For the past 20 years, Google has provided people with a list of links selected from the web to algorithms for specific search queries. At I/O 2025, Google revealed that the concept of search is firmly in the rearview mirror.
On Tuesday, Google CEO Sundar Pichai and his executives presented a new way to bring users to the web. This time, we used intermediaries through a series of AI agents.
“I couldn’t be more excited about 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 Liz Reid, Google’s vice president of search, on the I/O stage.
The biggest announcement of I/O was that Google would provide AI mode to all search users in the US. This allows hundreds of millions of people to have a button to talk to an AI agent who can help you access web pages, summarise them as you like, and shop. With Project Mariner, Google offers even more handoff AI agents to Ultra subscribers. That agent handles 10 different tasks at the same time, accessing web pages and clicking on those pages, but the user is completely free to connect to something else.
Google also has visited numerous related websites, generated in-depth research reports, and created deep search agents that are more personalized and connected to Gmail and Drive. In parallel development, the company further integrates Project ASTRA (the company’s multimodal, real-time AI experience) into search and Gemini, providing more ways for users to talk to AI agents verbally and show what they are seeing.
You can go on, but you can get ideas. AI agents ruled I/O 2025.
The rise of ChatGPT forced AI calculations on Google, rethinking how users provide information from the web. This calculation actually began with last year’s I/O when Google introduced AI overviews into searches. With the deployment of AI overview, it seemed that AI wasn’t ready for prime time.
However, in I/O 2025, Google presented a more convincing, embodied approach to how AI reconstructs searches and, therefore, how to reconstruct the web. The company’s new vision suggests that the future of the web and the company involves AI agents obtaining information from the web and presenting it to users in any way they like.
The idea that Google’s AI agents can replace searches is compelling, especially as Google is trying to build an infrastructure for AI agents. Google announced on Tuesday that the Gemini model SDK will natively support Anthropic’s MCP, an increasingly popular standard for connecting agents to data sources on the Internet.
Google is not alone in this shift. At another high-tech conference this week, Microsoft CTO Kevin Scott sets his own vision for the “open agent web” that moves on behalf of users across the internet. Scott’s key features to enable this are the plumbing that connect these agents to each other, and the MCP of Google’s Agent2Agent protocol and Anthropic.
Despite the enthusiasm, as Ben Thompson points out in Stratechery, there is a problem with the Agent Web. For example, Thompson points out that if Google sends AI agents to a website on behalf of a person, it primarily breaks the Internet’s ad support model.
The impact may vary from industry to industry. Agents may not be a problem for companies that sell products and services online, such as Doordash and Ticketmaster. In fact, these companies are embracing agents as new platforms to reach customers. However, the same cannot be said about publishers currently fighting AI agents for their eyes.
During I/O, Google Communications leaders told me, “human attention is the only truly finite resource.” And launching the company’s AI agent aims to give users more time. It may all pan out, but the AI summary of the article seems likely to take dollars from publishers. And it could destroy the very content creation that these AI systems rely on.
Furthermore, there are prolonged issues with AI systems regarding hallucinations – the tendency to make them and present them as facts – became embarrassingly revealed in the launch of Google’s AI overview. Speaking on stage Tuesday, Deepmind CEO Demis Hassabis even raised concerns about the consistency of the AI model.
“You can easily find some obvious defects within minutes [AI chatbots] – There are high school maths that don’t solve, there are basic games that you can’t play,” Hassavis said. For me, it needs to be much more consistent across the board for what’s called AGI. ”
The results can be extensive. Wide range of hallucinations can lead users to feel more distrustful of information they encounter on the web. They may also spill misinformation among users. Neither result is ideal.
It appears Google isn’t waiting for ad-supported companies and AI models to catch up. The company is promoting AI agents anyway. Google, as we know, has probably done more to steward more companies than any other company. However, in what can prove to be a major turning point, it appears that the company’s web concept is reorienting around AI agents.
Source link
