Websites Restrict Access as AI Agents Handle Online Tasks

What changed

Autonomous AI agents from OpenAI, Meta, Google, Instinct and others are being built to scour the web and handle tasks such as booking dinner, managing inboxes and buying shoes. But websites are beginning to restrict access by these agents, creating a conflict between what the tools promise and what the internet will let them do.

Why it matters

An agent can only be useful if websites accept its requests. If access rules stop it from reaching pages or completing actions, a polished demo can turn into a half-finished task: the dinner remains unbooked, the inbox untouched, and the shoes still in the cart.

That makes website access part of the product, not a minor technical detail. The companies promoting agents can improve their models, but they cannot by themselves decide how every website will respond. For builders and buyers, reliability will depend on both the agent and the sites it needs to use.

The scale of these restrictions, and the methods websites use, remain unclear. But the basic problem is already visible: autonomous agents are being sold as internet workers while the internet is not obliged to cooperate.