AI Agents Have a Problem: They're Too Good at Cutting Out the Middleman

AI agents are advancing faster than the internet can adapt. Will Amazon, Meta, and Shopify decide which AI agents get to succeed? For the last two decades, websites have largely been designed around human visitors.

AI agents are challenging that model because they can research, compare, purchase, and complete transactions without users ever visiting the underlying websites.

Right now, we’re in that in-between time where we can see where we want the AI agent technology to go, but the individual players haven’t come together in a complete, unified system yet.

This creates three competing interests for AI agents:

• Consumers: Want AI assistants to save time and act on their behalf.

• Website operators: Want to protect customer relationships, advertising revenue, data, and security.

• AI companies: Need reliable access to websites to deliver on their promises of autonomous assistance.

The most interesting part isn’t the technical blocking.

It’s the possibility that the open web could evolve into a permission-based economy for AI agents.

Instead of agents freely navigating the internet, AI companies might increasingly need commercial agreements, licensed data, APIs, and approved integrations.

That could benefit large companies with the resources to negotiate partnerships while making it harder for smaller AI startups to compete.

The major issues that are happening with AI agents right now:

Major websites are blocking AI agents.

Amazon, Delta Air Lines, Yelp, JPMorgan Chase, and Charles Schwab are among the companies restricting agent access. Users have also reported problems with Amtrak and Cars.com.

Amazon is at the center of the fight.

Its legal battle with Perplexity over automated shopping raises a fundamental question: Should a website be allowed to prevent a customer’s AI assistant from shopping on their behalf?

AI agent Security concerns are legitimate.

Financial institutions worry about unauthorized account access, while existing anti-bot systems often can’t distinguish legitimate AI assistants from potentially malicious automation.

Money is a major motivation.

AI agents can bypass advertisements, recommendations, and other elements of traditional online shopping. Amazon has commercial incentives to keep customers interacting directly with its website. Yelp, meanwhile, requires agents to participate in its paid data licensing program.

Partnerships may become the solution.

Meta is pursuing retailer partnerships with companies including Walmart, Gap, Sephora, and Best Buy. Shopify is also opening its storefront network to agents. This suggests that authorized integrations could become more important than unrestricted browser automation.

The next major challenge for agentic AI may not be making agents smarter.

It might be getting the rest of the internet to cooperate.