Google’s I/O updates turn AI into actionable daily tools

Google’s I/O – At its I/O developer conference on Tuesday, Google laid out how it intends to bring AI to consumers and businesses: faster Gemini models powered by its in-house AI chip infrastructure, personal “agents” that keep working in the background, and an “AI Mode” tha

On Tuesday, Google used its I/O developer conference to do something that’s harder than it sounds in the AI race: make the leap from impressive demos to tools people will actually use.

The company rolled out a set of new and updated products that ranged from personal AI agents and coding tools to search upgrades—and it paired those announcements with a blunt message about where the advantage comes from. Google’s approach leans heavily on the infrastructure built through search. the data it has accumulated over years. and the computing muscle it is already investing in.

At the center of the update were new DeepMind models and a bigger push toward what Google calls “world models” for generating video that stays physically accurate.

New models with speed, lower costs, and “tool use”

Google said its Gemini 3.5 Flash model powers many of the new products and features announced at I/O. The model is described as optimized for speed and efficiency. “four times faster than other frontier models. ” and priced so that it costs between one-half and one-third the price of comparable models. DeepMind’s prior best model. Gemini 3.1 Pro. is said to have been outperformed by 3.5 Flash “on nearly all benchmarks. ” with particular strength in coding and tool use.

Google also announced a Gemini 3.5 Pro model that will become DeepMind’s new flagship. Researchers are still studying its safety implications, and Google plans to release it publicly sometime in June.

During a call with reporters on Monday, Alphabet CEO Sundar Pichai said, “All our focus with the 3.5 series has been on taking the model intelligence and making sure tool use, instruction following, long-horizon use cases, and agent decoding all work well.”

The other major model track was Google’s push into “world models”—systems designed to create digital environments or video that remain true to real-world physical properties. Google’s entry is called Gemini Omni. It is multimodal. meaning it can generate outputs including video. images. text. audio. and more. based on prompts that include content in those same formats.

Google offered a concrete example: a user can provide an image of herself along with a video. and Gemini Omni uses high-level reasoning so the user’s likeness can stand in as a character in the video. Google is launching a small version of Omni, called Omni Flash, today. A larger Omni Pro model is currently in development.

Behind the scenes, Google pointed to infrastructure—and the economics

Before it spent much time on what the models can do, Google talked about the machine underneath them.

Google said it expects to spend up to $190 billion on new infrastructure this year. Much of that will go toward new data centers where Gemini models run on “hundreds of thousands” of Google’s own AI chips.

The company is now on its eighth generation of tensor processing units (TPUs). the chips designed for the billions of mathematical computations required by neural networks. As AI labs scale up computing resources. Google argued that the power and cost efficiency of the chips used to serve models increasingly affect the economics of running AI.

Google said training large AI models is no longer limited to a single data center. It can be distributed across “more than 1 million TPUs” globally, creating what Google describes as the world’s largest training cluster.

It also suggested a potential advantage in training data, with web crawling and knowledge infrastructure. Google said it very likely has the world’s most advanced web crawler—technology that continually scours and indexes web pages so they can be searched. Researchers train large AI models on massive amounts of this web content. and Google said the volume. quality. and composition of training data can directly impact a model’s intelligence.

Google’s crawlers may reach more web pages and content than those used by other AI labs. and Google captures much of this content in a “knowledge graph. ” which can serve information about people. places. organizations. products. events. and concepts. Google also said it has the full corpus of YouTube videos available for AI training. and that this content was very likely used to train the new Omni world model to understand how objects relate and move in the real world.

The human bargain—and the faith required

There’s a tightrope in Google’s pitch: AI is built on public trust, but much of the public doesn’t see the underlying process.

The company’s announcements land in a world where AI labs ask the public to take a lot on faith—faith that information will be kept secure; that companies will spend responsibly on AI safety; that technology won’t be used for harmful purposes such as autonomous weapons or mass surveillance; and that new data centers won’t spike energy prices or further tax the environment.

There’s also faith that the benefits of AI will be broadly distributed, and that the business itself will eventually generate enough market demand and revenue to survive.

Google isn’t described as perfect in the reporting of this strategy, but its pragmatic approach is presented as something that makes those promises feel credible—“and that there are, in fact, adults in the room.”

A consumer-first push, with personal agents that keep running

The dominant narrative around AI infrastructure has often focused on large enterprises—data centers built to power AI-infused business processes at scale. That’s why Google’s I/O emphasis stood out: a large share of the new work was aimed at consumers, plus the models and apps businesses can use.

Alphabet CEO Sundar Pichai said during Monday’s briefing that Google is trying to bring as much frontier intelligence to consumers as possible.

