AI agents tripled web traffic in a month. Now sites are banning them

The internet's visitor count just changed shape: agentic web traffic tripled between August and September after the launch of personal AI apps like OpenAI's Dots and Meta's Muse. Muse alone was downloaded 5 million times in its first month. And the web, built for human eyes and human clicks, is now fighting back. Amazon has blocked Muse's agents, Yelp won't let AI book restaurant reservations, and Pizza Hut has reportedly turned agents away at the digital door. The era of the personal AI agent has officially collided with a web that was never designed for it, and the fallout is just beginning to reach every e-commerce site, ad model, and customer service queue you interact with daily.

This is not the familiar bot-versus-captcha war. Until recently, the web had two kinds of users: people, who navigate with browsers and patience, and bots, which crawl pages under strict rules. Personal AI agents are a third category. They act for a specific human, moving at machine speed, doing legitimate tasks like comparing prices or negotiating refunds. But they don't fit the systems that distinguish humans from bots, so security-conscious platforms have defaulted to the bluntest available tool: a total ban. That leaves early adopters locked out of everything from fast-food ordering to restaurant reservations, and it is already spilling into court. Amazon is waging an appeal against AI startup Perplexity after a judge ruled in favor of Perplexity, with Amazon arguing that agents make it impossible to tell whether an ad was seen by a person or served to a bot, threatening its advertising business.

The consequences will not stay contained to early adopters. Wharton associate professor Ethan Mollick predicts that once the public realizes agents can be dispatched to request refunds or negotiate prices, customer service departments will be overwhelmed to the point of being unusable for real people.