Making the internet better for AI agents could make it worse for everyone else

The MIT Technology Review article, written by James O'Donnell and originally published in The Algorithm newsletter on October 19, 2026, argues that the mass arrival of personal AI agents on the web could transform the internet and, paradoxically, make it worse for people. The download contains the full text of the article (it is not a paywall teaser), although it is a short piece, so this summary is also compact.

The starting point is that, until recently, internet users fell into two categories: bots and people. Websites allow bots to perform automated tasks, such as scraping page content, but there are rules about what they can access and how they behave. The web is designed mainly for people, even if we sometimes have to solve puzzles and captchas to prove we are not bots. Personal AI agents, capable of doing things on the user's behalf such as booking flights or hotel rooms, upend that division.

According to the article, these agents have been rolled out at scale with apps such as OpenAI's Dots and Meta's Muse. Muse was downloaded 5 million times in its first month, and agentic web traffic tripled between August and September. The coexistence of agents and human visitors on a web designed to prioritize people, the text notes, has gone badly: many sites block the agents that users send to negotiate prices or plan vacations.

The author argues that agents are really a third category of visitor. They are not human and use the internet at machine speed, but they are not exactly bots either, because they act on behalf of specific people for legitimate and useful tasks; they simply work faster than a person. Because they do not fit into current systems for distinguishing bots from people, some platforms, especially those most concerned about security, have chosen to ban them outright.

As examples, the article cites TechCrunch's recent reporting on problems deploying agents on sites ranging from Pizza Hut to Yelp, which does not allow AI to book restaurant tables. Amazon has blocked Muse agents and is waging a legal battle against the startup Perplexity to block its agents as well. Although a judge recently ruled in Perplexity's favor, Amazon has filed an appeal arguing that AI agents threaten its advertising business by making it harder to distinguish between ads seen by humans and those served to bots.

Ethan Mollick, an associate professor at the University of Pennsylvania's Wharton School, predicts that these blocks, currently a nuisance for early adopters, will end up affecting the entire web. By his reasoning, once it becomes common knowledge that agents can be sent to request refunds or negotiate prices, they will overwhelm customer service departments until they become unusable for real people. Mollick wrote on LinkedIn that humans “will end up very annoyed when their agents aren't allowed to do this on their behalf.”

If demand for personal agents grows, platforms will find ways to let them in, especially since the vast majority of agentic traffic goes to e-commerce sites, where there is money to be made by allowing legions of agents to shop on behalf of people. Shopify, for example, already welcomes agents with dedicated payment rails that bypass pages designed for humans.

The underlying question is what will happen to the internet for the rest of us. Although the web is increasingly designed to exploit our attention, humans were still the intended users. That internet could give way to one optimized for agents that will eventually outnumber us, making it even more unpleasant for the humans who remain.

The article sketches a possible scenario: websites are neglected as destinations for people, and those who want to browse and compare products on their own find worse interfaces. Verification systems are imperfect, and one would constantly have to prove one is human and not an agent (even, the text jokes, with an iris scan). A sweater would not have a single price for everyone, but different prices shown to each person's agents. And you would never know whether an agent's recommendation is influenced by advertising or platform commissions.

That picture is far from the promise of Google, Anthropic and OpenAI, according to which agents will give consumers more power and bargaining leverage. The author considers the promise not far-fetched: agents can genuinely help, for example by fighting improper medical bills or finding cheaper car insurance. Whether they are used that way depends on the rules set for how they interact with websites and what they are allowed to do.

Those rules are still being written. Some will benefit everyone: neither consumers nor businesses want agents exposing credit card details all over the web, and security standards are a good thing. But the most contested ones, such as whether companies can block agents or prevent them from negotiating better deals for consumers, will determine whether the web becomes a highway of agents working for us or against us.