As of October 27, 2026, Walmart has moved further into agentic commerce with three updates announced on Oct. 21: Agent Profiles that let shoppers authorise an AI agent and set spending limits, an Agent Gateway built on Model Context Protocol that supports product discovery, add to cart, purchasing and order tracking, and new purchasing rules that bar agents from acting beyond a shopper's instructions or receiving payment-card details. The retailer says agent purchases will require customer approval by default. For marketplace sellers, the more consequential change is not the checkout flow but the pressure it puts on product data quality.
1. What Walmart actually announced
The three pieces fit together as a permission system rather than a new storefront. Agent Profiles give the shopper a way to authorise an agent, set a ceiling and see which orders came from a machine. The Agent Gateway gives the agent a direct connection instead of forcing it to click through pages designed for people. The purchasing rules define what the agent may not do, including exceeding instructions, touching unauthorised information or holding card details.
Walmart frames this as a trust play. Daniel Danker, executive vice president of AI acceleration, product and design, said the goal is to make the experience the most trusted one for shoppers who choose to delegate. The framing matters because adoption is still thin: in an NMI survey of 1,000 U.S. adults published in August, only 11% had used AI to complete a transaction, while 70% wanted the ability to review or override an AI purchase before checkout.
2. Why spending limits are the real feature
A spending cap sounds like a minor control, but it is the mechanism that makes delegation acceptable. Shoppers can approve each purchase individually or allow an agent to spend up to a preset limit, with Walmart noting it may still require additional approval. That combination, per-transaction review plus a ceiling, converts an open-ended mandate into a bounded one.
| Control | What it prevents | Seller implication | How to check |
|---|---|---|---|
| Spending limit | Runaway basket size | Bundles must fit under common caps | Test basket value against $50 to $200 |
| Per-purchase approval | Unwanted substitutions | Listing accuracy decides selection | Compare listing to shipped item |
| Order attribution | Agents hidden among human orders | Agent orders become measurable | Segment orders by profile |
| Agent restrictions | Access to payment and account data | No workaround via agent messaging | Review agent-facing content |
The table points to a practical consequence: basket economics change. If a meaningful share of agent purchases happens under a $50 or $100 cap, multi-pack offers priced above those thresholds may lose visibility in agent-mediated sessions. Sellers should test how their catalogue looks when an agent is optimising for a basket under a fixed ceiling, not just how it looks to a human browsing a category page.
Approval by default changes conversion logic
When every agent purchase needs a human tap, the agent becomes a recommender rather than a buyer. That shifts the competition from checkout speed to the quality of the recommendation itself: the product an agent proposes is the product most likely to be approved. Clear titles, accurate attributes and honest availability data become conversion levers rather than housekeeping.
3. The gateway changes how agents shop
Today most agents shop through a browser, following steps designed for people. Walmart describes that as slow and error-prone, with agents getting stuck along the way. The Agent Gateway, built on the Model Context Protocol that Anthropic developed, gives agents a structured connection to discovery, cart, purchasing and tracking instead of a visual interface they must interpret.
Walmart also says the gateway can help an agent find products a customer regularly buys without exposing the broader shopping history, citing the example of a preferred brand of milk or bread. That is a narrow data permission, and it is the kind of design choice that determines whether shoppers trust agents with repeat purchases. The gateway will launch in a limited beta for companies and developers, with timing and participants unannounced.
Product data becomes the interface
Once agents connect through a structured gateway rather than a rendered page, the attributes a seller provides stop being decoration and start being the interface. Missing size, material or compatibility fields do not just look untidy, they remove the product from consideration. Sellers who treat attribute completion as a compliance chore will be invisible in exactly the sessions that are growing fastest.
4. What it means for marketplace sellers
Walmart's own marketplace leadership has been explicit about the upside and the cost. Manish Joneja, senior vice president of Walmart Global Marketplace and Walmart Fulfillment Services, said agentic commerce could put the right products in front of the right customers, including brands without broad awareness. He also said accurate product information, competitive prices, availability and reliable delivery promises will all become more important.
That combination describes a shift in what gets rewarded. Historically, marketplace ranking leaned on sales velocity and promotional spend. In agent-mediated sessions, the deciding factors are closer to structured data quality and fulfilment reliability, because those are what an agent can verify before recommending. It is an early example of a broader consumer trend: the machine-readable catalogue is becoming the storefront, and consumer demand is shifting toward sellers who can document their products clearly.
5. A 30-day checklist for sellers
The controls are announced but not fully live, which gives sellers a window to prepare. The work should start with the catalogue rather than with promotions, because promotions cannot fix a listing that an agent cannot parse. A month is enough to audit attributes, test basket sizes and rewrite the content an agent is most likely to read.
One caution on scale: agentic AI experiences remain a small slice of retail traffic, and Walmart has not disclosed how many shoppers will opt into profiles. The reason to prepare now is not that agent purchases dominate today, but that the requirements they impose, structured data, honest availability and reliable delivery, are the same requirements that improve human shopping. Sellers who fix them for agents end up fixing them for everyone.
FAQ
Do shoppers have to let agents buy without review?
No. Walmart says agent purchases require customer approval by default. Shoppers can approve each purchase individually or set a spending limit such as $50, $100 or $200, and the company says it may still require additional approval in some cases.
What can an agent not do under Walmart's rules?
Agents cannot act beyond a shopper's instructions, access unauthorised information or receive payment-card details, and they cannot add or change payment methods. Agents that violate the terms are blocked from the site, and unintended purchases follow the existing returns process.
Why does this matter for sellers outside Walmart?
Because the underlying shift is platform-neutral. Once agents read structured attributes instead of rendered pages, catalogue quality and fulfilment reliability become the deciding factors on any marketplace that opens an agent gateway.
Is agentic commerce big enough to plan for?
Not yet in volume. Agent experiences remain a small share of traffic, but referrals that arrive through AI channels carry higher buyer intent and average order value, which is why the data groundwork is worth doing before volumes grow.
