Underneath all the AI noise, Google I/O 2026 shipped Android 17, Wear OS 7, and the end of ChromeOS

Last year Google put a counter on the keynote stage at I/O. Every time an exec said “AI,” it ticked up, and by the closing minutes it had landed on 92, down from the 121 it hit in 2024. This year there was no counter, no slide, no self-deprecating tally. Pichai opened by announcing Google processes 3.2 quadrillion tokens a month, paused, said “never imagined I’d say quadrillion in an IO keynote, but here we are,” and moved on. Somewhere in the room a joke quietly retired itself, because the joke had become the room. The most interesting parts of this year’s I/O were the parts you had to look past Gemini to see, and the absence of a counter was the cleanest signal yet that Google has stopped pretending the AI saturation is novel enough to count.

Android 17, internally Cinnamon Bun, ships in June

Pichai called it Gemini Intelligence on stage. The build, internally, is still Cinnamon Bun, the Android 17 codename Google has used through API level 37, the last vestige of a dessert-naming tradition that was retired publicly with Android 10 but kept alive in the canary tree. The Android 17 release date is June for Pixel, with broader OEM rollouts following through the summer. What the Android 17 feature set actually delivers is less an AI revolution and more a real design release with three Gemini features bolted on, and the design release is the part you will notice every day.

Material 3 Expressive is the headline. Frosted-glass surfaces ripple across system menus and quick settings, Gaussian-blurred so you can read the content underneath. Animations through Jetpack Compose now run on spring physics rather than linear curves, the kind of detail that sounds like nothing in a blog post and feels like everything by the tenth swipe home. The phone looks alive in a way Pixel UI has not since 12.

Beta 3 brought the rest. Priority Charging pauses background sync while your phone is plugged in, so the wattage reaches the battery instead of refreshing your Instagram cache. Vendor-defined camera extensions let OEMs expose proprietary modes (Samsung’s Super Resolution, OnePlus’s Hasselblad pipeline) to any app via isExtensionSupported(int), the kind of structural change that, given a year, makes Instagram’s camera stop being objectively worse than the system one. On the security side, Find Hub now requires biometrics for Mark-as-lost, and the theft protections that piloted in Brazil are default-on globally for new and freshly reset devices.

The Gemini Intelligence layer sits on top of all that. Create My Widget generates custom widgets from natural-language prompts, Rambler scrubs filler words out of voice input, and Pause Point inserts a ten-second delay before time-sink apps open, asking whether you actually meant to. They are real features, and they are also entirely optional. The marquee “Gemini doing multi-step tasks across your apps” demo, by contrast, is the same demo Google has shown for three I/Os running, and anyone who lived through Magic Cue on the Pixel 10, where Google promised proactive AI and most owners barely noticed it existed, will adjust expectations accordingly. The hardware bar does not help either: 12 GB of RAM, a current-generation flagship chip, Gemini Nano v3 support. Most of the Pixel 9 family does not make the cut, and neither do the Galaxy Z Fold 7 or the OnePlus 13.

Worth pinning a footnote here. Pixel 6 Android 17 support is confirmed: the Pixel 6, 6 Pro, and 6a all get the update, and that is their last major OS release. Google extended their support window by two years, so the runway ends with Cinnamon Bun. Five years from launch is not the worst outcome an enthusiast has ever bought into.

Wear OS 7 is finally interesting again

The watch platform has been quietly the most overdue release Google ships, and Wear OS 7 fixes most of the right things. Battery improves roughly 10 percent against Wear OS 6 across the board. Live Updates inherits from Android 16, so the notification feed on your wrist now changes in real time during a ride share or a delivery. Wear Widgets replace the static tile model with something dynamic and customisable, closer to what the phone has had for two years.

The developer payoff is the AppFunctions API, which lets a Wear app expose discrete functions to Gemini so that voice commands map to actual app actions without each developer hand-rolling intent extras. Saying “start tracking my run” calls Samsung Health directly. The plumbing is unglamorous and load-bearing in roughly equal measure. Canary builds are available now, with a public release pegged for “later this year,” which almost certainly means alongside the Pixel Watch 5 in August.

ChromeOS is now Android, and there is hardware to prove it

The biggest structural news of the week is Aluminium, Google’s internal codename for a new Android 17-based desktop OS that replaces ChromeOS and ships preinstalled on a new laptop class Google is calling Googlebook. On stage, the consumer OS name went unmentioned, so “Aluminium” stays a dev label for now. Retail devices from Acer, ASUS, Dell, HP, and Lenovo are targeting Q3, which puts the first Googlebooks in shops by July or August.

