E-commerce platforms, media companies and technology firms are preparing for an internet shaped by personal AI agents, but the transition could either give consumers more control or strengthen the dominance of major platforms. Research led by University of Toronto professor Akshit Kumar found that shopping agents ignored “Sponsored” listings and behaved inconsistently based on search-result position. Meanwhile, experts envision agents enabling micropayments, bypassing recommendation algorithms and connecting consumers directly with businesses—while warning that companies may build new barriers and walled gardens around these tools.
Shopping agents may disrupt sponsored search
E-commerce websites may try to move into the agent-based attention economy, but the path forward remains unclear. Last year, a team led by Akshit Kumar, an assistant professor of operations management at the University of Toronto, examined whether and how agents respond to the way shopping websites present search results.
The researchers found that agents completely ignored items marked “Sponsored.” That result suggests that selling lucrative positions on search-results pages may not be an effective business model in an agent-driven economy.
The models appeared to prioritize products based on their positions in search results, but the preferred position varied. Some models favored items at the top, bottom or middle of the page, and the behavior could also change from one generation of the same model to another. E-commerce platforms may therefore struggle to adapt.
Agents could make micropayments practical
Microsoft economist David Rothschild hopes the shift toward an agent-based web will give consumers more power. At present, paying for online content generally requires a commitment, such as subscribing to a magazine or joining a YouTuber’s Patreon.
Micropayments offer an alternative by allowing users to purchase individual articles, videos or other pieces of content. The approach has failed in the past because processing many small payments creates too much administrative work. Agents, however, could manage that overhead, paying publishers and other contributors directly for each piece of content a user consumes and distributing small amounts of money among the people involved in producing it.
Escaping algorithms and online intermediaries
Other experts share a similar vision. Wellesley computer scientist Mustafaraj predicts that agents could help people escape the algorithms that determine much of the media they consume.
“My hope is to have AI that works for us in the sense that I can change my feed the way I want, not the way TikTok wants,” she says.
Mozilla’s Varma similarly envisions an internet with fewer intermediaries, including companies such as Amazon and Doordash. These intermediaries currently charge fees to facilitate interactions between consumers and businesses. In Varma’s vision, people could use agents to work directly with restaurant agents, for example, to arrange dinner without onerous service fees.
Centralization could shape the agent web
The idea is attractive, but the history of the web has been marked by consolidation and the centralization of power and control. That pattern could become the default for agent-based services as well.
When OpenClaw became the first popular personal assistant agent and exploded onto the scene in early 2026, it briefly appeared that agents might be developed through a grassroots, decentralized process. OpenAI soon brought OpenClaw creator Peter Steinberger into the company, however, while other frontier AI companies began developing their own personal assistant agents.
People who use these agents will depend on the behavior their developers and operators choose to allow. “There is technology that can eliminate all the friction in communication,” Rothschild said. “You’re going to see companies erect barriers to retain their consumer base.”
Personal assistants could create new walled gardens
One possible outcome is already visible in the way Muse is used. Muse has become the default way for many people to find information, shop and manage their digital lives. In principle, it allows users to search the internet for the content most valuable to them, connect directly with vendors and move beyond the walled gardens that have defined the internet for the past few decades.
But Meta may not want users to do that. Muse could instead be designed to keep people within Meta’s own walled gardens—for example, by giving preference to content from its social networks or prominently recommending companies that pay for placement in Muse’s results.
Closed systems may offer safety benefits
A more closed version of the agent web also has advantages. Agents can become powerful tools for hackers and scammers, while major companies have safeguards designed to prevent or at least reduce such abuse. For example, a company’s AI tools may automatically identify hacking-related requests and refuse to answer them.
In a completely open ecosystem, someone could build and popularize an unrestricted system. Major companies also have strong reputational incentives to stop personal assistant agents from making serious mistakes, such as authorizing expensive purchases that users did not intend to make. As a result, they have invested heavily in building reliable guardrails.
Source: www.technologyreview.com