Over the past two decades, Google has brought a list of links people choose from algorithms listed on the web for any given search query. In I/O 2025, Google made it clear that the concept of search is firmly in its rearview mirror.

On Tuesday, Google CEO Sundar Pichai and his executives proposed a new way to bring users to the network, this time through a series of AI agents.

"We're very excited about this chapter of Google Search, and you can really ask any (…) your easiest, toughest questions, deepest research, your personalized shopping needs," said Liz Reid, Google's vice president of search on I/O. "We think AI will be the most powerful engine ever on the web."

The biggest announcement for I/O is that Google now provides an AI model for every search user in the United States. This provides hundreds of millions of people with a button to talk to an AI agent that will visit web pages, sum them up in any way, and even help them shop. With Project Mariner, Google is offering more manual AI proxy to its Ultra subscribers. The agent will handle 10 different tasks simultaneously, visit web pages and click on those pages, while the user can insert other content completely.

Google is also making its in-depth research agent that visits dozens of related websites and generates thorough research reports, is more personalized, and connects them to your Gmail and Drive. In parallel development, the company is integrating Project Astra (the company's multi-mode, real-time AI experience) into search and Gemini, giving users more ways to talk verbally with AI agents and let it see what they see.

I can go on, but you get it - AI agents dominate I/O 2025.

The rise of Chatgpt forced AI to estimate on Google, which led the company to rethink how to get user information from the network. This estimate does start with I/O last year, when Google introduced AI overview to search, which is masked by its embarrassing hallucinations. The launch of AI overview makes it seem like AI isn't ready for prime time, and we know it will stay here.

But in I/O 2025, Google has proposed a more compelling and fulfilling approach to how AI will reshape search and the web. The company's new vision shows that the future of the network and the future of the company involves AI agents taking information from the network and presenting it to the user in whatever way they want.

The idea that Google's AI proxy could replace search is a compelling search, especially because Google is trying to build infrastructure for AI proxy. Google announced Tuesday that the Gemini model's SDK will now support human MCP locally, an increasingly popular standard for connecting agents to data sources across the internet.

Google is not alone in this transformation. At a different technology conference this week, Microsoft CTO Kevin Scott proposed his vision for an "open proxy network" where agent representatives take action on behalf of users on the Internet. Scott noted that the key function of achieving this is to connect these agents to each other's pipelines and data sources - namely Google's Agent2agent protocol and numerous MCPs.

As Ben Thompson pointed out in Stratechery, despite the enthusiasm, there are still problems with the proxy network. For example, Thompson pointed out that if Google sends AI proxy to a website instead of people, this largely breaks the Internet's ad-supported model.

The impact of the impact may vary. Agents may not be an issue for companies that sell goods or services on the internet, such as Doordash or Ticketmaster - in fact, these companies are using agents as new platforms to attract customers. But, for publishers, it cannot be said that these publishers are now competing for attention with AI agents.

During I/O, a Google Communications head told me that “human attention is the only truly limited resource” and that the company’s AI agent aims to get users back more time. This may all be phased out, but the article's AI summary may seem likely to take the dollar from publishers and potentially undermine the content creation on which these AI systems rely.

Furthermore, there is a lingering problem surrounding illusions that they tend to make up and present them as facts – and it becomes embarrassingly clear about this as Google launches AI overview. DeepMind CEO Demis Hassabis spoke on the stage, even raising concerns about the consistency of AI models.

"You can easily find some obvious flaws (AI chatbots) in a few minutes - some high school math stuff that can't be solved, some basic games it can't play," Hassabis said. "It's not difficult to find those holes in the system. To me, to call AGI, it will be more consistent across the stage."

The consequences can be far-reaching. Extensive hallucinations may make users more distrustful of the information they encounter on the web. They may also seed misinformation between users. Neither result is ideal.

Google doesn't seem to be waiting for ad-supported businesses or AI models to catch up - the company is pushing AI agents anyway. Google may do more than any other company to housekeepers, as we know. However, this could be a major turning point, and the company's concept of the network appears to be repositioning around AI agents.