Two of the biggest retailers in America looked at the same question this month and gave opposite answers. The question: should you let an AI agent walk into your store and buy things for its owner? Amazon said no. On the twenty-first of October, Walmart said yes, and published the tools to prove it.
The centrepiece is the Walmart Agent Gateway, a developer beta built on the Model Context Protocol, the open standard that lets AI systems plug into outside tools. Through it, an approved agent can search Walmart's online inventory and place orders for the person it works for, across Walmart and Sam's Club. The pitch is speed. Agents clicking through a web page meant for human eyes are slow and brittle, and Walmart would rather give them a proper door than make them climb through the window.
Rules before access
What makes this more than a press release is the guardrails. Starting in November, each agent gets its own Agent Profile, a kind of sub-account separate from the customer's main one. The customer decides how much rope it gets. You can demand approval for every basket before it checks out, or set a standing spending cap at 50, 100, or 200 dollars, and you can cut the agent off entirely whenever you like.
Walmart has also written terms of use for agents and says it will block the ones that do not comply. Approved agents see the same prices human shoppers do, with no secret surcharge, and the system is designed to surface a customer's usual items without handing over their full purchase history. It reads less like an open invitation and more like a bouncer with a clipboard.
A deliberate split from Amazon
The contrast is the point. We have spent recent weeks chronicling the other direction of travel, with Amazon, Delta and Schwab shutting third-party agents out and banks locking them away from moving customer money. The nervousness is understandable. An agent that misreads an instruction can order the wrong thing, or far too much of the right thing, and the customer is the one left holding it.
Walmart is making a different wager. Its bet is that agent-driven shopping is coming whether retailers like it or not, and that the smart move is to meet it on your own terms rather than pretend it is not happening. Give agents a controlled lane now, the thinking goes, and you own the relationship when the habit sticks. Amazon is betting the opposite, that the risks still outweigh the rewards. One of them will look prescient in a year. We will see which.
Commentarii · 0