Google has placed a new layer of artificial intelligence directly under Android. Called Gemini Intelligence, it lets software agents read screens, click buttons, fill forms and complete purchases without a human in the loop. The change, detailed today by Search Engine Land, turns websites from destinations into back-end services that agents consume on users’ behalf.
Search results once delivered lists of blue links. Now agents handle the middle steps. They research flights, compare prices, add items to carts and pause only for final approval. Commerce follows the same path. A single Universal Cart spans Search, Gemini, YouTube and Gmail. One click, or none at all, finishes the buy.
Agents don’t browse like people. They demand structured data. Sites that speak their language win. Those that don’t fade from consideration.
The shift gathered speed at Google I/O 2026. Executives rolled out Gemini 3.5 Flash as the default engine for AI Mode. Google’s official blog described a redesigned search box that swallows text, images, videos, files and open Chrome tabs. It expands dynamically. It suggests intent before users finish typing. And it hands complex jobs to persistent agents that keep working in the background.
Elizabeth Reid, who leads Search, told the audience that AI Mode now exceeds one billion monthly users. Those users receive personalized dashboards, information agents that monitor apartment listings or sneaker drops, and shopping assistants that check stock by calling stores. The New York Times captured the stakes three days ago. More than 60 percent of U.S. searches already end without a click. Its opinion column warned that the new system compresses the old meandering journey into an immediate answer, often at the expense of publishers.
But Google sees revenue, not just answers. Conversational ads now appear inside AI responses. Sponsored products come with Gemini-generated explanations. Users can chat directly with ad units. The model turns advertising from interruption into helpful continuation. MarketingProfs reported on May 22 that these formats aim to feel native to the conversation. The roundup noted that many advanced features launch first for Google AI Pro and Ultra subscribers.
Under the hood sits a technical stack built for agents. WebMCP, a new API, lets sites declare their actions as callable tools. Inventory checks, checkout starts, price lookups all become functions an agent can invoke directly. The Universal Commerce Protocol, or UCP, standardizes product data, carts and payments across retailers. Shopify, Walmart, Target, Etsy, Wayfair, PayPal and Stripe helped shape it. The protocol launched in January. Chrome added support in version 149. Firefox follows this quarter. Safari arrives later this year.
A 2025 research paper quantified the gains. Agents given declared tools instead of raw web pages cut processing time by 67.6 percent, lowered costs between 34 and 63 percent, and achieved success rates of 97.9 percent compared with 98.8 percent for traditional browsing. The preprint, available on arXiv, examined online shopping, authentication and content management tasks. Sites that prepare for this approach gain clear advantages.
Chrome’s Auto Browse feature, first shown in January 2025, already demonstrates the pattern. Powered by earlier Gemini versions, it completes multistep workflows, compares products side by side, scans documents and pauses for user confirmation on purchases. The latest models make those pauses rarer. They handle longer horizons. They remember context across days.
Retailers face immediate choices. They must audit every customer journey for agent compatibility. Lighthouse now includes an Agentic Browsing score that measures how easily automated systems can parse and act on pages. Structured data, clear call-to-action labels, fast APIs and explicit permission flows matter more than ever. SEO teams that once optimized for rank now optimize for execution.
The implications stretch beyond marketing. Personal intelligence features pull from Gmail, Photos, Calendar and past searches to anticipate needs. A user who mentions a dinner party might receive a generated shopping list, reservation options and even outfit suggestions drawn from their wardrobe photos. Gemini Spark, an always-on agent, drafts emails, generates reports and books travel while the user sleeps.
Critics question privacy and accuracy. The New York Times piece highlighted how AI summaries sometimes distort nuance. X users today echo the concern. One post noted that even Gemini 3.5 Flash still fabricates details when retrieving fresh web data. Another praised the system for delivering quality reading material without context loss. The debate continues in real time.
Yet adoption accelerates. Over 900 million people use Gemini, according to figures shared at I/O. They have generated more than 50 billion images. The assistant now lives inside every major Google product. Android, Search, Workspace, Chrome, YouTube. Each surface feeds the same underlying intelligence.
Merchants who adopt UCP and WebMCP report higher conversion inside agent flows. Nike, Sephora, Target and Walmart serve as launch partners for Universal Cart. The brand remains the merchant of record whether checkout finishes on their site or through Google Pay. Background monitoring surfaces price drops and stock alerts without users opening apps.
Google Cloud chief Thomas Kurian tied the announcements to broader enterprise tools. Gemini 3.5 Flash outperforms prior models on agentic benchmarks while running at lower cost. The co-design of model and hardware, using custom TPUs, delivers faster reasoning on long tasks. Enterprises can now build their own agents that act across internal systems and external web services.
The transition carries risks for the open web. If agents complete transactions without ever loading a publisher’s page, traffic and ad impressions collapse further. Some brands already see AI citations favor their own controlled content. An analysis of 6.8 million citations across ChatGPT, Gemini and Perplexity found 86 percent came from brand-owned sources. Forums contributed just 2 percent.
Still, Google insists the system benefits users first. Agents reduce friction. They compare options across dozens of sites in seconds. They remember preferences. They act when users lack time. The company frames this as progress from search to completion.
Preparation advice from experts remains practical. Map every user action an agent might perform. Expose those actions through structured APIs. Test with Lighthouse. Monitor how often agents succeed or abandon. Update content for clarity because agents parse text differently than humans. And accept that the visit metric may soon measure agent sessions instead of browser loads.
Today’s Search Engine Land article crystallizes the moment. Gemini Intelligence is not another chatbot. It is an operating system upgrade that places autonomous agents between users and the internet. Search becomes a background service. Commerce becomes a conversation. Websites become toolkits.
The companies that redesign for agents will capture the next wave of transactions. Those that wait risk becoming invisible to the systems users rely on most. Google has drawn the line. The race to stand on the right side of it has begun.
WebProNews is an iEntry Publication