Google AI Agents Set to Redefine Web Interaction

For over two decades, Google has delivered algorithmically curated web links in response to search queries. At I/O 2025, the company signaled a definitive shift away from traditional Search as we know it.

During Tuesday's keynote, Google CEO Sundar Pichai and executive team unveiled new approaches for delivering web content—this time facilitated entirely through AI agents.

"We're incredibly excited about this new chapter for Google Search, where you can ask anything—from simple questions to complex research and personalized shopping needs," announced Google's VP of Search Liz Reid on the I/O stage. "We believe AI will become the most powerful discovery engine the web has ever seen."

The flagship announcement revealed Google now offers AI mode to all U.S. Search users, giving hundreds of millions access to conversational AI agents that browse websites, generate customized summaries, and assist with shopping. For Ultra subscribers, Project Mariner delivers an even more autonomous agent capable of handling ten simultaneous tasks—navigating web pages and interacting with content while users focus elsewhere.

Google is also enhancing its Deep Research agent, which scours dozens of relevant sites to produce comprehensive reports, with deeper personalization and integration to Gmail and Drive. Simultaneously, Project Astra—the company's multimodal, real-time AI interface—is being woven into Search and Gemini, enabling more natural voice interactions and visual understanding.

The message was unmistakable: AI agents dominated I/O 2025.

The emergence of ChatGPT triggered Google's AI awakening, compelling the company to reimagine how users access web information. This transformation began at last year's I/O with the troubled rollout of AI overviews, whose inaccurate hallucinations suggested AI wasn't yet ready for mainstream search.

Join us at TechCrunch Sessions: AI

Secure your spot at our premier AI industry gathering featuring speakers from OpenAI, Anthropic, and Cohere. Limited-time offer: full-day access to expert talks, workshops, and valuable networking for just $292.

Exhibit at TechCrunch Sessions: AI

Showcase your innovations to 1,200+ decision-makers at TC Sessions: AI. Limited exhibitor tables available through May 9 or until sold out.

At I/O 2025, Google presented a more sophisticated vision for AI-driven Search transformation. The company's new direction suggests a future where AI agents retrieve web information and present it according to user preferences.

The potential for AI agents to replace traditional Search is particularly compelling given Google's infrastructure developments. The company announced Gemini models' SDK will now natively support Anthropic's MCP—an emerging standard for connecting agents to online data sources.

Google isn't alone in this transition. At another tech conference this week, Microsoft CTO Kevin Scott outlined his vision for an "open agentic web" where agents perform actions across the internet on users' behalf. Scott emphasized the importance of connectivity protocols like Google's Agent2Agent and Anthropic's MCP for enabling this ecosystem.

Despite the enthusiasm, Stratechery's Ben Thompson highlights inherent challenges. For instance, if AI agents replace human website visitors, the internet's advertising-supported model could collapse.

The effects would vary across sectors. While e-commerce platforms like DoorDash and Ticketmaster are embracing agents as new customer channels, publishers face direct competition for audience attention.

A Google communications lead noted during I/O that "human attention remains our only finite resource," suggesting AI agents could give users more time. However, AI-generated article summaries may redirect revenue from publishers—potentially undermining the content creation that fuels these AI systems.

Hallucinations remain another persistent issue, as demonstrated by Google's problematic AI overviews launch. DeepMind CEO Demis Hassabis even acknowledged consistency concerns during his presentation: "Within minutes, you can identify obvious flaws—unsolved math problems, basic games they can't master. For true AGI, we'd need far greater consistency."

The implications are significant. Widespread hallucinations could erode trust in online information and accelerate misinformation spread—neither outcome being desirable.

Google appears unwilling to wait for either advertising models or AI systems to mature—pushing forward with agent deployment regardless. While no company has done more to shape the modern web, Google now appears to be reorienting the web's very conception around AI agents.