TLDR;
Walmart is building a controlled route for outside AI agents to search, choose and buy products from Walmart and Sam's Club. Customers will be able to approve every order or give each agent a spending limit, while the gateway starts as an invitation-only developer beta. This matters because retailers now need checkout permissions, audit trails and product data designed for software shoppers as well as people.
What happened
Walmart announced three connected changes on 21 October: Agent Profiles inside customer accounts, the Walmart Agent Gateway for approved developers and companies, and terms for shopping agents. Together they let customers shop through assistants such as ChatGPT, Gemini and Muse while Walmart controls identity, permissions, data and checkout.
Agent Profiles are due to launch within a month. Each agent gets a separate profile linked to the customer's Walmart account, so its orders do not disappear into the customer's normal purchase history. Approval is required by default, but a customer can instead set a spending limit, with $50, $100 and $200 given as examples. Access can be revoked and Walmart may still ask for extra approval inside the limit.
The Agent Gateway is a direct connection built on Model Context Protocol. Walmart says agents currently navigate pages designed for people, which is slow and error-prone. Its early beta promises current product and fulfilment data, verified identities, controlled permissions, observable actions, rate limits and category restrictions. Access remains invitation-only.
Walmart's terms keep card details and payment-method changes away from the agent. The agent may access only information needed for the task, must stay within the shopper's intent and can be blocked for breaking the rules. Walmart also says its returns process applies if an agent does not behave as intended, and agents will see the same product prices as human shoppers.
Why it matters
This is a merchant-controlled answer to agentic commerce. Instead of letting a browser agent scrape pages and imitate clicks, Walmart is creating an entry point with enforceable rules. That should improve reliability, but it also means the retailer controls participation, product information and how customer history resolves requests such as buying the usual milk.
For brands sold through Walmart, product discovery can now happen without a customer viewing a conventional category page or product grid. The agent may request a shortlist based on price, availability, delivery speed, pack size, previous purchases or constraints expressed earlier in a conversation. Missing or inconsistent catalogue attributes become a commercial problem because software cannot infer a dependable answer from a glossy image alone.
Separate agent identities, default approval and revocable spending limits create a visible chain from customer intent to order. Enterprise retailers need similar controls for substitutions, restricted products, recurring purchases, refunds and disputes. A successful payment is not enough if nobody can explain why the agent chose the item or whether it stayed inside the mandate.
The rollout is still limited. Agent Profiles are not live yet, the gateway is an early beta and Walmart has not published its full tool schema, participant list or commercial terms. Reuters reporting and independent retail coverage confirm the direction, but brands should treat this as infrastructure entering controlled testing rather than an open sales channel with proven demand.
How your brand can benefit / be affected
Brands selling through Walmart should start with their catalogue. Check that titles, variants, pack quantities, dietary or technical attributes, prices, stock, fulfilment choices and return conditions are complete and consistent. Agents need structured distinctions they can compare, especially when several near-identical products compete for one request.
Retailers building a gateway should copy the control pattern before the protocol. Give every agent a verifiable identity, narrow scopes, explicit approval rules, idempotent orders and an event log linking the customer's request to each tool call. Keep payment credentials behind the merchant boundary and design revocation, cancellation and returns before permitting unattended purchasing.
Measure agent orders separately from normal ecommerce. Track the requesting agent, selection path, approval mode, order changes, returns and customer support contacts without exposing unnecessary personal data. That makes it possible to compare conversion and operational cost while detecting agents that generate errors or repeatedly choose products customers reject.
Finally, watch who controls ranking. Walmart's gateway may become a valuable route to customers, but brands still need evidence about how products are selected, whether sponsored listings appear and what reporting suppliers receive. Treat the beta as a signal to improve product data and governance now, while waiting for transparent access and measurement before calling it a new revenue channel.
News date: 21 October 2026. Editorial review: 23 October 2026.