Agentic Coding Platform

Google Antigravity got a major 2.0 upgrade, establishing it as a direct competitor to GitHub Copilot, Cursor, and Claude Code.

What's New in Antigravity 2.0

New standalone desktop app — multi-agent orchestration in a single UI; you can run parallel subagents handling different parts of a codebase simultaneously
Background task scheduling — set agents to run automatically without manual intervention
Antigravity CLI — terminal-based agent control for developers who prefer the command line. Note: this replaces the previous Gemini CLI. If you're using Gemini CLI today, migration is required
Antigravity SDK — programmatic control over the Antigravity agent harness; deploy it on your own infrastructure with custom agent configurations
Voice commands — native voice input support, similar to Gmail and Docs integrations
Android bundle — install the Android CLI and skills directly within Antigravity during onboarding or via settings

Antigravity 2.0 is powered by Gemini 3.5 Flash. Google used it to co-develop the model itself — a nice demonstration of the flywheel they're building.

Managed Agents in the Gemini API

For developers who don't want to manage their own agent infrastructure, Managed Agents in the Gemini API provides a hosted execution environment. A single API call provisions a fully functional agent with a remote sandbox, built-in tooling access, and persistent state — no infrastructure to configure.

This is Google's answer to the friction of building agentic systems from scratch. If you're already using the Gemini API in your backend (including via Firebase AI Logic), Managed Agents is worth evaluating as a way to expose agentic capabilities without standing up your own orchestration layer.

Web Platform: WebMCP, Modern Web Guidance, and HTML-in-Canvas

While these are web-focused announcements, they're worth noting for Android developers who maintain web apps or PWAs alongside their native apps.

WebMCP is a proposed open web standard that lets developers expose structured tools (JavaScript functions, HTML forms) so browser-based AI agents can execute tasks with reliability and precision. An experimental origin trial starts in Chrome 149. If you're building any kind of web tool that agents should be able to interact with, this is the API to watch.

Modern Web Guidance provides expert-vetted skills for building performant, accessible, secure web experiences — covering over 100 use cases. It integrates with the Baseline browser compatibility signal. Install it in Antigravity or via npx modern-web-guidance install.

HTML-in-Canvas (origin trial) lets you embed real DOM elements inside a WebGL/WebGPU canvas, enabling accessible and searchable 3D web experiences. Primarily relevant for game or immersive web app developers.

AI Content Verification: SynthID + C2PA

Google announced that SynthID (its AI-generated content watermarking technology) and the C2PA standard are coming to Google Search and Chrome. OpenAI has joined as a verification partner, meaning AI-generated content from multiple platforms will carry verifiable metadata that Search can surface.

For developers: if you're generating images or video in your Android apps using Gemini's generative capabilities, being aware of SynthID's watermarking behavior is important — especially for apps in media, news, or content creation verticals.

Key Takeaways for Android Developers

Android CLI 1.0 is stable — integrate it into your agent-based dev workflow now, regardless of which AI coding tool you prefer
Open-source Android Skills are available — use them to improve your LLM's Android-specific accuracy
Android Bench gives you an objective way to evaluate LLMs for Android tasks before committing to one
Migration Agent in Android Studio (preview) can accelerate iOS or React Native to Android Kotlin migrations
Antigravity CLI replaces Gemini CLI — if you're using Gemini CLI in any automation scripts or workflows, plan your migration
Android XR DP4 is live — if you're targeting the fall 2026 Samsung or XREAL glasses launch, now is the time to start building
Android 17 lands in June — start testing your apps against the beta now, especially if you're targeting XR or Googlebook form factors
AppFunctions EAP remains the most important thing to register for — Gemini Spark's expansion into third-party apps is the next phase, and early access gives you a head start

What Comes Next

With Android 17 stabilizing in June and the first Googlebooks and XR glasses shipping in fall 2026, the second half of this year is where all the platform bets placed at I/O begin to pay off — or not. The developer tooling story (CLI, Bench, migration agent, Antigravity 2.0) is the most mature it has ever been, and the agentic shift in how Android apps are discovered and used through AppFunctions is real.

We'll continue covering the rollout of these features here as they hit production. Stay tuned.