Most big retailers have treated AI shopping agents like a nuisance: something to rate-limit, block or ignore. On October 21, 2026, Walmart went the other way. CEO John Furner announced Agent Gateway and Agent Profiles at the Masters of Scale Summit, a pair of tools that let approved third-party AI agents search Walmart and Sam's Club inventory and, with the shopper's permission, buy things on their behalf (Source: AI Weekly).
The interesting part is not that an agent can click Buy. It is the rulebook around it: a separate profile per agent, spending limits the shopper picks, approval prompts, and a published set of terms agents must follow. This post explains how the system is described, which parts are still beta or undisclosed, and what it signals about where AI shopping is heading.
Quick summary: Announced October 21, 2026, Walmart Agent Gateway is an MCP-based connection point that lets approved agents shop Walmart and Sam's Club. Agent Profiles give each agent its own identity and purchase history inside a shopper's account, with approve-every-purchase or spend-within-a-limit modes (example limits: $50, $100, $200). Agents never receive card details, and shoppers can delete a profile at any time. The Gateway starts as a beta for companies and developers, and Agent Profiles are described as launching within about a month. Fees and the list of supported agents were not disclosed.
What Agent Gateway and Agent Profiles are
Agent Gateway is a dedicated connection point built on MCP (Model Context Protocol), the open standard that lets AI agents talk to outside tools. It is meant to replace slower, error-prone browser-based shopping, where an agent drives a website like a person would. According to one write-up it supports product discovery, add to cart, purchasing and order tracking, and it begins as a beta for companies and developers (Source: Stellagent). A second report describes it as a route to Sparky, Walmart's own shopping assistant, and says agents work through Sparky rather than touching the catalog and checkout directly (Source: eesel AI).
Agent Profiles are the shopper-facing half. Each agent gets its own profile in the customer's Walmart account, so what it buys is tracked separately from the shopper's own orders. The first time a customer shops through an agent using the Gateway, Walmart creates a profile automatically, the customer signs in to authorize it, and then sets a limit (Source: Stellagent).
The controls: caps, approvals and a human gate
Shoppers choose one of two modes: approve every purchase, or let the agent buy within a spending limit. Walmart's example limits are $50, $100 and $200 (Source: Stellagent). Agents never receive card information, profiles can be deleted at any time, and per TechCrunch's reporting as relayed by Stellagent, the feature launches within roughly a month and does not require a paid Walmart+ membership (Source: Stellagent).
The eesel write-up goes deeper on how a purchase flows, based on Walmart's announcement, beta page and terms rather than published API documentation, so treat the mechanics as that author's reading. In that account, Walmart checks a user-defined guardrail on its own servers (per-transaction spend, category limits and purchase frequency), checkout uses a short-lived token tied to one merchant, cart, amount and session, and an order that falls outside the limits triggers a Human Consent Gate that asks the shopper instead of simply failing (Source: eesel AI).
| Control | How it is described | Source |
|---|---|---|
| Approval modes | Approve every purchase, or buy within a limit ($50, $100, $200 examples) | Stellagent |
| Card details | Agents never receive them | Stellagent, eesel AI |
| Profile removal | Shopper can delete a profile at any time | Stellagent, eesel AI |
| Browser agents vs Gateway agents | Browser agents need consent before every checkout; Gateway agents may check out within limits | eesel AI |
| Rule-breakers | Non-compliant agents get blocked | AI Weekly, Stellagent |
The terms agents must accept
Walmart also published Terms for AI Agents. In eesel's summary, agents must identify themselves and their company on request, may not pretend to be human, must not act beyond what the shopper intends, and must label substitutions and recommendations. They may receive only the information a task needs, and bulk scraping or using Walmart data to train models or target ads is prohibited. The same summary says Walmart states it is not liable for decisions made by an agent, and reports a $50 liability cap that we could not confirm elsewhere (Source: eesel AI). Read these points carefully: they are one outlet's reading of the legal terms, so check Walmart's own document before relying on it.
What is still unclear
- Which agents will support Agent Profiles at launch has not been announced, and Walmart's announcement reportedly names existing integrations with several assistants without saying which use the Gateway.
- The Gateway beta's start date, participants and any fees were not disclosed.
- Walmart may require extra approval even inside a spending limit, but the conditions were not disclosed.
- Public developer docs, a tool list and a general-availability date had not been published as of the eesel write-up's October 28 edit.
- Critics argue the design lets Walmart keep the purchase data, recommendations and ad economics because outside agents route through Sparky. That is an analyst's view, not a Walmart statement.
Why this matters beyond Walmart
AI shopping has had a trust problem on both sides. Shoppers worry an agent will buy the wrong thing at the wrong price. Retailers worry about bots scraping prices, ignoring terms or placing orders nobody can trace. Walmart's design answers both with the same ingredients: identity (each agent is a named profile), limits (a ceiling the shopper sets), consent (a gate that cannot be waived by an agent's own instructions, per eesel) and an exit (delete the profile). Whether other retailers copy this or keep blocking agents will shape how useful shopping assistants become.
There is also a practical lesson for any agent you use today, shopping or otherwise. Treat spending authority like a corporate card: start with a low limit, review what the agent bought, and widen permission only after it has earned it. The profile-and-cap pattern Walmart describes is a good template even outside Walmart.
If you try any shopping agent, begin with approve-every-purchase mode or the lowest cap on offer, and check the order history after the first few buys before raising the limit.
FAQ
01What did Walmart announce?
On October 21, 2026, Walmart announced Agent Gateway, an MCP-based connection for approved AI agents, and Agent Profiles, which give each agent its own profile and spending controls inside a shopper's account.
02Can an AI agent spend any amount of my money?
No. You choose either to approve every purchase or to let the agent buy within a limit. Walmart's example limits are $50, $100 and $200. Agents never receive your card details.
03Do I need Walmart+?
Per TechCrunch's reporting as relayed by Stellagent, Agent Profiles do not require Walmart+.
04When can I use it?
Agent Gateway starts as a beta for companies and developers. Agent Profiles are described as launching within about a month, but Walmart has not said which agents will support them.
05What if an agent buys the wrong thing?
Per Stellagent and eesel, Walmart's regular returns process applies, including when an agent does not act as intended. You can also delete the agent's profile at any time.
Shopping agents are only as useful as the model behind them, and the best model differs by task. With LumiChats you get 40+ premium AI models in a single subscription, so you can compare how different assistants handle a product search or a price check before you trust any of them with a budget.
