TL;DR Summary:
Two Paths Ahead: AI agents will reach websites through familiar browsing and specialized protocols, with protocols offering greater scalability but taking longer to standardize.Commerce Gets Disrupted: Google’s Universal Commerce Protocol could enable direct purchases through Gemini, while merchants and publishers navigate an uncertain shift in traffic, ads, and business models.Agent Access Expands: Google is widening Gemini’s agent capabilities, raising fresh questions about robots.txt, bot identification, and whether websites are discoverable and citable by AI systems.Will AI agents shop and browse websites the same way humans do, or will sites need new technical doors built just for bots? Google CEO Sundar Pichai answered that question directly in a podcast interview, laying out his view on AI agents browsing and protocols reaching the open web. The timing matters because merchants, publishers, and ad buyers are all trying to figure out how to prepare their sites before agents become a major source of traffic.
Pichai’s Two-Path View On AI Agents Browsing And Protocols
Pichai told Alex Heath, host of the Sources podcast, that he expects AI agents to reach websites in two separate ways. The first is browsing, where an agent moves through a page much like a person clicking and scrolling. The second is through protocols, which are standardized technical connections built specifically for agents to use instead of a visual browser. Pichai said both approaches will keep growing. He called protocols “a far more scalable approach” because they work anywhere and respond instantly, but he said they will take longer to build than browsing tools that already work today.
Why Protocols Take Longer To Scale Than Browsing
Pichai pointed to momentum behind MCP (a protocol that lets AI models connect to outside tools and data), payment protocols, and Google’s own Universal Commerce Protocol, known as UCP. UCP is an open standard Google built with a group of other companies to let merchants enable direct buying through AI Mode and Gemini. Google’s UCP page lists lodging booking and food ordering as features coming soon, each with its own waitlist. Pichai predicted “a chaotic evolution” before the industry agrees on shared standards, comparing this period to the early internet, when people expected home buying and selling to transform overnight. Twenty years later, he said, much of that process still works the same way because of “real constraints in the real world.” As MCP and UCP gain traction, site owners are left wondering whether their own pages are even set up to be found and cited by these agents in the first place, rather than just ranked in a traditional search results page. This is the gap that Nuwtonic SEO is built to address, giving site owners visibility into whether their content is actually discoverable and citable by AI systems operating through these emerging protocols.
What AI Agents Browsing And Protocols Mean For Ads And Commerce
Heath asked Pichai how Google protects its ad business, which depends on user engagement, once agents start doing the browsing instead of people. He raised Amazon’s move in September 2025 to block Meta’s Muse agent from shopping on Amazon.com, with Amazon saying Muse did not identify itself while browsing. Pichai said the right long-term answer is to put users first and let business models adjust around that. He said markets will reach an “equilibrium” that shifts over time, and that AI agents will likely create “micro equilibriums” that prove even more disruptive than past internet shifts. He did not explain how Google’s ad system will function once agents, rather than people, view the pages.
How Google’s Own Browsing Agent Works Today
On the technical side, Google already runs a hosted browsing agent called Google-Agent. Google lists it as a user triggered fetcher, meaning it only activates when a person asks it to complete a task. Because the end user requested the action, Google says these fetchers generally do not follow robots.txt rules, the file site owners use to block automated crawlers. Google also warns that agent user-agent strings can be faked by others, so it publishes the IP address ranges its own fetchers use, giving site owners a way to verify real traffic.
What Comes Next For Gemini And Agent Access
Pichai said Google plans to expand Gemini’s agent features beyond Ultra and Pro subscribers over the coming months, widening access to the tools driving this shift. Watch for how merchants respond to UCP adoption and whether robots.txt rules gain any real authority over agent traffic. The core issue Pichai raised, balancing user needs against existing ad and business models, remains unresolved and worth tracking closely. As Pichai’s two-path vision plays out, tools like Nuwtonic SEO are already helping site owners track AI citation signals and GEO visibility, so they aren’t left guessing whether agents can actually find and trust their pages.
