Walmart Opens Its Doors to AI Shopping Agents as Amazon Shuts Them Out

For most of the past year, the big question in agentic commerce was whether retailers would let AI agents in at all. This week Walmart gave its answer. On October 21 the retailer announced Agent Profiles, a Model Context Protocol (MCP) based Agent Gateway and a written set of terms that agents must follow on Walmart and Sam's Club. The move puts Walmart on the opposite side of the line from Amazon, which has been blocking third-party shopping agents. For enterprise leaders well outside retail, the more important story is the template: Walmart is treating agents as a new kind of customer-facing identity that needs its own account, its own limits and its own rules.

What did Walmart announce?

Walmart CEO John Furner presented the plans at the Masters of Scale Summit in San Francisco, TechCrunch reported. The package has three parts.

Agent Profiles, due within about a month, give each AI agent a dedicated profile inside the customer's Walmart account, so its purchases show up separately from the person's own. Customers can require approval for every order or let the agent buy within a spending limit, and they can delete the profile at any time. Walmart compared the setup to adding a cardholder to a credit card, according to The Next Web.

The Agent Gateway is a direct technical channel for agents, built on MCP, so they no longer have to click through the website like a person. It starts as a beta for companies and developers. Walmart says the gateway can help an agent find a customer's usual brands of bread or milk without handing over the customer's broader purchase history.

The terms set the boundaries. Purchases require customer approval by default, and Walmart may require more approvals even inside a limit. Agents cannot act beyond what the shopper intends, cannot see payment card details or change payment methods, and may use only the data a task needs. Walmart's normal returns process applies when an agent gets something wrong. Agents that break the rules will be blocked.

Why are Amazon and Walmart taking opposite approaches?

Amazon has moved in the other direction. In September it began blocking Meta's Muse agent from its store after Meta declined a request to remove it, Bloomberg reported in a story carried by The Star. An Amazon spokesperson argued that third-party apps buying for customers should operate openly and respect a business's choice about whether to participate, much as food delivery apps and travel agencies do. Amazon also sued Perplexity last year, accusing it of hiding its shopping agents after Amazon asked it to remove them.

The two positions reflect a real trade-off. Walmart itself named the problem with today's agents, as PPC Land summarized: they shop through the customer's own login, which mixes human and software purchases in one order history, and they navigate the site through a browser, which is slow and error-prone. Blocking them keeps the retailer in charge of the customer relationship, though Bloomberg noted that Amazon could lose sales and advertising revenue if shoppers start buying through agents in volume. Opening a controlled door, as Walmart is doing, accepts that agents will bring traffic anyway and tries to set the terms.

Walmart is not alone in drawing lines. The Next Web noted that Schwab and JPMorgan have reportedly blocked AI agents from moving customers' money. Walmart also faces critics: Lindsay Owens, who runs the Groundwork Collaborative, accused the company of "sneaky pricing practices," claims Walmart disputed as misleading, The Next Web reported.

Why is agent identity the hard part?

The week's other data point shows why Walmart built a gateway rather than simply waving agents through. Arkose Labs, a bot detection firm, published a report on October 20 finding that only 2.3% of legitimate AI agent sessions it observed carried a verified Web Bot Auth signature, PPC Land reported. The figures are vendor-reported and unaudited, but they point to the core problem. Most agents arriving at a website today cannot prove who they are or whom they represent.

That is the same problem enterprises face inside their own walls. "Traditional identity and access management says what you can access," Midships CEO Ajit Gupta said at The AI Enterprise Conference 2026, arguing that what is missing is control over whether an agent should be doing something at a given moment. His remedy for banks, checking each agent's intent and holding a named human accountable, maps closely onto Walmart's design, where spending caps, default approvals and a separate profile tie every agent action back to an account holder.

AI Connect's expert panel on where humans belong in the agent loop made a related point in August: give each agent scoped, least-privilege credentials rather than a shared account, and match oversight to the risk of the action. A $50 grocery reorder and a change of payment method are not the same decision, and Walmart's terms treat them differently.

What should enterprises outside retail take from this?

Walmart has been building toward this for some time. At COLLIDE 2025, Walmart's Sashi Gandavarapu described how the company anchors its generative AI work to four super agents for customers, associates, merchants and sellers, and developers. "There are no AI projects and non-AI projects anymore," he said. Opening a governed channel to outside agents is a logical next step for a company organized that way.

For other industries, the lesson is that external agents will show up whether or not a company plans for them. Snowplow's Nick Stanchenko told a AI Connect webinar earlier this year that simply blocking bots no longer works "because agents actually bring you users and they bring you revenue." In his session on how agents are breaking web analytics, he urged teams to identify and measure agent traffic before deciding how to treat it.

Practical steps follow from Walmart's model. Decide which customer journeys agents may complete and which they may not. Give agents a distinct identity in customer accounts, with limits the customer controls. Offer a structured interface, whether MCP or an API, so agents stop scraping pages. Publish terms and enforce them. And keep the data boundary tight, so an agent gets what a task requires and nothing more.

Several questions remain open, including which agent companies are in the gateway beta, how Walmart will detect agents that ignore its terms and whether spending limits apply per order or per period. The answers will shape whether Walmart's approach becomes a standard for agentic commerce or one retailer's experiment. Either way, the companies that define how agents identify themselves, and what they are allowed to do, will have more say over the customer relationship than those that only try to keep agents out.

Frequently asked questions

What is Walmart Agent Gateway?

It is a direct channel, built on the Model Context Protocol, that lets approved AI agents search and buy on Walmart without navigating the website like a person. It starts as a beta for companies and developers.

How do Walmart Agent Profiles work?

Each agent gets its own profile in the customer's account, with purchases shown separately. Customers can approve every purchase or set a spending limit and can delete the profile at any time.

Why does Amazon block AI shopping agents?

Amazon says third-party apps that buy for customers should operate openly and respect a business's choice to participate; it blocked Meta's Muse after Meta declined to remove it.

Filed under: Agentic AI, AI Governance & Trust, AI Use Cases & Applications