I see Google’s unveiling of Gemini Intelligence at the May 12 Android Show as a significant step toward an agent-powered future. Announced alongside a new laptop called the Googlebook, Gemini Intelligence is designed as an underlying layer that works across the Android operating system on laptops, phones, watches, and glasses.
The Googlebook makes that vision tangible to me. Built from the ground up around an AI agent, it can understand what is on the screen and act on it. I could point to a date in an email and have the agent schedule a meeting, or select furniture in an app and see how those pieces might look in my living room.
I believe this ability to complete tasks without requiring someone to open a webpage will fundamentally change how people search, discover information, and conduct commerce. Here is how I expect that shift to affect the search industry.
What the shift to an agentic operating system means
Until now, I have viewed search as a familiar sequence: someone has a question or intent, enters it into a search engine, receives a list of links, and chooses one. Earning a prominent position on that list was the prize, and much of the SEO industry was built around winning that click.
Gemini Intelligence starts from a very different assumption. Search intent still exists, but an AI agent can handle the steps between the request and the outcome. It can read pages, complete forms, and increasingly finish the entire task. Instead of visiting a website myself, I may have an agent visit and use it on my behalf.
When I look for an early example, Chrome Auto Browse stands out. Launched in January and built on Gemini 3, it can manage multistep tasks such as researching flights, filling out forms, scheduling appointments, and managing subscriptions. It then pauses for approval before making a purchase.
That efficiency gives me a clear reason to believe ecommerce will continue moving toward agentic AI.
A 2025 preprint supports this view. Researchers evaluated the declared-tools approach across online shopping, authentication, and content management. They found that giving an agent pre-structured interaction data reduced processing requirements by 67.6% and lowered costs by 34% to 63% compared with parsing a complete HTML document. Task success declined only slightly, from 98.8% with the traditional method to 97.9%.
The architecture behind Gemini Intelligence
To me, the architecture is as important as the interface. AI agents naturally favor websites they can interact with cleanly and efficiently, and Gemini Intelligence can only deliver on its promise if those agents can perform tasks reliably.
I see two protocols as central to making that possible. WebMCP turns a website’s actions into callable tools, while the Universal Commerce Protocol (UCP) allows an agent to complete a sale. Together, they enable an agent to finish a task without requiring a person to load and navigate the underlying webpage.
WebMCP
I think of WebMCP as a labeled menu for AI agents. The API allows a website to declare functions as structured tools an agent can call, including searching inventory, beginning checkout, or submitting a support request.
Google co-developed WebMCP with Microsoft. An origin trial is live in Chrome 149, Firefox has committed to the third quarter of 2026, and Safari is expected to follow in the fourth quarter.
Universal Commerce Protocol (UCP)
I see UCP as the transactional counterpart to WebMCP. It gives AI agents a shared language for discovering products, building a cart, completing checkout, and managing orders without requiring someone to visit the merchant’s website.
Google also offers a consumer-facing layer called Universal Cart. It can collect items as I move across Search, Gemini, YouTube, and Gmail, creating a more connected shopping experience across Google’s products.
The range of companies behind UCP shows me how seriously the industry is taking this shift. Google, Shopify, Walmart, Target, Etsy, Wayfair, PayPal, and Stripe co-developed the protocol, which launched in January.
How I would prepare for agentic AI
My main takeaway is that websites are rapidly evolving from destinations into backends—from places people actively visit into systems agents quietly use. As the operating system becomes a search and action layer, I no longer think ranking is the only question that matters. I also need to ask whether an agent can actually use the site.
To prepare, I would begin by auditing the site’s most valuable actions, whether that means submitting a lead form, completing a booking flow, or reaching checkout. I would determine whether an agent could complete each action reliably and check the site’s Lighthouse Agentic Browsing score much as I would review Core Web Vitals. The goal is to understand whether an agent can use the site, not merely read it.
If I ran an ecommerce business, I would confirm whether the checkout process is accessible through UCP or ACP. I would also continue investing in retrieval and visibility because an agent still needs to find and trust the business before it can act on anyone’s behalf.
Dig deeper: Are we ready for the agentic web?
Inspired by this post on Search Engine Land.
Leave a Reply