AI agents are reshaping how the web works

AI agents are changing how people find information, shop, and manage tasks online, pushing websites to serve software alongside human visitors. Unlike earlier visions of a machine-readable internet, language models can interpret ordinary web content without special annotations. Multimodal agents can also navigate visual interfaces, clicking links and using menus.

Competing approaches are shaping this emerging agentic web. Browser-based agents operate through existing interfaces, while connectors let assistants access services through text commands and code. Connectors can impose safeguards: Muse’s integration with Stripe’s Link payment service keeps credit card details hidden from the agent and requires human confirmation before purchases. More capable browser agents, however, make it harder for websites to separate human and automated access.

The economic consequences could be substantial. Publishers face declining traffic as AI-generated answers replace visits, while shopping agents may bypass advertising and sponsored listings. Time is experimenting with agent-targeted ads, and businesses including Etsy and Expedia are developing integrations with AI companies. Amazon has blocked Meta’s Muse, illustrating tensions over access and control.

Agents could support direct payments to creators, reduce reliance on intermediaries, and improve accessibility for older and disabled users. They could also concentrate power within dominant platforms, expose users to security risks, and leave nonusers excluded from some services. Whether human browsing and automated activity coexist openly will depend on platform incentives, safeguards, and what users choose to delegate.