AI Agents’ Next Challenge: Gaining Access to Websites

Meta description: Amazon blocks Meta Muse AI agent as personal agents hit retail barriers, while Walmart, Stripe and partners draft open standards for online shopping.

Tags: Meta Muse, AI agents, Amazon, Walmart, agent commerce

Featured image: abstract technology style network of AI agent pathways connecting to retail checkout gateways, no text, no people, no logos

The Meta Muse AI agent can no longer browse or buy from Amazon’s product catalog after the retailer began blocking it from its retail site. That puts a practical limit on consumer agents such as Meta’s Muse, Instinct, and ChatGPT’s Dots, which are positioned to book flights, make dinner reservations, and order groceries without requiring users to set up OpenClaw or another agent on their own computer.

Background: From Chat to Task Completion

Personal agents promise more than answers. They act on behalf of users, so a request can run through a retail, travel, or food-service website like a browser session. When access fails, the user sees a blocked checkout or login, not a clear reason. Some blocks are intentional, like Amazon’s, while others come from traditional anti-bot systems built to stop spam and malicious activity. That ambiguity frustrates users toward both the agent provider and the merchant.

Development: Retail, Travel, and Verification Walls

TechCrunch saw complaints that Muse could not complete a purchase on Walmart’s retail website, and NBC reported the same issue in Walmart’s tests of the AI agent. A Walmart spokesperson said the failures were not intentional: Walmart is partnered with Muse, as announced at Meta’s Connect developer conference in September, and wants to be where customers are, including through AI agent experiences. The likely problem is a human-verification button. If the agent cannot click it, or if the flow is interrupted, the check fails and the agent is booted out. Kevin Wang’s October 5, 2026 post said Walmart’s attempt to stop an agent had gone too far, effectively blocking a human from purchasing or signing in.

Travel shows a different posture; some companies are less keen or have not fully embraced the technology. Flight booking is one of the main use cases being positioned for agents, yet users report airlines rejecting agent requests. Delta said it is protecting customers from unauthorized automated activity and that any opening to AI agents would have to be done with “security and the customer experience in mind.” Delta’s General Manager of Global Communications, Heena Chavda, said the company has no current partnership or integration enabling a third-party AI agent to shop for or book Delta flights on a customer’s behalf, while continuing to evaluate external AI agents. United pointed to Terms of Use prohibiting any robot, spider, other automatic device, or manual process used to monitor or copy the website or its content for an unauthorized purpose without United’s prior written permission; it did not answer whether it blocks specific personal AI agents, like Muse.

Complaints also name Yelp, eBay, Zillow, Pizza Hut, Adidas, and various airlines, with some users believing blocking has increased. One person said eBay suspended their account over agentic AI use; another said Google kicked out Instinct while it cleaned Gmail. Martin Kessler said on September 28, 2026 that Adidas would not get his business that night. Yelp said non-human traffic is not permitted unless an agent pays to access its content and data through its data licensing program, so quote requests, waitlist additions, or table bookings likely fail without formal partnership. eBay said it does not prohibit all third-party shopping agents from completing purchases, but restricts unauthorized agents and actions such as automated scraping and model training. Adidas, Zillow, and Pizza Hut did not respond or declined by publication.

Latest Progress: Standards, Cloudflare, and Connector Lists

Meta, Walmart, Stripe, Sierra, Genesys, Rocket, NiCE, and Decagon have begun working on an open standard for how AI agents communicate with businesses, aiming to separate good bots acting for users from bad bots. Meta said in a blog post the protocol will focus on agent-to-agent communication for online commerce. Infrastructure is also in play. Cloudflare offers sites protection from AI agents that could train on their data or manage automated activity. Some suspect Cloudflare and other content delivery networks, or CDNs, are disrupting Muse traffic, citing a September 15 change to crawler defaults that lets site owners block AI training while allowing other bots; sites previously set to block AI bots for security were shifted to block AI agents on pages with ads. Cloudflare said it had no specific data on personal AI agent blocking and pointed to Cloudflare Radar, a free public data hub that shows rising bot activity—a trend Cloudflare has also commented on—but does not distinguish good from bad bots. Cloudflare also runs a marketplace where AI bots pay for data and recently began testing a web browser built specifically for AI agents.

Meta’s interim strategy is partnership. Its Muse app maintains a list of “connectors,” or partners, and access failures at partner sites can be reported as bugs. The list continues to grow, although many newly announced partners have not yet appeared. Muse’s rapid expansion now includes partnered tools for small businesses, making a larger connector-list reorganization likely soon. Meta said: “Turning away a personal agent means turning away the customer behind it.” The company said the path should start with clear norms, give businesses more predictable control, and evolve toward more efficient agent collaboration.

Current Stage: A Splitting Web

The web is at a transition point, and consumer confusion remains high. Walmart and Meta push agent-friendly access; Amazon and Delta signal restrictions or evaluation; Yelp, eBay, United, and Cloudflare reflect policy, licensing, or infrastructure controls. Human-verification systems designed for people are colliding with an agent-first web. The next phase will be decided less by whether agents can shop or book and more by whether websites accept standards, paid data access, formal partnerships, or explicit Terms of Use.