Websites are blocking AI agents like Meta's Muse, and the web may be rebuilt around them
In its weekly AI newsletter The Algorithm, MIT Technology Review argues that the web, long split into bots and people, now has a third kind of visitor: personal AI agents that book flights or hotel rooms, negotiate prices or plan vacations for an individual user. Apps such as OpenAI's Dots and Meta's Muse have recently released such agents at scale. Muse was downloaded 5 million times in its first month, and agentic web traffic tripled from August to September. The piece gives no absolute traffic figure, only the tripling.
The mix of agents and human visitors on a web built mainly for people has gone poorly, the article says, with many websites blocking agents that consumers have sent out. Agents do not fit the systems used to tell bots from people. They are not human and they work at machine speed, but they are not quite bots either, because they act for individuals on tasks that can be legitimate and useful. So some platforms, especially security-conscious ones, have banned them outright. TechCrunch recently reported people having trouble deploying agents on sites from Pizza Hut to Yelp, which does not allow AI to book restaurant reservations. Amazon has blocked Muse's agents and is fighting a court battle against the AI startup Perplexity to block its agents too. A judge recently ruled in favor of Perplexity, but Amazon has filed an appeal arguing that AI agents threaten its advertising business by making it harder to tell ads seen by humans from ads served to bots.
Ethan Mollick, an associate professor at the Wharton School of the University of Pennsylvania, predicts the blocks will soon affect the whole web. Once it is common knowledge that agents can be sent to request refunds or negotiate prices, he argues, agents will overwhelm customer service departments to the point where real people cannot use them. He recently wrote on LinkedIn: "Humans will eventually get very annoyed when their agents aren’t allowed to do this on their behalf."
The article expects platforms to find ways to let agents in if demand grows. The vast majority of agentic web traffic goes to e-commerce sites, so platforms stand to make money if agents shop for people. Shopify is already welcoming agents with their own checkout lanes that bypass pages designed for humans.
The pressing question, the author writes, is what happens to the internet for everyone else. A web that has at least kept humans as the intended users might yield to one optimized for agents that will eventually outnumber people. The article sketches a hypothetical version of that: websites neglected as destinations for real people; worse interfaces for humans who want to browse and compare products themselves; constant demands to prove you are not an agent, perhaps even an iris scan; different prices shown to different people's agents; and agent recommendations you cannot tell are shaped by ads or platform commissions.
That is far from the future promised by Google, Anthropic and OpenAI, where agents give consumers more power and bargaining leverage. The article calls that promise not outlandish, since agents can genuinely help, for example by fighting bogus medical bills or researching cheaper car insurance. Which outcome wins depends on rules for how agents interact with websites, and those rules are still being written. Some will help everyone: neither consumers nor businesses want agents waving credit card information around, so security rules are a good thing. The contested ones, such as whether businesses can block agents or stop them from negotiating better deals for consumers, will decide whether the web becomes a highway for agents working for people or against them.
Key facts
- Meta's Muse was downloaded 5 million times in its first month, and agentic web traffic tripled from August to September; OpenAI's Dots is the other newly launched agent app named.
- Many websites are blocking agents: Yelp does not allow AI to book reservations, Amazon has blocked Muse's agents, and Amazon is appealing a recent ruling in Perplexity's favor, arguing agents threaten its advertising business.
- Wharton's Ethan Mollick predicts blocks will spread across the web and that agents will overwhelm customer service departments once their ability to request refunds or negotiate prices is common knowledge.
- Most agentic traffic goes to e-commerce sites, and Shopify already gives agents their own checkout lanes that bypass pages built for humans.
- The article's worst-case scenarios (worse human interfaces, constant human verification, per-agent pricing, ad-influenced recommendations) are hypotheticals; the rules that will decide the outcome are still being written.
Why it matters
The article frames agents as a new category of web visitor that existing bot-versus-human systems cannot classify. That mismatch is already causing blocks, and the author argues that if demand for agents keeps rising, the web will be increasingly redesigned for them instead of for people. Money points the same way: most agentic traffic goes to e-commerce, so platforms have a reason to let agents in.
Who it affects
Early adopters of agent apps such as Dots and Muse feel the blocks first. Online retailers and platforms such as Amazon, Shopify, Yelp and Pizza Hut must decide whether to ban agents or give them separate paths. Customer service departments could be swamped, per Mollick. Ordinary human visitors could end up with worse interfaces and more verification if sites tilt toward agents.
How to use it
This is an analysis piece, not a product guide, so it offers no setup steps. For readers who use or build agents, it shows where friction is reported: Yelp bars AI from booking reservations, Amazon blocks Muse's agents, and TechCrunch has reported trouble on sites from Pizza Hut to Yelp. Shopify's agent checkout lanes are the example of a platform welcoming agents. The article names tasks where agents can help: fighting bogus medical bills and researching more affordable car insurance.
How solid is it
The two hard numbers (5 million Muse downloads in its first month, agentic traffic tripling from August to September) are stated without an absolute traffic level or a source, and the Dots launch date and download count are not given. The Amazon and Perplexity dispute is reported as fact, though the court, the ruling date and the appeal timetable are not named. Mollick's view is a prediction. The article's forward-looking scenarios are the author's speculation, introduced with words like 'perhaps' and 'imagine'. The piece is an opinion-leaning newsletter analysis.
Risks and caveats
The bad outcomes described are possibilities, not reported events: neglected sites, constant human verification possibly with iris scans, different prices shown to different people's agents, and agent recommendations influenced by ads or commissions. The article also says the web 'might' shift toward agents, and that agents will outnumber humans 'eventually', with no timeline. The opposite outcome is plausible too: the Google, Anthropic and OpenAI vision of agents giving consumers more bargaining power is called 'not outlandish'. Rules on whether businesses can block agents or stop them from negotiating deals remain unsettled.
“Humans will eventually get very annoyed when their agents aren’t allowed to do this on their behalf”
— Ethan Mollick, associate professor at the Wharton School of the University of Pennsylvania, on LinkedIn