By Roman Aglaganov
This week the two biggest retailers in the US gave opposite answers to the same question. Amazon blocks AI agents from buying on its site. Walmart, on 21 October, opened an Agent Gateway that lets approved agents shop for customers, and promised each agent its own profile, a spending limit and a set of rules.
Walmart has the better answer, and not only because it suits Walmart.
Blocking does not last
Blocking assumes agents can be told apart from people. That gets harder every month: agents use real browsers, real accounts and real cards. A store that blocks them invites an arms race in disguise, and the customers caught in the middle are the ones who asked an assistant to reorder paper towels.
It also misreads who the customer is. The agent is not a stranger; it is acting for someone with an account and a payment method. Turning it away is turning away that person.
Badges work
Walmart’s model is the one banks settled on decades ago: an authorised user on a credit card. The agent gets its own identity, its spending shows up separately, the customer sets the limit or approves each order, and the shop can cut off an agent that misbehaves. Nobody has to pretend the agent is a person, and nobody has to pretend it does not exist.
That framework also answers the question that will matter most: who pays when an agent gets it wrong? Walmart says normal returns apply. It is not a complete answer, but it is a clear one, which is more than most of the industry has offered.
What should be standard
Every shop that takes agents should do three things: give each agent a visible identity tied to a customer, let customers set limits and see the agent’s history, and publish what agents may and may not do. None of this needs new law. It needs shops to decide that agents are customers’ helpers, not intruders, and to build the badge.
