Google AI Agents: Bringing the Web to You

The Evolution of Google Search: From Links to AI Agents

For the past twenty years, Google has provided users with a ranked list of web links in response to search queries. However, during I/O 2025, Google signaled a definitive shift, indicating that the traditional concept of Search is now largely in the past.

A New Approach to Web Access

Google’s CEO, Sundar Pichai, alongside other executives, unveiled innovative methods for accessing web content, now facilitated by a network of AI agents.

Liz Reid, Google’s VP of Search, expressed excitement about this new era, stating that users will be able to pose any question, conduct in-depth research, and fulfill personalized shopping needs. She believes AI will become the most potent tool for web discovery ever created.

AI Mode and Project Mariner

A significant announcement from I/O was the availability of AI mode to all Search users in the United States. This feature empowers hundreds of millions of individuals to engage in conversations with an AI agent capable of visiting websites, providing customized summaries, and assisting with purchases.

Furthermore, Google is introducing Project Mariner, an even more autonomous AI agent for Ultra subscribers. This agent can manage up to ten tasks concurrently, navigating and interacting with web pages while users focus on other activities.

Deep Research and Project Astra Integration

Google is enhancing its Deep Research agent, which compiles comprehensive research reports from numerous websites, by making it more personalized. It is also integrating this agent with Gmail and Drive for seamless access to user data.

In a related development, Project Astra – Google’s multimodal, real-time AI experience – is being integrated into both Search and Gemini, offering users additional ways to interact with AI agents through voice commands and visual input.

The Impact of ChatGPT and the AI Reckoning

The emergence of ChatGPT prompted a critical reassessment of AI within Google, leading the company to reconsider its approach to delivering web information. This reevaluation began at last year’s I/O, with the introduction of AI overviews, a launch marred by instances of inaccurate information.

The initial rollout of AI overviews suggested that AI wasn’t fully prepared for widespread use, and that the conventional Search experience would remain dominant.

A More Developed Vision for AI-Powered Search

However, at I/O 2025, Google presented a more refined and comprehensive strategy for integrating AI into Search, and consequently, the web. The company’s new vision centers on AI agents retrieving information from the web and presenting it to users in a customized manner.

Infrastructure for AI Agents

Google is actively building the necessary infrastructure for these AI agents. The company announced that the SDK for Gemini models will now natively support Anthropic’s MCP, a growing standard for connecting agents to internet data sources.

Microsoft’s Parallel Vision

Microsoft is also pursuing a similar path. At a separate tech conference, Microsoft CTO Kevin Scott outlined his vision for an “open agentic web,” where agents act on behalf of users across the internet. Scott emphasized the importance of connecting agents to each other and to data sources, highlighting Google’s Agent2Agent protocol and Anthropic’s MCP.

Challenges and Potential Disruptions

Despite the optimism, Ben Thompson of Stratechery points out potential issues with the agentic web. A key concern is that directing AI agents to websites instead of human users could disrupt the advertising-based revenue model of the internet.

The impact will likely vary by industry. Companies offering goods or services online, like DoorDash or Ticketmaster, may benefit from agents as a new customer reach platform. However, publishers could face challenges as agents compete for user attention.

The Finite Resource of Human Attention

A Google communications leader stated during I/O that “human attention is the only truly finite resource,” and that the launch of AI agents aims to return time to users. While this may be true, AI-generated summaries of articles could reduce revenue for publishers and potentially jeopardize the content creation that fuels these AI systems.

The Persistent Problem of Hallucinations

AI systems continue to struggle with “hallucinations” – generating inaccurate information presented as fact – a problem highlighted by the initial rollout of AI overviews. Even DeepMind CEO Demis Hassabis acknowledged concerns about the consistency of AI models.

Hassabis noted that it’s relatively easy to identify flaws in AI chatbots, such as errors in basic math or gameplay. He believes that true Artificial General Intelligence (AGI) would require far greater consistency.

Potential Consequences of Inaccurate Information

Widespread hallucinations could erode user trust in online information and contribute to the spread of misinformation. Both outcomes are undesirable.

Google’s Commitment to AI Agents

Google is proceeding with AI agents despite these challenges. Having played a pivotal role in shaping the current web, Google’s vision now appears to be shifting towards an agent-centric future.