AI companies promise that digital agents can handle online tasks for people, from booking reservations to shopping. But those agents depend on websites letting them in, and many services are restricting access. The result is a growing gap between what agents are supposed to do and what they can actually accomplish across the web.
Websites are setting limits
Users of Meta’s Muse, xAI’s Grok, OpenAI’s Dots, and Instinct have reported that their agents cannot complete tasks on some popular websites. Yelp told TechCrunch it allows only agents that pay for its data licensing program. Delta Air Lines is preventing agents from booking tickets for users, while JPMorgan Chase and Charles Schwab have confirmed blocking agents from account access over security concerns.
Other reported access problems involve Cars.com and Amtrak. Those reports suggest that restrictions may extend beyond companies that have publicly explained their policies. For users, the distinction between a site’s deliberate block and an agent that simply stops working may not be obvious: either way, the task remains unfinished.
Amazon has been a prominent battleground. It banned Perplexity’s shopping agent and sued the startup. Amazon said it had repeatedly asked Perplexity to stop allowing its agents to shop on the site, and argued that companies buying items for customers should operate openly and respect a service provider’s choice about participation.
Perplexity argued that an agent acts as an extension of the user, and said Amazon’s position challenged a person’s ability to hire an assistant. A judge later ordered Perplexity to stop its agents from shopping on Amazon, but an appeals court overturned that ruling in August. Amazon has since filed an amended complaint.
Access determines what agents can do
The dispute has not stopped major technology companies from launching their own agents. Meta, OpenAI, and Google have all moved into the space, but their products face the same basic constraint: an agent’s usefulness depends on the services it can reach.
Amazon has said Muse’s continued access violates its site policies. Reports also describe xAI’s Grok, Instinct’s agent, and OpenAI’s Dots having trouble using Amazon. Elon Musk responded by suggesting he might start a competitor to Amazon. For people considering AI agents, these conflicts make practical access as important as the capabilities companies advertise.
Chris Canal, CEO of AI safety evaluator EquiStamp, told The Verge that blocking access can weaken the perceived value of AI companies. If agents are presented as tools that can do what people do online, but businesses actively prevent that, there are fewer useful tasks those agents can perform.
Partnerships and safeguards shape the experience
Meta has said that turning away a personal agent also means turning away the customer behind it. Rather than trying to fight Amazon, the company has partnered with retailers including Walmart, Gap, Sephora, and Best Buy so Muse can shop on their sites. Shopify has also opened its network of storefronts to AI agents.
These arrangements show one way agents may gain access: through cooperation with businesses. But a partnership does not guarantee a smooth experience everywhere. Users have reported checkout problems on Walmart despite its relationship with Meta, and some Reddit users say Google blocked accounts using Muse because of “suspicious activity.”
Some failures may come from protections designed to catch bots, scams, and hacking attempts, rather than from a policy aimed specifically at AI agents. Canal said agents can trigger the same safeguards as other automated programs. That can leave users unable to tell whether a site rejected an agent on purpose or its security systems treated the agent as suspicious.
More bot controls could affect people too
Cloudflare has introduced tools that let websites block AI crawlers. It also began blocking crawlers that scrape sites for both search indexing and AI training. The company’s actions respond to concerns that AI bots are putting strain on internet infrastructure and causing problems such as inbox disruption.
Bot controls also touch activities where automated competition can frustrate people, including buying concert tickets, booking reservations, and finding certain products. Yet defenses against unwanted automation can create extra steps for ordinary visitors. If sites increasingly ask users to prove they are human, shopping or booking could become more cumbersome even when a person is browsing without an agent.
The web’s response will shape how much AI agents can do in practice. Some businesses are offering access through partnerships, while others cite policies or security risks. As these approaches spread, an agent’s reach may depend as much on individual websites’ decisions and defenses as on the technology behind it.