DeepMind’s Tulsee Doshi. senior director and product lead of generative AI and Gemini models. put it in terms of Google’s identity. In an interview with Fast Company. Doshi said. “As someone who grew up using Google search. I think Google’s whole ethos has been to organize the world’s information and make it universally accessible and useful. ” adding. “And now in the agentic era. you can add ‘help users take action on that information in a way that is thoughtful and intentional.’”.

Doshi acknowledged that a significant portion of the return on Google’s massive capital expenditure investment in data centers is likely to come from enterprise business.

One of Google’s sharpest moves into that “agentic” era was a personal AI agent called Gemini Spark. Google says Spark runs on Gemini 3.5 Flash and stays active in the background even when a user’s devices are off.

Google frames Spark’s main advantage as quick personalization. By connecting to Gmail. Docs. Slides. and other widely used Google Workspace tools. it can learn users’ interests. preferences. and work habits. The company says Spark can handle complex tasks such as drafting status updates from multiple documents or planning block parties.

It also supports multistep workflows—for example, parsing credit card statements; monitoring a Gmail inbox for time-sensitive information; or turning meeting notes into polished documents.

Spark also includes connectors to third-party tools including Canva, OpenTable, and Instacart. Google says more capabilities are coming this summer: the ability for users to text or email Spark directly, create custom subagents, and let Spark control a local browser.

Google also said users control which apps Spark can access, and the agent is designed to ask for confirmation before taking high-stakes actions like sending emails or spending money. Google says Spark will soon come to its Gemini mobile app so users can manage agents from anywhere.

Search and AI are becoming one layer

Google’s I/O storyline also returned to its core product: search.

At the beginning of the generative AI boom. many feared AI search would undercut Google’s search advertising business—its cash cow. Google’s classic model placed ads next to ranked search results. the familiar “10 blue links. ” and the industry wrestled with how advertising would work around AI-generated answers.

Google now argues the opposite: that radically improving search with AI encourages users to search more often, expanding advertising opportunities.

Google said users conducted more searches during the first quarter of the year than in any previous quarter. likely because of the conversational. multiple-query nature of AI search. It said “AI Mode” queries have been doubling every quarter. and that more than a billion people now use the tool each month.

The company previously began using large language models to interpret the intent behind searches. After ChatGPT arrived, Google introduced “AI Overviews” for some searches, packaging results into AI-generated summaries designed to answer user questions. Then came “AI Mode,” an advancement on the same idea.

Google’s position now is that AI is best understood as a permanent layer sitting on top of all Google search functionality.

It also said it does not need to invent an entirely new ad system. AI has been folded into the existing search advertising machine: Google still shows traditional search ads above and below AI-generated responses, and its existing ad auctions continue to function.

Among the search upgrades coming soon is “Ask YouTube.” Google says users will be able to talk to videos and ask questions about their content. building on the idea that users can already search for videos on a topic and sift through them for answers. Google also said YouTube may return custom search results that combine several videos with instructions or steps for completing a task.

On a web-wide level, Google says its goal is for its AI to analyze the world’s information, reason over it, and answer questions about it.

During the press briefing. Liz Reid. Google Search chief. said. “We’ve successfully combined the best of the search engine with the best of AI so that we can build a true AI search experience that brings together our most advanced Gemini models. our newest agent capabilities. and the full breadth of the world’s information.”.

The search improvements Google announced are powered by Gemini 3.5 Flash.

Google also said it is changing its legacy search box so it dynamically expands to accommodate longer and more detailed queries. In the coming months. users will be able to deploy “background agents” that continually monitor specific information on the web. and can even build personalized. persistent tools such as fitness trackers.

What’s still riding on search ads

For all the focus on agents and models, Google’s AI ambitions ultimately rest on the health of its core search advertising business.

Unlike some peers that rely heavily on revenue from AI model APIs or subscriptions, Google does not rely solely on AI model monetization to keep its core business running. The reporting characterizes AI as additive to search and as something Google can also sell through its thriving cloud business.

Even if Wall Street looks at the AI boom through a bubble lens—especially given the enormous capital expenditures associated with it—Google’s diversified business is framed as insulation against growing fears that the current AI spend won’t translate into durable demand and revenue.

Google I/O Gemini 3.5 Flash Gemini 3.5 Pro DeepMind Gemini Omni Omni Flash Omni Pro world model Gemini Spark AI search AI Mode AI Overviews Ask YouTube tensor processing units TPUs data centers Sundar Pichai Liz Reid search advertising Workspace Canva OpenTable Instacart