Pichai Says AI Agents Will Browse Sites And Use Protocols

What did Pichai say about how AI agents will reach websites?

Sundar Pichai told Alex Heath, on the Sources podcast, that he expects two separate methods for AI agents to access web content going forward. One is browser-based: an agent clicks through pages the way a person would. The other runs on protocols built specifically for agents.

Pichai said both methods will keep growing. But he was direct about which one wins on scale: protocols. He called them "a far more scalable approach" that works instantly and anywhere, while adding that they will take longer to develop than browser-based agents.

He compared the current moment to the early internet, when people assumed home buying and selling would transform almost overnight. Twenty years later, he said, most of that process looks the same, because of "real constraints in the real world." He expects AI agents to produce similarly uneven progress — arriving in "micro equilibriums" that shift over time, which he called even more disruptive than the early web.

What is the Universal Commerce Protocol, and why did Pichai bring it up?

The Universal Commerce Protocol, or UCP, is an open standard for agent commerce that Google built with a group of other companies. Google's UCP page says merchants can adopt it to enable agentic actions inside AI Mode and Gemini, starting with direct buying.

Pichai cited UCP as an example of the "lot of momentum" behind agent and payment protocols, while predicting "a chaotic evolution" before the industry settles on interoperable standards.

Google has already built out supporting infrastructure. A Merchant Center option lets merchants turn on cart transfer, which sends a shopper's items from Google to the merchant's own site to complete the purchase.

Heath also pressed Pichai on how Google protects its own interfaces once agents start doing the browsing on its behalf, noting that Google runs a large ads business that depends on engagement. Heath also cited Amazon's decision in September to cut off Meta's Muse agent from shopping on Amazon.com, with Amazon saying Muse didn't identify itself while browsing.

Pichai's answer to the ads question stayed at the level of principle. "The right answer longer term is you put users first," he said, adding that it has to work for all the businesses providing value and that the market will find an equilibrium — one that keeps shifting. He didn't address what happens to Google's ad model once agents, not people, are the ones viewing pages.

How does Google's own agent fetcher work right now?

Google already runs agents from its own servers using a user agent called Google-Agent. Google classifies it as a user-triggered fetcher — one that acts "to navigate the web and perform actions upon user request."

That classification matters for one practical reason: user-triggered fetchers generally don't follow robots.txt directives, according to Google, because a person asked for the fetch in the first place. Google also notes that user-agent strings can be spoofed, which is why it publishes the IP ranges these fetchers operate from.

In practice, that means a site's robots.txt file — the usual way to block crawlers — generally won't stop a Google-hosted agent from visiting a page on a user's behalf.