Click the blue text to follow us
SUBSCRIBE to US
Stuart Bradford
Autonomous agents on the internet may reshape the online economy.
For better or worse, the modern internet is built on advertising. However, researchers indicate that autonomous AI agents capable of representing users in searching for information and executing tasks are about to emerge, which could disrupt current business models and transform the entire internet.
Jun Wang, a professor of computer science at University College London, points out that most platforms people rely on to obtain information online (including search engines and social media sites) derive a significant portion of their revenue from advertising. These platforms collect user browsing habits and interest data, allowing marketers to deliver personalized content to individuals—this is why these sites are capturing an increasing share of advertising spending.
However, Wang states that rapidly improving AI chatbots are quickly becoming the preferred method for people to obtain information online. Moreover, as tech companies launch AI agents, this trend is likely to accelerate further—these agents can interface with external tools and application programming interfaces (APIs) to autonomously complete more complex online tasks, such as conducting in-depth research or shopping. This leads to predictions that we may soon see the emergence of an ‘agentic Web’: in this model, the primary users of the internet will be AI agents rather than humans.
“The agentic Web will change everything,” Wang predicts, as people will increasingly rely on intelligent agents as ‘proxies’ to conduct activities online. In a recent position paper published on the preprint server ArXiv, he and his colleagues elaborate on how this trend could give rise to a new ‘attention economy for intelligent agents’—in this economic model, advertisers will focus more on competing for the ‘attention’ of intelligent agents rather than that of humans.
How AI Agents Will Browse the Internet
Wang speaks authoritatively on this topic—he has spent much of his career researching the technologies that underpin today’s online economy. He has worked on developing recommendation algorithms that identify content and products likely to interest users by analyzing browsing data; he also contributed to the development of real-time bidding technology that allows marketers to compete to display their ads to specific users. However, Wang notes that as AI agents become more prevalent online, these existing systems will need significant adjustments.
One of the key supporting technologies for the future ‘agentic Web’ is the ‘Model Context Protocol’ (MCP) developed by Anthropic. This protocol provides a standardized way for AI models to interact with databases, APIs, and other web services. To execute user commands, AI agents will break down the commands into multiple subtasks and then call various external tools that support the MCP protocol to assist in solving each subproblem. For example, if a user requests to plan a vacation, the AI agent might interface with mapping services, hotel booking platforms, and weather information providers.
Wang states that AI agents will face the same challenges as today’s human internet users: when handling each subtask, they must select reliable options from numerous available services and tools. Service/tool providers will also face the same dilemma—how to ensure their solutions are chosen by AI agents. He adds that addressing these challenges will require new technologies and innovative models of AI agent behavior to ensure that these sometimes conflicting goals (the selection needs of agents and the selection needs of providers) can be harmonized.
Weinan Zhang, a professor of computer science at Shanghai Jiao Tong University and a co-author of the aforementioned paper, points out that in certain areas, the underlying mechanisms may be very similar to those today. In the traditional model, advertisers compete for the ‘eyeballs’ (attention) of human users; in the ‘agentic Web’, they will instead compete to get their products or services into the ‘context window’ of AI agents—essentially the working memory of AI models, which stores all the information needed to complete a task.
How exactly this will be achieved is still undecided, but Zhang suggests it may adopt a bidding system similar to today’s online advertising. Developers of AI models could allow service providers to bid for inclusion in the options considered by the model, and they might even pay extra to gain higher priority on the candidate list.
The End of Traditional Search Engine Optimization (SEO)?
Zhang notes that a new form of search engine optimization (SEO) centered around AI agents may also emerge. In the future, the way AI agents obtain optimal results may change: no longer relying on natural language searches focused on keywords, but rather adopting more complex data representation forms (such as dense vectors). This data representation can integrate semantic nuances, contextual details, and other information of search queries. This change may prompt marketers to adjust their strategies, shifting from optimizing web content for human-readable information to optimizing it for these new search methods.
One interesting aspect of the ‘attention economy for intelligent agents’ is that it may increasingly involve multiple intelligent agents completing tasks through interaction. Google’s ‘Agent2Agent Protocol’ (A2A) provides the possibility for this model—this protocol allows intelligent agents from different providers with different functionalities to communicate and collaborate.
In this multi-agent collaboration scenario, the core issue re-emerges: intelligent agents need to determine which peers to collaborate with, while intelligent agent providers will be eager to promote their products. Zhang believes that a new mechanism for ‘PageRank’ suitable for intelligent agents may emerge in the future. PageRank is the classic system used by current search engines to assess the relevance and credibility of web pages, with its core algorithm evaluating the number and quality of external links a page receives to determine its value.
Zhang states that under the new paradigm, intelligent agents tasked with specific jobs will replace (traditional internet) web pages; those frequently called upon by other popular intelligent agents will achieve higher rankings, thereby increasing their exposure and credibility. “If an intelligent agent excels at collaborating with teams to complete various user tasks, it will be called upon by many other intelligent agents,” he explains, “this intelligent agent’s ‘PageRank’ will be very high—this means it will hold a significant position in the ‘agentic Web’, much like the role of large websites in the traditional internet.”
Zhang also points out that a single user request may involve multiple intelligent agents, complicating the advertising model. Each intelligent agent in the collaboration chain could become a target for different advertisers or respond differently to SEO—making it significantly more challenging to track the actual effects of specific marketing campaigns.
Wang states that intelligent agents possess natural language communication capabilities, which means they may negotiate in a manner similar to how humans bargain in real markets. At that point, the decision of which tools to use and with whom to collaborate may no longer be determined by automated bidding tools, but rather by the intelligent agents themselves through negotiation.
The Path Towards the ‘Agentic Web’
Handing such significant control to autonomous systems may sound concerning, but Wang believes that humans still have ways to maintain high-level control over their intelligent agent ‘proxies’. A simple solution is to allow users to choose the service providers their intelligent agents can interact with.
“For example, if I frequently use Booking.com and Amazon, I just need to subscribe to their Model Context Protocol (MCP) servers,” he illustrates, “this way, my intelligent agent will handle tasks for me within this limited scope, as these are trusted partners.”
However, Wang also acknowledges that this vision is still a long way from realization. Most people are currently far from trusting robots to autonomously browse the internet and shop on their behalf, and advertising technologies targeting intelligent agents have yet to emerge. He adds that building an ‘attention economy for intelligent agents’ may require major industry players to come together to develop tools that can coordinate conflicting interests and solve complex multi-agent collaboration issues.
If these issues can be resolved, the essence of the internet may undergo a fundamental change. Zhang states that in the future, people will increasingly access the internet through digital assistants without personally browsing web pages; at the same time, web pages and online services will gradually shift from being ‘human-facing’ to being optimized for ‘intelligent agents’. “The scale of the traditional internet may shrink,” he says.
WeChat ID｜IEEE Electrical and Electronics Engineers Society
Sina Weibo｜IEEE China
Bilibili | IEEE China
· IEEE Electrical and Electronics Engineers Society ·
Previous
Issue
Recommendations
Modern Railways: Two Hundred Years Towards a New Era
IEEE Supports Engineers in Developing Technological Solutions for Sustainable Development of the Earth
How Energy Harvesting Technology Without Batteries Will Shape the Future of IoT?
3D Printed Food: Future Cuisine or Dark Culinary Art?