- Walmart announced an Agent Gateway that lets approved AI agents search its inventory and buy for shoppers, while TechCrunch reports Amazon has begun blocking agent purchases.
- The design treats an agent like an authorized card user: its own profile, separate purchase records and spending limits set by the shopper.
- Leaders should ask whether their own sites block agents by default, and who is accountable when an agent buys the wrong thing.
Blocking agents is a choice. So is admitting them.
Two retailers have made opposite bets on the same technology. TechCrunch reports that Amazon has begun blocking AI agents from making purchases for online shoppers. It adds that various travel and booking sites have locked agents out completely. Walmart announced on Wednesday that it is going the other way.
The lesson here is that the choice is not really open versus closed. The harder question is how a company knows who is acting, on whose behalf, and who pays when it goes wrong. Walmart's announcement is an attempt to answer that, and the answer deserves a close look.
What Walmart announced
Walmart's president and CEO, John Furner, presented the new tools in a talk at the Masters of Scale Summit in San Francisco, according to TechCrunch. He said the company is preparing for a world where AI agents become mainstream. The tools cover shopping at Walmart and at Sam's Club, its discount warehouse chain.
The centrepiece is the Agent Gateway. TechCrunch describes it as Walmart's new MCP server, built to let approved agents connect over the model context protocol. Anthropic originally created that protocol. It lets an AI agent reach outside data sources, tools and workflows. Here, it lets an agent search Walmart's online inventory and buy on its user's behalf.
Walmart also claims a privacy benefit. A customer could ask an agent to reorder familiar brands or items without handing the agent their personal details or past purchases. TechCrunch notes that Walmart had already announced a partnership with Meta's Muse. It also offers integrations with ChatGPT, Gemini and other AI apps.
How the controls work
The mechanism that matters is not the connection. It is the identity layer on top of it.
Agent Profiles roll out next month. The first time a customer shops through the gateway with a given agent, a dedicated profile is created. The customer links it to their Walmart account and sets the controls. They can require approval for every purchase. Or they can allow purchases up to a spending limit they choose. They can delete the profile at any time. Purchases made by an agent show up separately from the customer's own.
Walmart compares this to adding an authorized user to a credit card. That is a useful way to see it. The card network does not ask whether a person is trustworthy in general. It asks which card, which holder and what limit. Walmart is applying the same logic to software.
Agent support is open to anyone with a Walmart account. A Walmart+ membership is not required.
Rules for the agents themselves
Walmart is also publishing a set of terms for agents. They spell out how an agent must behave and what permission it has to complete transactions on the site. The company says its returns process applies even when an agent does not act as intended. TechCrunch calls that a customer-friendly condition that could reassure early adopters. A misbehaving agent should not leave a shopper stuck with unwanted goods.
Walmart also says agents cannot take actions beyond what the shopper intends. They cannot access payment information. They can only see the information a given task needs. Agents that break the terms will be blocked.
What remains unanswered
These are Walmart's statements about its own system. TechCrunch's account does not say how Walmart decides which agents are approved. It does not say how Walmart checks that an agent stayed within its limits, or how disputes are handled beyond returns. TechCrunch also does not explain why Amazon chose to block agents.
Those gaps matter. A spending cap protects a shopper from a runaway purchase. It does not by itself prove that an agent followed the shopper's intent.
Questions to put to your team
Few companies sell on Walmart's scale, but every company with a website faces the same fork. Start with these questions.
First, do we know whether our site blocks agents, admits them or has simply never decided? Second, if an agent buys, books or changes an account on a customer's behalf, can we tell it apart from the customer? Third, who is accountable for a mistake: the customer, the agent's maker or us? Fourth, are our refund and dispute terms written with agents in mind?
Locking the door is a policy. So is opening it with rules attached. Leaving the question unanswered is also a policy, and it is the only one nobody chose.
Produced by the WebPulse Newsroom with AI assistance from the original reporting credited below, and checked against that source by our editorial review. How we use AI.
Original reporting: TechCrunch.
