Walmart now gives AI agents their own profiles — with spending caps and separate purchase history
The retail giant launched technical infrastructure on Wednesday, October 21, that lets approved AI agents search inventory and complete purchases on customers' behalf — even as rival Amazon blocks exactly such agents.
Walmart went public on Wednesday, October 21, 2026, with three interconnected launches: Agent Gateway, a server based on the Model Context Protocol (MCP) that approved AI agents can connect to; Agent Profiles, a control mechanism that gives customers per-agent settings; and a set of terms governing how the agents behave. The announcement came from Walmart's president and CEO John Furner at the Masters of Scale Summit in San Francisco, according to TechCrunch.
The timing is worth noting. The launch comes shortly after Amazon blocked Meta's shopping agent Muse — less than two weeks after Muse launched — and some months after a federal appeals court in August vacated Amazon's preliminary injunction against Perplexity's Comet browser. While Amazon shuts the door, Walmart opens its own — but with significant caveats, and so far only in an invitation-only beta.
How Agent Gateway works
The Gateway is a platform that lets AI assistants such as OpenAI's ChatGPT, Google's Gemini, and Meta's Muse connect directly to Walmart's and Sam's Club's systems to search for products and complete purchases, writes Reuters via Kansas City Star. The technical underpinning is an MCP server, which according to TechCrunch is designed to help approved agents connect to Walmart over the Model Context Protocol. The technology supports agentic commerce for both Walmart and the company's low-price chain Sam's Club.
The Gateway is also meant to enable agents to purchase a customer's regular items — such as sourdough bread or a particular type of milk — without the customer's personal details or broader purchase history being shared with the agent, according to Unite.AI, which cites Walmart's announcement and beta page.
One practical detail shows why the control mechanisms are necessary: In the current setup, the agent signs in with the customer's own account. That means purchases made by the agent and purchases made by the customer appear together, with no form of differentiation, according to the announcement Unite.AI reproduces. The agent's purchase is thus, for now, indistinguishable from the customer's own orders in the purchase history.
Agent Profiles: separate visibility and spending caps
Agent Profiles, which is rolling out "next month" according to TechCrunch, is designed to solve exactly this. The customer will be able to give each AI agent its own dedicated profile in their account, so that the agent's purchases appear separately from the customer's own. In addition, the customer gets control over how purchases are approved: either by requiring approval for each individual transaction, or by authorizing transactions within a spending limit the customer sets themselves. A Walmart+ membership is not required to use the features.
Both models address the classic problem of agentic commerce: giving an agent enough authority to act usefully, without handing it a blank check over the customer's money and data.
The terms Walmart imposes
Walmart has also published its own terms for AI agents, according to TechCrunch:
- The agents may not do more than the shopper has asked — no actions beyond intent.
- The agents may not access the customer's payment information.
- The agents get access only to the information they need for the task at hand.
- Agents that fail to comply with the terms are blocked from access to Walmart's website.
Walmart further states that the returns process also applies when an agent has not behaved as expected. On prices, the Every Day Low Price commitment applies regardless: the prices an AI agent sees will be the same as those a human customer sees on Walmart.com, with relevant offers and fees shown before purchase.
It is worth noting that these are Walmart's own terms, reported through secondary sources — not a regulatory regime. Enforcement in practice consists of Walmart being able to deny the agents access, which gives the company full control over who gets in.
The contrast with Amazon
Walmart's open approach stands in contrast to Amazon's. Amazon blocked Muse from browsing and buying products on the platform less than two weeks after its launch, according to Axios via MSN.
The legal basis for blocking agents is disputed, however. In the Perplexity case, Amazon won a preliminary injunction in March 2026, according to Newsweek, which reviewed court documents — but in August a federal appeals court vacated the injunction on the grounds that it was not Perplexity using Amazon's website — it was their customers, with the agent as a helping tool.
The analysts see different motivations behind the two companies' strategies. Neil Saunders of GlobalData Retail tells Axios that Amazon is trying to preserve "its closed ecosystem," and likely fears that agentic commerce could undermine the advertising business. David Bellinger of Mizuho has, according to Reuters' reporting, pointed to tools of the kind Walmart is now launching as able to build confidence in agentic purchases.
Walmart itself points to the growth of its in-house assistant Sparky, which recently saw 60 percent quarter-to-quarter growth, according to David Guggina, described as head of Walmart U.S. in Axios' reporting, at a Goldman Sachs conference in September.
What remains open
There are so far no numbers on usage, transactions, or revenue through the new tools. Nor has it been documented how much of the existing traffic from Muse, ChatGPT, and Gemini actually flows through the new gateway, rather than through existing browser-based flows. TechCrunch's reporting suggests Walmart already had integrations with these assistants before the gateway launched.
The beta is moreover reserved for invited participants, and according to Unite.AI, citing the beta page, it can in principle be suspended at any time. Agent Profiles is rolling out "next month," but there is no more precise timeline.
The most important unanswered question, though, is whether Walmart's model works in practice: whether approved agents actually stay within the terms, whether customers use the control mechanisms, and whether Amazon — which now has to contend with both a court ruling that undermines its blocking strategy and a chief competitor opening the door — changes course. For now, the only certainty is that the two companies have chosen entirely opposite answers to the same question: Should AI agents be allowed to shop?