AI is beginning to move beyond answering questions toward navigating, comparing and completing online tasks, potentially placing software between people and the websites they once visited directly.
For most of the web’s history, people have done the navigating. They search, open websites, compare information, fill out forms and eventually make a decision. Agents could increasingly rearrange that process by allowing people to describe what they want while artificial intelligence handles more of the journey.
Instead of opening several travel sites to compare hotels, for example, someone could ask an AI to find three properties meeting specific requirements, check current availability and organize the results. Shopping agents can already research products according to detailed constraints, while emerging systems can interact with websites or commerce infrastructure and, in some circumstances, help complete transactions.
The shift is still developing, and today’s systems generally require more human involvement than the fully autonomous “personal agent” often imagined. But it raises a larger question for the internet itself: what happens to websites when software increasingly becomes the visitor?
Agents Are Moving From Answers to Actions
The distinction between a chatbot and an AI agent is what happens after the answer.
A conventional chatbot might explain how to book a trip. An agent can potentially perform parts of the process: search multiple sources, compare choices, navigate pages and take actions according to instructions.
The technology is no longer entirely experimental. OpenAI introduced Operator in January 2025 as a system capable of using a browser by visually interacting with webpages, including clicking, scrolling and typing. Its browser capabilities were subsequently integrated into broader agent systems. OpenAI has also developed shopping experiences that compare products and, for eligible merchants and products, support checkout without requiring the customer to complete the traditional journey through a retailer’s website.
Google is pursuing a similar direction. In May 2026, the company announced additional agent capabilities in Search, describing systems intended to carry out tasks rather than merely return information. Google said AI Mode had surpassed one billion monthly users, although using an AI search product does not mean those users are necessarily delegating autonomous tasks to agents.
Commerce is becoming an important testing ground. OpenAI’s Agentic Commerce Protocol is designed to allow merchants and AI systems to exchange structured product and transaction information. Google’s work on agent commerce includes protocols intended to establish what a user authorized an agent to purchase and under what conditions.
The implications extend well beyond shopping.
An agent could theoretically coordinate travel, research products, organize information, interact with business software or perform repetitive administrative work. What makes the model important is not any individual task. It is the possibility of replacing a sequence of human interactions with a single instruction.
Today, someone might search for “best hotel in Montreal,” examine search results, open several websites, compare rooms and then book.
An increasingly agent-driven version of that process could begin with: Find me a highly rated downtown Montreal hotel under my budget, with parking, for these dates.
The AI could perform much of the research before the person ever sees a webpage.
What Happens to Websites When Agents Become Visitors?
This creates an unusual economic problem.
Much of the open web was designed around attracting human attention. Publishers depend on readers visiting articles. Retailers want shoppers entering stores and websites. Search optimization is built partly around earning clicks. Advertising, subscriptions, affiliate links and customer relationships can all depend on direct visits.
AI can interrupt that path.
A system might retrieve information from several sources, synthesize it and give the user an answer without requiring them to visit each source individually. In commerce, however, the effect is not necessarily the disappearance of referrals. Retailers are already reporting potentially valuable traffic from AI services.
Shopify reported that, among product-detail-page sessions in its Q1 2026 commerce data, AI-referred visitors converted at nearly 50% higher rates than organic-search visitors. Orders attributed to AI-powered search also had 14% higher average order values. Those figures come from Shopify’s own platform data and should not be assumed to represent the entire web, but they suggest AI referrals can bring consumers who are already further along in deciding what to buy.
Retailers nevertheless face a trade-off. Reuters reported in August that companies were seeking greater visibility in AI shopping systems while also trying to preserve direct customer relationships and the valuable behavioural data generated when people shop on their own websites.
Publishers face a related but potentially harder question. An AI system can benefit from information on a webpage without necessarily delivering the traditional pageview that supports advertising or subscriptions.
New technical and commercial responses are emerging. Cloudflare has developed controls allowing website operators to manage AI crawler access and is testing a “Pay Per Crawl” system through which participating publishers can set a price for automated access to their content. The program remains in closed beta rather than an established web-wide payment model.
The distinction between an AI crawler and an AI agent also matters. A crawler may systematically collect or index web content. An agent may visit a website in response to a specific person’s request, much as that person might have visited it themselves. Websites will increasingly need ways to distinguish legitimate automated customers, search systems and assistants from unwanted scraping, fraud and malicious bots.
None of this establishes that conventional websites or search engines are about to disappear. As of mid-2026, Forrester assessed most agentic-commerce experiences as still primarily conversational, with humans continuing to make decisions and complete most purchases.
But the direction introduces a significant change in what it could mean to publish something online.
For decades, websites have competed to persuade people to visit them. In an agent-driven internet, they may increasingly need to be understandable not only to people and search engines, but also to software acting on people’s behalf.
That could alter search optimization, advertising, online commerce and publishing. A website’s value may increasingly depend on whether its information can be accurately discovered, interpreted, attributed and acted upon by machines, even when the human asking the question never sees the page itself.
The web would still connect people with information and services. The difference is that an AI agent could increasingly stand in the middle.
About the Author Brad Socha is the founder of The Universal Record, focused on sourced, factual global reporting. Coverage includes international news, geopolitics, technology, and major developments.
