For over two decades, Google Search has relied on the familiar model of ranking blue links based on complex algorithms. But at I/O 2025, Google signaled a clear departure from this long-standing approach.
Highlights
The company is now embracing AI agents—smart, autonomous tools designed to interact with the web on your behalf. These agents don’t just help you find information; they find it, summarize it, and even complete tasks for you.
This isn’t just a new feature—it’s a fundamental shift in how the web works.
From Search Engine to Digital Assistant
During the keynote, Google described this transformation as the next chapter in its mission. Liz Reid, VP of Search, explained that users can now “truly ask anything,” whether they need detailed research or personalized product recommendations.
The focus is moving away from delivering links and toward providing actionable answers, directly and conversationally.
The rollout has already begun. AI Mode in Google Search is now available to all users in the U.S. This feature introduces a conversational interface where the AI can read and summarize pages, answer complex queries, and even act as a shopping assistant.
If you’re subscribed to Gemini Advanced, you get access to something even more powerful—Project Mariner.
Project Mariner
Project Mariner is a Chrome extension powered by Gemini 2.0. It’s designed to carry out web-based tasks on your behalf.
Whether it’s clicking through links, filling out forms, or extracting information from websites, Mariner can handle it. It was tested using the WebVoyager benchmark and achieved an impressive 83.5% success rate.
Imagine planning a dinner party. Mariner could look up a recipe, visit a grocery store’s site, and add all the ingredients to your cart—while you’re busy doing something else. It’s not just search assistance—it’s task automation at scale.
Project Astra
Underpinning many of these new features is Project Astra, Google’s multimodal AI system. Astra understands voice, visuals, and context.
That means you can point your phone camera at an object and ask questions about it—or speak naturally and get real-time help.
With enhancements from Gemini 2.0, Astra now supports multilingual input, better accent recognition, and faster, more contextual conversations. It can also remember what you said for up to 10 minutes, making your interactions feel more continuous and human-like.
New Internet Infrastructure
One of the more technical but critical pieces of this evolution is the Agent2Agent (A2A) protocol. Think of it as a common language that lets different AI systems talk to each other.
Google introduced this open standard to help AI agents—regardless of who built them—collaborate on tasks and share information securely.
This step suggests that we’re not just heading toward smarter assistants, but toward an entirely new model of the web: one driven by AI agents that work together behind the scenes.
Even Microsoft is pushing in the same direction. At its own developer event, CTO Kevin Scott highlighted the importance of agent interoperability—where AI tools can connect across services, platforms, and use cases.
It’s clear this isn’t just about smarter tools; it’s about creating the operating system for the AI-powered web.
Not Everyone Is on Board
But this shift has raised serious questions—especially from content creators and publishers. Analysts like Ben Thompson from Stratechery warn that as AI agents deliver answers directly, users may skip the source websites entirely.
Some reports already show a drop of nearly 30% in clickthrough rates from AI-assisted search. That’s a big deal for ad-supported websites that rely on traffic for revenue.
While Google says that competition from new AI tools could balance out market power concerns, the economic impact on publishers and content-based platforms remains unclear.
And Then There’s the Trust Factor
Even with all these advancements, there’s still the issue of reliability. AI agents—despite getting better—can still hallucinate, or generate incorrect information.
DeepMind CEO Demis Hassabis admitted that these systems sometimes struggle with basic reasoning. So while the technology is impressive, it isn’t flawless.
Until these problems are consistently addressed, trust in AI-generated answers will be a major sticking point for both users and developers.