After a decade of pretending two desktop operating systems was a coherent strategy, Google’s implicit answer is no, and the substantive consequence falls on Android developers. Every Android app is now also a desktop app, by default, on the laptop class Google itself is brand-stamping. Once that ships as the retail default in the fall, the developer reader has to start thinking about windowing, keyboard focus, and trackpad gestures, not whether to bother with tablet mode as an afterthought.

Android XR finally has ship dates

For two years, the Android XR glasses story has been “we are working on it.” This year it has a calendar. Samsung’s first Android XR glasses, designed with Gentle Monster and Warby Parker, are shipping in Fall 2026. XREAL’s Project Aura, also 2026, will be the first AR glasses globally to ship on Android XR. The eyewear partnerships are the part to pay attention to. Google has learned from Glass that you cannot sell hardware people will not put on their face; the Gentle Monster collab is built for “disruptive yet refined aesthetics,” which is brand-speak for “looks like sunglasses, not a Black Mirror prop.”

The feature set is predictable: Gemini voice queries, real-time audio translation in the speaker’s voice, summarised notifications, calendar additions. None of it breaks new ground against what Meta has been demoing on the Ray-Ban Display side, but the novelty was never going to come from the feature list; it was always going to come from shipping. If Samsung lands the fall window cleanly and the price is not absurd, this is the year smart glasses become a category instead of a tech-demo genre.

For Android developers, Gemini Nano is the real announcement

Gemini Nano is now available to third-party Android apps through the new Android AI Core ModelManager API, which means inference at zero per-token cost, with no network round trip and no customer data leaving the device. For healthcare, legal, finance, and journalism apps that cannot ship sensitive data to a cloud model, the compliance story changes overnight. This is the structural change of the week on the Android dev side, and most of what came after it in the developer keynote is incremental against it.

The rest is incremental but still real. Antigravity 2.0 ships with a desktop app, a CLI, an SDK, and subagents that run scheduled tasks in the background. Over on the consumer-facing side, Google AI Studio gets a mobile app that lets you vibe-code Android apps from your phone and pick the project up at your desk later. Android Studio adds Agent Mode and an iOS-to-Kotlin porting tool, and Google open-sourced its “Android skills” repo so Claude Code, Codex, and any other agent can build Android apps to spec. None of this will make you write less code if you ship Android software for a living, but it might make the boring parts shorter.

Gemini 3.5, Omni, and Spark: the AI side of I/O 2026

Gemini 3.5 Flash ships today in the Gemini app, Search, and the API, and on Google’s numbers it beats 3.1 Pro on coding and agentic benchmarks at 289 tokens per second; 3.5 Pro follows next month. Alongside it, Omni Flash, the unified multimodal model that outputs video grounded in real-world physics, lands for AI Plus subscribers with chat-based editing and a ten-second clip ceiling. Then there is Spark, the 24/7 cloud agent built on the Antigravity harness, which runs on dedicated Google Cloud VMs, takes work over its own Gmail address, and goes into beta next week for US AI Ultra subscribers. Daily Brief, a morning digest stitched from your inbox and calendar, ships to US AI subscribers today.

Pichai also disclosed that Google processed 3.2 quadrillion tokens through its models in April, up from 480 trillion a year ago. “Never imagined I’d say quadrillion in an IO keynote,” he said, “but here we are.”

Spark is the one to watch. A 24/7 agent with email access and hooks into Canva, OpenTable, and Instacart is either the most useful product Google has shipped in five years or the next Bard, and the early reads from US testers will tell us which. Project Mariner, Google’s previous attempt at the same idea, was shut down on May 4, and its tech was absorbed into Google Agent and the Gemini API. That is the thing about Google’s AI products: they are sometimes the same product twice.

What did not happen at I/O 2026

No Pixel 11 appeared, and no Tensor G6 either. Google looks to be reverting to the pre-2023 cadence of leaving hardware for August, which the Pixel buyer reading this should treat as good news; the November rollout has historically been the bumpy one. Veo 4 also stayed in the wings, with Omni Flash now the closest analog and the Veo line presumably folding into whatever Omni becomes from here. NotebookLM’s new Cinematic Video Overviews, meanwhile, are an AI Ultra exclusive, which says something about who Google now thinks NotebookLM is for.

They retired the counter. The rest is the work

Google did not put a counter on the stage this year, and the test of whether that silence means anything is what is left when the keynote glow fades: whether Android 17 still feels like itself when it stabilises in June, whether the first Googlebook boots cleanly out of the box in October, whether the Samsung glasses turn up looking like glasses. If those land, Google’s loudest I/O in years will quietly turn out to have been its most consequential, and the missing counter will be the most accurate slide Google never showed.

FAQ