Google I/O 2026 Preview: What to Expect on May 19 — Gemini 4, Android 17, and the AI-Everything Pivot

Google I/O 2026 is seventeen days away, and the signals the company has been sending in the weeks leading up to it — through earnings calls, leaked job listings, developer blog posts, and the careful breadcrumb trail of confirmed session titles — paint a picture of a developer conference that is less about product launches and more about a company trying to answer a single question: is Google still the most important platform for building AI-powered software?

The event runs May 19-20 at the Shoreline Amphitheatre in Mountain View, California, with keynotes and sessions livestreamed on io.google. The Google developer keynote — the one aimed at engineers and builders rather than the general press audience — kicks off at 1:30 PM PT on May 19.

The Pichai Tease

The strongest signal about the tone of I/O 2026 came from Sundar Pichai’s remarks during Alphabet’s Q1 2026 earnings call. After discussing the company’s cloud compute constraints and the $725 billion in aggregate capex that hyperscalers are committing to AI infrastructure this year, Pichai said Google was “excited to share more about Search at I/O” and highlighted the company’s focus on “pushing the next frontiers of foundation models, including intelligence, agents, and agentic coding.”

In a separate blog post, Pichai disclosed a data point that speaks to how thoroughly AI has penetrated Google’s own engineering culture: AI now generates 75% of all new code written at the company. That figure positions I/O as not just a showcase of AI products but as a company reporting back from the frontier of actually living with these systems at scale.

Gemini 4: Expected but Not Confirmed

The most-watched potential announcement at I/O 2026 is Gemini 4. Google has not officially confirmed a Gemini 4 reveal for the event, but the timing lines up: the company tends to use I/O as its premier model announcement vehicle, and the current Gemini 3.1 Pro — released in February with a 1-million token context window and strong reasoning capabilities — has had several months to establish itself in the market. A next-generation model reveal with “concrete details rather than benchmark slides” is what AI analysts are hoping for.

What would Gemini 4 need to demonstrate? At minimum, meaningful progress on the dimensions that have kept the Gemini 3 series from fully eclipsing OpenAI’s and Anthropic’s flagship offerings: multi-step agent reliability, fewer hallucinations on complex reasoning chains, and more consistent tool-use performance in production environments. A demonstration of Gemini Nano 4 — a version small enough to run on-device on Pixel flagships — would also signal a push into the on-device AI space that Apple has been aggressively cultivating with its own neural engine roadmap.

Android 17: The Adaptive Everything Platform

Android 17 is essentially confirmed for I/O. Google pre-seeded developer interest with a preview of session topics, including one titled “Adaptive development for the expanding Android ecosystem” — language that frames Android 17 as completing a transition to what Google internally calls “Adaptive Everywhere.”

The vision is ambitious: a single Android platform that spans phones, automotive displays (Android Automotive), large-screen tablets, television (Google TV), and immersive environments (Android XR). Sessions confirm that Android 17 will bring performance improvements, new camera and media capabilities, expanded support for desktop and large-screen apps, and deeper integration of agentic automation — AI functionality that can take multi-step actions on behalf of users without requiring them to explicitly authorize each step.

For developers, the most significant change may be the formalization of what Google has been calling “predictive back” and deep links into a coherent app-to-agent handoff model: when a Gemini model or third-party AI agent wants to accomplish a task inside an app, Android 17 provides the standardized plumbing for that interaction. This is how Google plans to make Android the default substrate for the agentic AI applications it is betting will define the next era of mobile computing.

Aluminium OS: The ChromeOS Endgame

The most potentially disruptive announcement at I/O 2026 may not be a model or a mobile OS update — it may be the formal reveal of Aluminium OS, Google’s ground-up rethinking of what a desktop operating system looks like when you build it from scratch with AI at its center.

Leaked job listings, infrastructure code, and developer documentation have revealed a project internally codenamed ALOS: an Android-based operating system designed specifically for laptops and desktops. Crucially, this is not ChromeOS with an Android skin applied on top — it’s a purpose-built desktop-class platform engineered to take on Microsoft Windows and Apple macOS directly. The project has been described in Google’s own leaked materials as “Android-based” and “built with Artificial Intelligence at the core.”

