Google used I/O 2025 to show a version of Search that looks far less like the familiar page of blue links and far more like an AI-mediated service. The company’s message was clear: instead of only pointing people to the web, Google wants AI agents to go into the web on their behalf.
That shift is not a small interface change. It suggests a new role for Google Search, Gemini and related AI systems: finding information, visiting pages, summarizing material, helping with shopping and producing research outputs before a user ever needs to click through.
Search moves from links to agents
For roughly the last two decades, Google Search has been built around returning algorithmically chosen links for a query. At I/O 2025, Google presented a different direction, with executives describing Search as something users can converse with and delegate to.
Google CEO Sundar Pichai and other executives introduced several products and integrations built around AI agents. Google’s VP of Search, Liz Reid, described the company’s ambition onstage, saying users could bring Search their simplest and hardest questions, deep research needs and personalized shopping needs.
“We believe AI will be the most powerful engine for discovery that the web has ever seen.”
The biggest announcement was that Google now offers AI mode to every Search user in the United States. That gives hundreds of millions of people access to a button for interacting with an AI agent that can visit web pages, summarize them in different ways and help with shopping.
Google is also bringing Project Mariner to Ultra subscribers. That agent is designed to be more hands-off, handling 10 different tasks simultaneously while visiting web pages and clicking around on them as the user works on something else.
The agent layer is spreading across Google
The announcements were not limited to one Search feature. Google also said its Deep Research agent, which visits dozens of relevant websites and generates thorough research reports, is becoming more personalized. The company is connecting it to Gmail and Drive.
Project Astra, Google’s multimodal, real-time AI experience, is also being further integrated into Search and Gemini. The point is to let users speak with an AI agent and allow it to see what they see.
Taken together, these moves show Google trying to place AI agents between users and more online tasks. In this model, the user does not always begin by scanning results, opening tabs and evaluating pages directly. The agent can do a growing share of that work first.
The company is also working on infrastructure for this shift. Google announced that the SDK for Gemini models will now natively support Anthropic’s MCP, a standard for connecting agents to data sources across the internet.
Google is not moving alone
The broader technology industry is also pushing toward an agentic web. At a different tech conference in the same week, Microsoft CTO Kevin Scott described a vision for an “open agentic web,” where agents act for users across the internet.
Scott pointed to the importance of the technical plumbing that lets agents connect with each other and with data sources. He specifically cited Google’s Agent2Agent protocol and Anthropic’s MCP.
That matters because AI agents need more than polished chat interfaces. To be useful across the web, they need ways to retrieve information, act on pages and connect with services. Google’s I/O 2025 announcements showed the company trying to build both consumer-facing agent features and the underlying systems that could support them.
Publishers face a harder tradeoff
The agentic web also creates pressure for the businesses that supply the content these systems use. As Ben Thompson notes in Stratechery, if Google sends AI agents to websites instead of people, it can disrupt the ad-supported model of the internet.
The impact may not be the same for every type of company. Businesses that sell goods or services online, such as DoorDash or Ticketmaster, may see agents as a new platform for reaching customers. The source notes that these companies are embracing agents.
Publishers are in a more difficult position. If AI agents summarize articles for users, fewer people may need to visit the original page. That could reduce the attention and advertising value publishers depend on, even as AI systems continue to rely on published content.
During I/O, a Google communications leader told the source that “human attention is the only truly finite resource.” The company’s argument is that agents can give users more time back. But that same time-saving feature may shift dollars away from publishers and weaken the content creation that AI systems depend on.
Trust remains an unresolved problem
Google’s new agent push comes after a difficult earlier rollout. At last year’s I/O, Google introduced AI overviews into Search, but that launch was overshadowed by embarrassing hallucinations.
Hallucinations remain a central risk for AI systems. The source defines the issue as the tendency of these systems to make stuff up and present it as fact. If agents become a major way people access the web, that problem becomes more consequential.
DeepMind CEO Demis Hassabis raised concerns about the consistency of AI models onstage Tuesday. He said users can quickly find obvious flaws in AI chatbots, including a high school math problem they do not solve or a basic game they cannot play.
“It’s not very difficult to find those holes in the system. For me, for something to be called AGI, it would need to be much more consistent across the board.”
If hallucinations spread through agentic search experiences, users may become more distrustful of information they encounter online. They may also be exposed to misinformation. Neither outcome fits neatly with Google’s claim that AI will become a powerful engine for discovery.
Google does not appear to be waiting for every business model or every AI reliability issue to be solved before moving forward. The company has played a major role in shaping the web people know today. At I/O 2025, it showed that its next version of the web is increasingly organized around AI agents.