Google I/O 2026 Opens May 19 — What to Expect from Gemini 4 and the Future of Search

Updated after the event with what Google actually announced. We previewed this keynote expecting a headline act called Gemini 4. What Google delivered on 19 May was different, and in truth more consequential: not one flagship model, but an attempt to make its AI act on your behalf, everywhere, all the time. Here is what actually launched, and why Europe should read it closely.

Google I/O 2026: What Google Actually Delivered - Two New Models: Omni and 3.5 Flash

Google did not unveil a "Gemini 4." It unveiled two models pointing in a sharper direction. Gemini Omni is the ambitious one — a system built to create from any input, beginning with video, and pitched as a leap in multimodal understanding and editing.

Gemini 3.5 Flash is the workhorse: fast, cheap to run, and explicitly built for agentic, long-horizon tasks, which Google says outperforms its previous Pro model on demanding coding and agent benchmarks. The naming matters less than the intent. Both models were designed less to answer questions and more to take actions. That is the thread running through everything Google showed.

Spark: The Agent That Works While Your Phone Is Off

The announcement that deserves the most attention, and got the least scrutiny, is Spark. Running on Gemini 3.5 and Google's new Antigravity platform, Spark is presented as an autonomous, around-the-clock personal agent that navigates your digital life and acts for you.

The detail worth pausing on is Google's own claim that it works in the background on your phone or laptop even while they are turned off. Its roadmap includes custom sub-agents and authorised budgeting for payments — an AI that can spend your money on your behalf.

For now it is gated to trusted testers and then US subscribers on the premium AI Ultra tier. But the direction is unmistakable, and it is the same one we later saw in the Pixel 11's screen-reading assistant: the move from an AI you consult to an AI that quietly runs errands across your life.

The End of the Search Box as We Knew It

Google also rebuilt the thing it is most famous for. The search box received what the company calls its biggest upgrade in over 25 years — now reasoning across text, images, files, videos and even open Chrome tabs at once, with Gemini 3.5 Flash installed as the default engine of AI Mode globally.

More striking still is generative UI: rather than returning a list of links, Search can now assemble a custom answer on the fly — tables, interactive visuals, even simulations built in real time. This is the quiet endgame of the AI Overviews story. The link, the thing the open web was built on, is being demoted from the destination to raw material for an answer Google composes itself.

The European Perspective

For Europe, I/O 2026 was not a product showcase; it was a preview of three collisions. The first is data protection. An agent that runs while your device is off, reads across your digital life, and is authorised to make payments is a GDPR question of the first order — and, predictably, the most autonomous features launched in the US first, with Europe left to wait, as it so often is, precisely because its rules force harder answers.

The second is the future of the European web. When Search stops sending clicks and starts composing its own answers from everyone else's work, the traffic that funds European publishers, retailers and creators thins out — and there is still no settled model for compensating the sites whose content trains and feeds these answers. The third is dependency.

Every announcement deepened the same reality: Gemini is becoming the invisible layer beneath search, phones, shopping and work, and almost none of that layer is built, owned or governed in Europe. The lesson of I/O 2026 was not that Google built impressive AI. It was that Google is quietly becoming the operating system of everyday digital life — and Europe's role, once again, is to write rules for a machine it does not build.

We are not first. We are right.