What makes Aluminium OS more than a branding exercise is the structural shift it represents. ChromeOS has been a solid but limited platform — powerful for web apps and light workloads, weak for creative professionals and enterprise software that requires native application support. An Android-based successor could, in principle, run the enormous catalog of Android apps natively while extending Google’s cloud and AI services into every interaction at the OS level. Gemini would not be an app you open on an Aluminium OS machine — it would be ambient infrastructure built into the window manager, file system, and clipboard.

Whether Google formally announces Aluminium OS at I/O or keeps it to a roadmap tease will be one of the most watched moments of the conference.

Project Astra: The Universal Assistant, Live This Time

Project Astra, Google’s multi-modal universal AI assistant that can see through a phone camera, hear ambient audio, and maintain persistent context across interactions, was demonstrated at I/O 2025 in a form that impressed researchers but wasn’t publicly available. The expectation at I/O 2026 is different: an Astra demo that shows the technology working on a real task in a live environment, not a carefully choreographed lab demonstration.

The session schedule confirms that Google will show “persistent context across a real task” — the demo challenge that makes AI assistants feel genuinely useful rather than impressive in limited scenarios. If Astra can reliably pick up a thread from a conversation that happened hours earlier, associate it with a file the user was editing, and take actions in apps to advance the task — that is the moment when “AI assistant” becomes something people actually change their workflows around.

Project Astra is also the product that connects most directly to Google’s business model: if Astra becomes the primary interface through which users interact with Google Search, Gmail, Maps, and Drive, Google recaptures the ambient interface position that it has been ceding to ChatGPT and Claude.

The Firebase and Developer Stack Overhaul

For developers building applications — the core I/O audience — the most consequential set of announcements may come from changes to Google’s developer infrastructure. Session copy describes Firebase as evolving into an “agent-native platform,” with an end-to-end path from AI prototyping through to production deployment on Google Cloud.

The specific tools mentioned in confirmed sessions: AI Studio (Google’s low-code AI development environment), a new tool called Antigravity for building full-stack applications, and deep integration between Android Studio, Gemini, and Firebase for AI-native Android development. The stated developer keynote theme of “agentic coding” — AI tools that handle routine development tasks, freeing engineers to focus on architecture and product strategy — is the developer-facing articulation of the same agentic shift that consumers will see in Astra and Android 17.

Android XR and the Hardware Wild Card

Android XR — Google’s mixed reality platform developed in partnership with Samsung and Qualcomm — is expected to appear at I/O, potentially with a hardware reveal. The platform has been developing relatively quietly since its announcement; I/O 2026 is the natural venue for Google to show developers what building for Android XR looks like in practice and to give the first concrete sense of what a consumer Android XR headset experience might be.

Google has, notably, kept Android XR off the official session list as of this writing. That absence could mean it’s too early, or it could mean Google is saving it for a surprise keynote moment — the kind of “one more thing” that elevates a developer conference into a cultural moment.

What Google Needs to Prove

Google enters I/O 2026 in a position that is simultaneously stronger and more precarious than it has occupied in years. Stronger, because Gemini is a genuine frontier model, because Android remains the world’s dominant mobile operating system, and because Google’s cloud business is growing strongly. More precarious, because OpenAI has GPT-5.5, Anthropic has Claude Opus 4.7, the Musk-Altman trial is consuming AI news cycles, and the implicit question hanging over every Google AI demo is whether the company’s search business will be cannibalized by the very AI products it is building.

What Sundar Pichai needs to communicate on May 19 is not just that Google’s models are good — the benchmarks already support that. What he needs to communicate is that Google’s ecosystem — the integration of Gemini across Android, Search, Cloud, and Workspace, the developer toolchain, the on-device inference story — gives it a structural advantage that pure AI labs cannot replicate. That is the argument that I/O, more than any other Google venue, is designed to make.

Seventeen days to go.