Websites Are Slamming the Door on AI Agents

Personal AI agents can now book flights, order groceries and reserve tables for you. A lot of the time, though, the website they’re trying to use won’t let them in. TechCrunch AI reports that Amazon recently started blocking Meta’s Muse agent from its retail site, so Muse can no longer browse or buy from Amazon’s catalog. Amazon isn’t the only one, and that problem may decide how quickly agents become something regular people use every day.

🚧 What’s actually happening

According to TechCrunch AI, early adopters on social media keep saying their agents get stopped partway through a task. The brands they mention include Yelp, eBay, Zillow, Pizza Hut, Adidas and several airlines. Some users say eBay suspended their accounts for using agentic AI. Another user said Google kicked out the Instinct agent while it was cleaning up their Gmail inbox.

What stands out is that the blocks come from two very different places:

  • On purpose. Amazon’s block of Muse was deliberate. Yelp told TechCrunch it doesn’t allow non-human traffic unless the agent pays for access through its data licensing program. United pointed to terms of use that ban “any robot, spider, other automatic device” without written permission.
  • By accident. Walmart is actually a Muse partner, but shoppers still hit walls. That’s because the site’s “verify you’re human” button can fail when an agent clicks it, and then the agent gets booted out.

Users can’t tell which kind of block they’ve hit. So they end up annoyed at both the retailer and the agent.

🔍 Why this matters now

The web’s security tools were built to tell humans from bots, and that worked when bots were mostly scrapers and spammers. Now a growing share of bots are working for paying customers. A CAPTCHA can’t tell which is which.

Infrastructure players are part of the mess too. TechCrunch AI notes that some people suspect Cloudflare and other content delivery networks (CDNs) are disrupting Muse traffic. They link it to a September 15 change in crawler defaults that lets site owners block AI training while still allowing other types of bots. When rules change at the network level, they can hit thousands of sites at once, often without the site owners even knowing.

Under all of this sits a business question. Delta told TechCrunch it doesn’t currently have a partnership that lets third-party agents book its flights, and that any opening would come with “security and the customer experience in mind.” Put simply, airlines and marketplaces want to control the customer relationship. They don’t want to become a back-end API (a way for other software to plug in) for someone else’s agent.

🤝 The fix being built

A group including Meta, Walmart, Stripe, Sierra, Genesys, Rocket, NiCE and Decagon is working on an open standard for how agents talk to businesses. According to Meta’s blog post, it’ll focus on agent-to-agent communication for online commerce. The goal is to give websites a reliable way to separate good bots acting for a real user from bad ones.

Stripe’s involvement is a big deal. Payments are where trust gets real, and a standard that payment rails support has a much better chance of catching on than one pushed by the agent makers alone.

🔮 What the next 1-3 years likely look like

  1. Partnered agents win. Agents with formal deals (like Muse with Walmart) will work smoothly. Unpartnered agents will keep failing on the big platforms.
  2. Access gets paid. Yelp’s model, where agents pay for data access, will spread. Expect “agent access” to become a line item in licensing deals.
  3. Identity replaces CAPTCHAs. Human checks will move toward verified agent credentials tied to a real user and a real payment method.
  4. Holdouts lose ground. If competitors open up, airlines and marketplaces that keep blocking agents will feel it in sales.

✅ What to do about it

  • If you run a site: Check your bot rules and CDN settings. You might be blocking agents you’d actually want, like customers ready to buy. Decide on purpose what you allow.
  • If you build agents: Don’t count on scraping a site that doesn’t want you there. Work on partnerships and keep an eye on the open commerce standard.
  • If you’re a business user: Before you rely on an agent for a workflow, check that it has official access to the services involved.

The agent-first web is coming, but it won’t arrive all at once. It’ll come one deal at a time. TechCrunch AI’s original report has the full rundown of company responses.

Scroll to Top