Over the past two decades, Google has brought people a list of links selected by web algorithm for a given research request. At E / S 2025, Google clearly indicated that the research concept is firmly in its rear view mirror.
On Tuesday, the CEO of Google Sundar Pichai and its leaders presented new ways of bringing users to the web, this time intermediate via a series of AI agents.
“We could not be more excited by this chapter of Google Search where you can really ask anything […] Your simplest and most difficult questions, your deepest research, your personalized shopping needs, “said Google’s vice-president of research, Liz Reid, on stage at E / S.” We believe that AI will be the most powerful engine for the discovery that the web has ever seen. “”
The biggest E / S announcement was that Google now offers AI mode to each research user in the United States. This gives hundreds of millions of people a button to converse with an AI agent who will visit web pages, summarize them in the way they would like, or even help them buy. With Project Mariner, Google provides an even more prohibited AI agent for its ultra subscribers. This agent will simultaneously manage 10 different tasks, visiting web pages and clicking on these pages while users are free to connect to something else.
Google also manufactures its in -depth research agent, which visits dozens of relevant websites and generates in -depth, more personalized search reports and connects it to your Gmail and your reader. In a parallel development, the company integrated more Project Astra – the multimodal experience of AI in real time of the company – in research and Gemini, giving users more means to speak verbally with an AI agent and to let it see what they see.
I could continue, but you have the idea – IA agents have dominated the 2025 E / S.
The rise of Chatgpt forced a calculation of the AI at Google, which forced the company to rethink the way it provides information on the web of users. This calculation really started in the E / S of last year, when Google introduced the previews of the AI in research, a launch which was overshadowed by its embarrassing hallucinations. The deployment of IA’s overviews gave the impression that AI was not ready for prime hours, and this research as we know to stay.
But at E / S 2025, Google presented a more fascinating and expanding approach to the way IA would reshape research, and therefore the web. The new vision of the company suggests that the future of the web, and the company, implies that AI agents are to recover information on the web and present it to users in the way they would like it.
The idea that Google AI agents could replace research is convincing, especially because Google tries to put an infrastructure for AI agents. Google announced Tuesday that the SDK for Gemini models was now going natively support the anthropic MCP, An increasingly popular standard for the connection of agents to data sources on the Internet.
Google is not alone in this change. During another technological conference this week, the CTO Microsoft Kevin Scott exposed its own vision for an “web open on the web”, in which agents take measures on behalf of the users on the Internet. Scott noted that a key characteristic to make this possible would be the plumbing that connects these agents and the data sources – namely the agent2agent protocol of Google and the MCP of Anthropic.
Despite the enthusiasm, as Ben Thompson Note in Stratechery, The agency web has its problems. For example, Thompson notes that if Google sends AI agents to websites instead of people, this largely breaks the Internet model supported by advertising.
The impacts could vary from industry to industry. Agents may not be a problem for companies that sell goods or internet services, such as Doordash or Ticketmaster – in fact, these companies adopt agents as a new platform to reach customers. However, the same thing cannot be said for publishers, who are now fighting with AI agents for eyeballs.
During the E / S, a leader in Google communications told me that “human attention is the only really finished resource”, and the launch by the company of AI agents aims to make users more time. All this can take place, but the summaries of AI articles seem likely to keep the dollars away from publishers – and potentially devastate the creation of content even on which these AI systems depend.
In addition, there is a persistent problem with AI systems around hallucinations – their tendency to invent things and to present it as a fact – which has become embarrassing clear with the launch of Google of the IA previews. Speaking on Tuesday on stage, the CEO of Deepmind Demis Hassabis even expressed his concerns concerning the coherence of AI models.
“You can easily, in a few minutes, find obvious faults with [AI chatbots] – Something mathematics in high school that it does not solve, a basic game that he cannot play, “said Hassabis.” It is not very difficult to find these holes in the system. For me, for something to be called AG, it would be necessary to be much more coherent in all areas. »»
The consequences could be a large range. The widespread hallucinations could lead users to be more suspicious of the information they encounter on the web. They could also sow disinformation between users. One or the other result is not ideal.
Google does not seem to wait for companies to advertise or IA models are caught up – the company goes ahead with AI agents. Google has probably done more than any other company to manage the web as we know it. But in what could prove to be a major turning point, the design of the company on the web seems to redirect around AI agents.
