Websites Face a New Reality as AI Agents Blur the Line Between Humans and Bots

Until recently, the web’s gatekeepers mostly sorted visitors into two buckets: people and bots. Automated software has always been part of the online ecosystem—search crawlers indexing pages, scripts checking prices, tools pulling public information—but websites developed a set of norms and technical rules to manage that activity. For the most part, the internet was still built around humans, even if that sometimes meant proving your humanity through captchas and other hurdles.

That tidy division is breaking down as personal AI agents arrive at scale. These systems are designed to carry out everyday online tasks on a user’s behalf—booking travel, ordering food, comparison-shopping, or trying to secure a discount. Their footprint is growing quickly, fueled by consumer-facing launches such as OpenAI’s Dots and Meta’s Muse. Muse, for example, was downloaded 5 million times in its first month. Across the wider internet, “agentic” traffic jumped sharply too, tripling from August to September.

As these agents mingle with human visitors on sites built for human behavior, the results have been messy. Many companies are responding by blocking the new arrivals outright, particularly when agents are used to bargain, request refunds, or plan and purchase travel. TechCrunch recently documented users struggling to deploy agents on everyday services including Pizza Hut and Yelp, which does not permit AI systems to book restaurant reservations.

Major platforms are also drawing a line. Amazon has blocked Muse’s agents and is fighting in court to stop agent access from the AI startup Perplexity as well. Although a judge recently ruled in Perplexity’s favor, Amazon has appealed. One argument Amazon is making is that agents could undercut its advertising business by muddying a critical distinction: whether ads are being served to real people or to automated traffic.

Part of the tension is that agents represent something genuinely new—neither a traditional bot nor a human user. They operate at machine speed, but they are also acting for an individual customer, executing tasks that may be entirely legitimate. Today’s defenses were built to detect and deter bots, not to accommodate software delegates that are “authorized” by a person yet behave like automation.

The current wave of blocks may feel like a nuisance affecting only early adopters, but Ethan Mollick, an associate professor at the Wharton School of the University of Pennsylvania, has warned that the consequences could spread across the entire internet. Once it becomes widely known that anyone can send an agent to contest fees, demand refunds, or negotiate prices, he argues, customer-service channels could be flooded to the point that human users can’t get through. In a LinkedIn post, Mollick predicted that people will eventually become frustrated if companies refuse to let agents act on their behalf.

If consumers keep embracing agents, businesses will have incentives to accommodate them—especially in online retail. Most agent-driven browsing is concentrated on e-commerce sites, where every additional shopper represents potential revenue. Shopify has already signaled where things may be headed by creating dedicated “checkout lanes” for agents that bypass web pages designed for human eyes and hands.

That raises a larger question: what happens to the web experience for everyone else? A future internet optimized for machine intermediaries could leave human visitors with degraded interfaces and more friction. Sites might become less welcoming as destinations for browsing and comparison-shopping, while verification systems—already imperfect—could escalate into constant identity challenges aimed at distinguishing people from agents.

Why it matters

Personal AI agents could shift the internet’s design priorities away from human browsing and toward machine-to-machine commerce and negotiation. Consumers may gain convenience, but they may also face more verification barriers and a web that feels less usable without an agent. Companies, regulators, and platform operators will need to decide whether to treat agents as legitimate users—and how to prevent abuse—before the volume of automated “personal” traffic overwhelms systems built for humans.