Walmart gives AI agents a profile, a spending limit, and written rules: what online sellers can learn
On October 21, 2026, Walmart announced how AI agents will shop on its platform: a dedicated profile for each agent with purchase approval or a spending limit, a technical entry point via MCP, and written rules, from the same prices for agents and people to card details the agent never sees. What changes and what online sellers can learn.
In brief
- Walmart has announced three tools for people who shop with an AI agent: a dedicated profile for each agent inside the customer’s account, a technical entry point built specifically for agents, and written rules that agents must follow.
- Customers decide whether to approve every purchase or give the agent a spending limit, such as $50, $100, or $200, and can delete the profile at any time. The agent never receives card details and sees the same prices as a person.
- Walmart decides for itself how agents get in, and puts it in writing. For online sellers, it is the most complete example so far of what getting ready means.
The problem Walmart starts from is concrete. Today, when a customer asks an AI agent to do the shopping, the agent logs into the site with the customer’s account, and every purchase it makes blends in with the person’s own. There is no way to tell who bought what, and the customer has no simple way to tell the agent how far it can go. Walmart has announced how it plans to fix this, with profiles, a dedicated entry point, and written rules. We read the announcement to understand what changes and what it can teach online sellers.
Index
One profile for each agent
The first tool is called Agent Profiles and, according to Walmart, it is “launching in the next month.” Each agent the customer uses will have its own profile inside the customer’s account, and its purchases will stay separate from everything else. The profile is created automatically the first time the customer uses an agent to buy from Walmart; the customer links it to their account by signing in to Walmart, and manages it from there.
Control is the part that matters. The customer can choose to approve every single purchase, or give the agent a spending limit, such as $50, $100, or $200, within which it can buy on its own. They can see each agent’s purchases and delete the profile at any time. Walmart uses a simple comparison: it is like adding another person as an authorized user on your credit card, with a spending limit and the ability to see how they use it.
An entry point built for agents
The second tool addresses a technical problem that Walmart describes bluntly: the site is built for people, who browse, click, and sign in with their own account. An agent that uses it the same way follows steps designed for a human, so it is slow, makes mistakes, and sometimes gets stuck. That is why Walmart is building the Walmart Agent Gateway, an entry point reserved for agents and based on MCP, the protocol AI models use to talk to external services. It will launch as a beta for businesses and developers.
The dedicated entry point also protects privacy. If a customer usually buys a certain bread or a certain brand of milk, the agent can find those regular items without Walmart handing over the purchase history. The agent gets the list of what needs reordering; the rest stays with Walmart.
The rules agents must follow
The third tool is the terms of use for AI agents, which set out what an agent can and cannot do when it shops on behalf of a Walmart customer:
- purchases must be approved by the customer, unless the customer has chosen otherwise, and Walmart may ask for extra confirmation even within the spending limit;
- the agent must not do more than the customer wants;
- the agent does not receive payment card details and cannot add or change payment methods;
- the agent sees only the information needed for the task and authorized by the customer;
- Walmart’s return policy applies even when the agent does not behave as expected;
- agents that do not comply with these terms are blocked.
Then there is a sentence about prices, aimed at a specific fear: that the price changes depending on who, or what, is buying. Walmart writes: “We price products, not people.” The prices an agent sees are the same ones a person sees on Walmart.com, with deals and costs shown before checkout.
A door already open, now with rules
Walmart is not starting from scratch. It already sells through its own assistant, Sparky, and through integrations with Muse, Meta’s agent, as well as ChatGPT and Gemini. The announcement puts rules on a door that was already open, and it does so just as agents are starting to buy for real.
A comparison with other large companies makes the choice clearer. Starting October 30, Meta gives itself, by contract, the right to read the websites of businesses that are on Facebook and Instagram, with no way to opt out. Walmart does the opposite: it decides how the agent gets in, what it can do, and who is responsible if it makes a mistake, and it makes the conditions public.
What online sellers can learn
The first lesson is that you can decide. Between blocking every agent and letting them do as they please, there is a third way: recognize them, give them a path and limits, and put everything in writing. Walmart’s rules are a checklist that any online seller can adapt: what the agent sees, what it can buy, who approves, and what happens if it makes a mistake.
The second is that agents need readable data. A site built for human eyes is hard for a model to use. Prices, availability, and product pages written so that a machine can understand them are the foundation, and a dedicated entry point comes later. Customer behavior points the same way: people who use AI before buying mostly use it to compare prices and summarize reviews.
The third is about trust. Payment networks are building the tools that let an agent pay safely; Walmart is building the ones that let it shop inside a store. A spending limit, approval, and guaranteed returns are what customers need before they delegate. Whoever offers them first will be ready when agents start bringing in customers, as has already happened to Target, whose traffic from AI platforms grew twentyfold in a year.
Frequently asked questions
They are profiles dedicated to AI agents inside a Walmart customer’s account, announced on October 21, 2026, and coming within a month. Each agent has its own profile, its purchases stay separate from the person’s, and the customer can approve every purchase or set a spending limit, such as $50, $100, or $200. The profile can be deleted at any time.
Only within the limits the customer chooses. By default, purchases must be approved by the customer; alternatively, the customer sets a spending limit within which the agent buys on its own, and Walmart may ask for extra confirmation. The agent never receives card details and cannot add or change payment methods.
No. Walmart states that the prices an agent sees are the same ones a person sees on Walmart.com, with deals and costs shown before checkout. The announcement puts it this way: “we price products, not people.” The price depends on the product, not on who is buying.
