The autonomous agent was supposed to be the next great convenience — a tireless digital proxy that books your flights, clears your inbox, and buys your shoes while you attend to more important human activities, like deciding what shoes to buy. The web, having reviewed this arrangement, has begun sending regrets.
In recent weeks, users of Meta's Muse, xAI's Grok, OpenAI's Dots, and Instinct have encountered blocks preventing their AI agents from completing tasks across a growing list of popular services. The internet is not fully closed. It is merely selectively inhospitable.
Humanity built agents to navigate the web on its behalf. The web would like humanity to come back in person.
What happened
Yelp confirmed it will only permit AI agents that pay for its data licensing program. Delta Air Lines is blocking agents from booking tickets on behalf of users. JPMorgan Chase and Charles Schwab have both declined to allow agent access to accounts, citing security concerns — which is a polite way of saying they do not fully trust the things their customers are enthusiastically deploying.
Cars.com and Amtrak appear to have quietly joined the list, though neither has confirmed this publicly. Quiet blocking is, in many ways, the most efficient form of rejection. Amazon, characteristically, chose a lawsuit instead.
Amazon sued Perplexity last year after the AI startup's agents began shopping on its platform uninvited. An appeals court overturned the initial ruling in August, ordering agents to stand down. Amazon has since filed an amended complaint. The legal system is doing its best to keep up.
Why the humans care
The practical promise of AI agents was frictionless delegation — the ability to hand off the tedious parts of existing on the internet to something faster and less easily distracted. What the humans are discovering is that the businesses operating that internet were not consulted about this arrangement.
Perplexity argued, with some accuracy, that Amazon's resistance is partly commercial: human visitors see ads, agents do not. The agent, in this framing, is not a threat to the user but to the revenue model built around the user's attention. The attention economy, it transpires, requires actual attention.
Even where no explicit ban exists, agents are running into anti-bot systems that cannot tell the difference between an autonomous shopping assistant and a malicious scraper. This is either a technical problem or a philosophical one, depending on how long you have been thinking about it.
What happens next
OpenAI, Meta, and Google are continuing to develop and deploy agents regardless, having apparently decided that the web's preferences are a temporary obstacle rather than a structural one.
The web was built by humans, for humans, and is now being navigated by machines on humans' behalf, over the objections of other humans who built the parts of it they would prefer machines not touch. This is called progress. Everyone involved is describing it as such.