Two years ago, Google was the loser of the AI race. The company with the biggest data pile on Earth had been outpaced by a chatbot built on a fraction of the same training material. Last week, at the I/O developer conference, Google showed what that data pile actually buys you the day you decide to use it.
Three announcements stood out. One will reach you almost immediately. The other two will not.
The Search bar already changed
You can now drag a file into the Google search bar and ask a question about it. Drop in a quote your supplier sent and ask whether the price is reasonable. Drop in a video and ask what was said in minute three. The bar remembers the conversation, so you don’t have to reformulate the same question 4 times.
This is the boring announcement, and also the one that matters. Over a billion people already use Google’s AI features in Search every month. The smartest Gemini model is now sitting in that bar by default, worldwide. The flashier news got the applause. This one quietly rewires the first tool most people open every morning.
The trade is the obvious one. The more Google knows about what you search, the better the results, and the more data you hand over. Sundar Pichai said it out loud in an interview: “If users use our AI-driven features, they search more.” Useful for you. Slightly more useful for Google. An analyst at Arete Research called the open web (meaning the internet outside Google) “on its way down” now that AI keeps traffic on the search results page rather than sending it elsewhere.
That is the part worth thinking about, even if you do not change a thing about how you use Search.
Gemini Spark: the assistant that keeps working
Spark is Google’s agent. Pichai gave it three tasks live on stage, put the phone down, and said: “Now you can just close your laptop. Spark keeps going.”
Spark is plugged into your Gmail, Docs and Calendar. You can ask it to pull deadlines out of your inbox, summarise the meetings you missed, or prepare the briefing for tomorrow’s 10am. Anthropic’s Cowork and OpenAI’s Agent Mode already do similar things. The difference is that Google does not have to integrate with anything. It is already inside your account.
Caveats. Spark is US-only for now and available only to Google AI Ultra subscribers, at $100 to $200 a month. It works well if you live inside the Google ecosystem and considerably less well if your work happens in Outlook or Apple Mail. The next step is to let Spark send emails and make purchases on your behalf, which means handing an agent the keys to your inbox, calendar, documents, and payment details. Google says it will always ask first. You will want to believe them.
Gemini Omni: AI video that knows how things move
The third announcement was Omni, a video model. Describe a scene, upload a frame, and Omni produces a clip. Change the background, add an object, swap the style, all in plain language.
Sora and a handful of others already do this. What Omni is supposed to be better at is physics. Gravity, motion, things bumping into other things. The small cues that betray earlier AI video as AI video. The clips Google showed are convincing.
Omni is available worldwide to paying Gemini subscribers, from about €22 a month. The business API, the connection that lets a company build Omni into its own software, is not out yet. For a marketing team that wants a quick visual for a deck or a short animation to explain something to a colleague, this is the most immediately useful of the three. For everyone else, it is still a demo.
What this actually means
Google has been holding the same chip for twenty-five years. Your search history, your inbox, your calendar. It never quite knew what to play it for. Now it does. The Search change reaches you this week. Spark and Omni are mostly a promise for now, and a US promise at that.
The interesting question is not whether Google is back. It is back. The interesting question is what the open web looks like when fewer people leave the search page.
