- Google launched AI Mode to every U.S. Search user, replacing links with agent-driven answers.
- Anthropic’s MCP and Google’s A2A protocol are becoming the plumbing that connects AI agents to the web.
- The ad-supported internet breaks when the “user” clicking through pages is a machine, not a person.
- Publishers face an existential threat as agents summarize content without sending traffic.
Google Just Told You the Future of Search Isn’t Search
At I/O 2025, Google did not present a better search engine. It presented a replacement for one.
AI Mode is now available to every Search user in the United States — hundreds of millions of people who can converse with an AI agent that visits web pages, summarizes them, and shops on their behalf. Project Mariner handles 10 tasks simultaneously, clicking around websites while users do something else. Deep Research generates reports from dozens of sources. Project Astra lets users speak to an agent and share what they see through their camera.
“We believe AI will be the most powerful engine for discovery that the web has ever seen,” said Liz Reid, Google’s VP of Search. What Google described is not search. It is delegation.
The Plumbing Is Already Being Laid
At Build 2025, Microsoft CTO Kevin Scott laid out his vision for an “open agentic web” — agents acting across the internet on behalf of users, booking flights, comparing prices, managing workflows, without anyone opening a browser tab.
Google’s Gemini SDK now natively supports Anthropic’s Model Context Protocol (MCP), the emerging standard for connecting agents to data sources. Google’s own Agent2Agent (A2A) protocol handles inter-agent communication. Together, MCP and A2A are becoming the TCP/IP of the agentic web.
A decade ago, businesses without a mobile-friendly website might as well have not existed. Now the same logic applies to APIs. If a business doesn’t expose structured, machine-readable data, AI agents can’t find it. The primary consumer of the web is no longer a person with a browser. It’s a model with an API call.
The Ad Model Breaks When the User Is a Machine
The internet’s economy runs on human attention: clicks, impressions, scroll depth, time on page. AI agents don’t have eyes. They don’t see banner ads. They don’t click sponsored links. When Google sends an agent to a website instead of a person, the ad-supported model collapses.
Companies that sell goods — DoorDash, Ticketmaster, Amazon — may adapt. But publishers, whose revenue depends on traffic, face an existential reckoning. An AI summary of a 2,000-word article captures the value but sends zero clicks. The content creation that feeds the web loses its funding mechanism.
“Human attention is the only truly finite resource,” a Google communications lead said during I/O. That time the company gives back comes at someone else’s expense.
The Dead Internet Gets Closer to Real
The “dead internet theory” — most online activity already generated by bots — has been a fringe concern for years. The agentic web makes it mainstream. Gartner projects 80% of customer service interactions will be handled autonomously by 2029. The agentic AI market is expected to grow from $28 billion in 2024 to $127 billion by 2029.
But the technology is not ready. DeepMind CEO Demis Hassabis acknowledged the problem at I/O. “You can easily, within a few minutes, find some obvious flaws with these systems,” he said. Hallucinations remain unsolved. Trust remains unearned.
For two decades, Google organized the web around links, PageRank, and human curiosity. Now it is dismantling the model that made it work. The only question is who gets left behind when the primary user of the internet is no longer human.
