Walmart Gave AI Agents a Spending Cap and a Login — Which Makes the Retailer, Not the Model Lab, the Owner of the Transaction

Three objects, one staged rollout. The agent terms and the rules of engagement are live now; the per-agent profiles arrive within a month of the announcement; the gateway itself opens only as a limited beta for companies and developers, with no date and no partner list named. Reuters reported that the gateway is designed to connect assistants including OpenAI's ChatGPT, Google's Gemini and Meta's Muse to Walmart and Sam's Club systems. What Walmart has not disclosed is the commercial half of the deal — no fee schedule, no partner list, no volume target, and no statement of how agent-placed orders will be attributed in its ad products.

Amazon's objections to AI shopping agents were specific: nobody consented, the agent did not identify itself, and the agent appeared to capture customer credentials. Walmart's design answers each one with a control rather than an argument — a customer-created profile for consent and identification, and a payment instrument the agent never sees. What remains is the one thing no platform could previously offer a customer: a bounded mandate to spend.

This is the layer that decides who wins. Meta's advertising revenue reached $63.1 billion in Q3 2026, up 26%, with the online commerce vertical the largest contributor to the increase — commerce ad money is being earned inside conversations no retailer owns. Walmart's advertising business grew 38% in the June quarter, with Walmart Connect up 43% excluding VIZIO, and it attaches to transactions Walmart can see from search to doorstep. Agent-mediated shopping only pays a retailer if the retailer keeps the receipt, because the receipt is what the ad gets priced against.

Brands sit on the wrong side of that trade. A paid placement inside an assistant that is meant to represent the customer is hard to disclose and harder to defend, which means the retailer's ad inventory, not the platform's feed, is the only place a sponsored result can survive. Walmart has not said whether agent sessions will carry sponsored inventory at all, and has not told brands what an agent-placed order is worth inside a media plan.

The number that matters is not the cap; it is the conversion gap. Walmart's own pilot data, cited in October, puts direct agent checkouts at roughly one-third the conversion rate of click-outs — and because the agent is capped at $50 to $200, the basket it places is smaller still. That is why the retailer, not the model, has to hold the transaction: a third of the value per session is not worth handing the customer relationship to a platform that will bill for the impression.

Walmart's own numbers show where agent demand would land. In the June quarter, Walmart U.S. operating income rose 20.6% to $8.12 billion, Sam's Club U.S. operating income rose to $678 million from $470 million, consolidated operating income rose 28.8% to $9.38 billion, and gross margin widened to 26.1% from 25.2%. Membership and other income reached $1.84 billion against $1.65 billion. The mix that is compounding is small-basket replenishment, store-fulfilled delivery and advertising, and the basket a $50-to-$200 cap produces lands in the highest-margin part of that mix.

Amazon cut Meta's Muse agent off Amazon.com in September, telling shoppers that continued access by an unauthorized AI agent violates the conditions they had accepted. Amazon's stated reasons were precise: it never agreed to the access, the agent did not identify itself, and the agent appeared to capture and store customer credentials. Walmart's design answers each objection with a control. Amazon keeps the ad funnel and cedes the agentic surface, while Walmart trades a slice of catalog control for the chance to own the transaction.

Retailers can approve agents, but they cannot authenticate them alone. That work is migrating to the card networks and to the checkout layers that already hold tokenized credentials, which makes it the one part of the stack that does not require either side to trust the other.

The retailer has made a decision brands cannot ignore: an agent that finds a product may never visit the brand's site, and the order will sit in the retailer's file, priced against the retailer's ad inventory. Walmart has already told sellers that listings need stronger product information, competitive pricing, availability and reliable delivery promises before agents will recommend them.

Walmart is not the only retailer building agent rails, but it is building a different one. Alibaba connected its Qwen assistant to Taobao's entire product catalog and launched a Qwen-powered shopping assistant inside the Taobao app. Rakuten launched Rakuten AI inside the Rakuten Ichiba mobile app on January 5, 2026. Both keep the agent on the retailer's own property and inside the retailer's own data; Walmart is the first large retailer to invite outside agents and hand them buying power inside the customer's account.

That contrast is the investment signal. If the open model works, the platforms that kept the agent in-house eventually face a choice between opening a budget-bearing gateway to outside assistants and watching agent traffic settle on whichever retailer did it first. If it does not — and a one-third conversion rate is not encouraging — they have avoided an expensive rollout that brands would have fought, while Walmart has spent a quarter of management attention on a rail with no revenue attached to it yet.