Walmart Opens Its Doors to AI Shopping Agents: What the Agent Gateway Means for Your Online Store

Walmart has decided not to wait and see how AI shopping plays out. On 21 October the retailer announced Agent Profiles, the Walmart Agent Gateway and a set of terms for AI shopping agents, three tools that together let customers shop at Walmart and Sam's Club through an AI assistant of their choice, with spending limits and protections built in from the start.

The timing is not accidental. Amazon is currently blocking prominent outside agents from its store, Target is quietly letting approved providers in, and retailers everywhere are trying to decide whether an AI agent arriving with someone's shopping list is a customer or a threat. Walmart has picked its side, and the choice says a lot about where online retail is heading.

In this post we look at what Walmart actually launched, how the Agent Gateway works under the hood, and what all of this means for anyone who runs an online store. Let's sink in!

What Walmart Actually Announced

The announcement has three pieces. The first is Agent Profiles, expected to launch in November. Every authorized agent gets its own dedicated profile inside the customer's Walmart account, so purchases made by that agent sit separately from everything else the customer orders. Walmart compares it to adding an extra cardholder to a credit card: the agent can spend, but it is always clear who spent what, and the arrangement can be cancelled at any moment.

The customer stays in control of the money. They can require approval for every single purchase, or let the agent buy on its own within a set spending limit. Walmart's examples were 50, 100 and 200 dollars. Permissions can be switched off at any time, and the whole profile can be deleted with a couple of taps.

The second piece is the Agent Gateway itself, which opened as an invitation-only beta for companies and developers. The third is a formal set of terms for AI agents, which we will get to below. In its press release Walmart was refreshingly direct about the whole thing: "Shopping with an agent won't be for everyone, and it doesn't have to be." The goal, the company says, is to be the most trusted place for those who do want it.

How the Agent Gateway Works

Until now, an AI agent that wanted to buy something at Walmart had to drive a web browser the way a human does: load pages, find buttons, fill in forms. Walmart itself calls that approach slow and error-prone, and it is also expensive for the retailer, because agent traffic looks like scraping and gets treated as such.

The Gateway replaces all of that with a structured connection built on MCP, the Model Context Protocol, the same open standard that is quickly becoming the default way to plug AI models into external tools and data. Through the Gateway an approved agent receives product and fulfillment data directly from Walmart instead of scraping it from web pages, and can search inventory and place orders through Sparky, Walmart's own assistant layer.

The beta page lists what comes with that connection: verified agent identities, controlled permissions, observable actions, category restrictions, rate limits and compliance guardrails. One detail worth noting is privacy. Walmart says an agent can reorder a customer's usual items without ever receiving the customer's personal information or purchase history. The agent asks, the Gateway answers, and the data stays on Walmart's side.

Pricing stays boring on purpose. Agents see exactly the same prices a human would see on Walmart.com, with any offers and fees shown before purchase, and Walmart says its Every Day Low Price commitment is unchanged. There is no special machine-only price list, at least for now.

The Rules Agents Have to Follow

The terms for AI agents read like a code of conduct, and they are strict. An agent must identify itself as an agent and must not disguise automated activity as human browsing. It must follow rate limits. It cannot act beyond the shopper's actual intent, cannot receive payment card details, and cannot add or change payment methods on the account. It only gets access to the information a specific task requires and the customer allows.

Walmart also keeps a number of levers for itself. It can require extra approvals for purchases even when they fall within an agent's spending limit, it can cap transaction value, quantity or frequency, and it can decline or reverse transactions and suspend access entirely. Agents that break the terms get blocked. Helpfully for shoppers, the standard returns process applies even when an agent did not act as intended, so a rogue reorder of forty tins of beans can be undone the usual way.

There is a flip side for everyone else. Walmart's updated Terms of Use now explicitly prohibit unauthorized agents from scraping the site, placing orders or using its materials to train AI models without written consent. The front door is open, but the windows are being locked.

Open Doors at Walmart, Closed Doors at Amazon

What makes this launch interesting is the contrast with the rest of the market. Amazon has been blocking prominent third-party shopping agents from its store. Target is permitting a list of approved providers. Walmart has gone furthest in the opposite direction: it already has conversational shopping integrations with ChatGPT, Gemini and other assistants, and the Gateway now gives developers a sanctioned technical route in.

The logic is not hard to reconstruct. Agents are going to shop whether retailers like it or not, and a retailer that blocks them simply pushes the activity into scraping, where it has no identity, no telemetry and no rules. A controlled front door means Walmart knows which agent did what, can bill or block it, and keeps the customer relationship. Walmart said it wants to help shape what shopping becomes rather than wait to see.

Merchants in general are less prepared than they would like to be. Research from PYMNTS Intelligence published this autumn found that merchants broadly expect agentic commerce to become a real, paid sales channel, but many admit they have no plan for connecting with agents. The big retailers are now writing the playbook that everyone else will copy.

What AI Shopping Agents Mean If You Sell Online

If you run an online store, the immediate takeaway is that agentic traffic is becoming a legitimate channel rather than a nuisance. Within a year or two, a noticeable share of your visitors may be software acting for a person, arriving with a task, a budget and no patience for pop-ups. Those visitors will judge your store the way machines judge things: by structured data, accurate stock and pricing information, and a checkout that does not fight back.

The second takeaway is about expectations. Walmart has just taught millions of shoppers that an AI agent comes with a spending cap, purchase approvals and a separate purchase history. Customers who get used to those controls at Walmart will expect something similar wherever they shop, including from the agents that buy from your store.

The third is that you will probably not build any of this yourself, and you do not need to. TikTok's recent agentic checkout push was built with Stripe and Shopify, and platforms such as Wix, Shopify and BigCommerce are the natural place for Gateway-style rails to appear for ordinary merchants. What you can do today is make sure your store is ready when they ship.

How to Get Your Store Ready for Agentic Shopping

Start with your product data, because that is what agents actually read. Clean titles, honest descriptions, correct prices, live stock levels and proper schema.org markup turn your catalogue into something an agent can parse without guessing. If your product feed is an afterthought, fix that before anything else.

Then decide your policy. Walmart wrote explicit terms for agents; Meta recently updated its terms around crawling. You should know your own answer: do you welcome agent purchases, and under what conditions? Put the answer in your terms of use and your robots rules, so the decision is yours and not an accident.

Finally, think through the boring operational questions. How does a return work when an agent placed the order? Can your support team see that a purchase was agent-made? Does your analytics separate agent traffic from human traffic? None of this is difficult, but the merchants who think about it now will have a quiet advantage over those who meet their first AI shopper unprepared.

Final Notes

Walmart opening a gateway for AI shopping agents is one of those announcements that looks small and technical today and will look obvious in hindsight. The world's largest retailer has concluded that agents are customers, written house rules for them, and built them their own entrance. Amazon has, for the moment, concluded the opposite, which means the experiment is running in public and we will all see the results.

For most businesses the right response is neither panic nor a big project. Keep your product data clean, know your policy on agents, and lean on your platform for the heavy lifting. The shops that treated mobile traffic seriously in 2012 did rather well a few years later. Agentic traffic is shaping up to be the same kind of quiet deadline.