As of October 27, 2026, the clearest signal about agentic commerce comes from a retailer that chose to write the rules before handing over the cart. Walmart said on October 21 that AI agents shopping for its customers will get their own profiles, preset spending ceilings and a machine-readable gateway into Walmart and Sam's Club, a combination that turns shelf availability and price parity into hard requirements rather than reporting hygiene.
What Walmart put on the table on October 21
The announcement bundles three moving parts. Agent Profiles give every authorised agent a separate identity inside a customer account, so its orders no longer dissolve into the household's own order history. The Agent Gateway gives approved agents a direct machine-to-machine route instead of a browser session. The third piece, the terms agents must follow, decides what an agent may see, buy and touchDigital Commerce 360.
Profiles are due within the next month. A shopper can demand approval for every single purchase, or let an agent spend up to a ceiling, with 50, 100 and 200 dollars offered as illustrations rather than fixed tiers. Setup runs through three steps: the profile is created the first time a shopper buys through the gateway, the shopper signs in to link it, and the shopper then sets the limit, reviews agent orders and can delete the profile at any moment.
The gateway swaps browser clicks for a protocol
What the gateway actually exposes
Today an agent shops by driving a browser through pages built for human hands, which Walmart calls slow and error-prone. The gateway relies on Model Context Protocol, the standard Anthropic developed for connecting AI applications to outside tools, and it covers discovery, add to cart, purchasing and order tracking. It opens as a limited beta for companies and developers, with no date announced for general release.
The gateway also settles a quieter problem: repeat purchase. Walmart says the connection can help an agent locate the brand of milk or the loaf of bread a household usually buys without handing over the wider shopping history. That is a narrow lookup, yet it is the part that determines whether an agent basket resembles a household basket or a random sample of the category.
Where the shelf meets the bot
Two sentences in the announcement matter more to an operations team than the feature list. The first is that agents see the same prices a human sees, with offers and fees displayed before purchase. The second is that agents breaking the terms get blocked from the site. Together they promote price consistency and item-file accuracy from back-office hygiene into a commercial requirement that shows up in conversionDigital Commerce 360.
An agent cannot walk an aisle. It buys whatever the item file says is sellable, at whatever price the system returns. If a promoted SKU is out of stock in a fulfilment node but still flagged as available, the agent will order it, the substitution will land in a picker's queue, and the return will be booked against a store that never saw the demand coming. Availability stops being a shelf question and becomes a data question.
| Control | What it constrains | Effect on a store network |
|---|---|---|
| Agent Profiles | Who may buy, and up to which ceiling | Basket size becomes a permission rather than a forecast |
| Price parity clause | What the agent sees at checkout | Any gap between shelf price and app price becomes a visible error |
| Usuals lookup | Which repeat items an agent can find | Core SKUs gain weight against promotional SKUs |
| Returns policy | What happens after a wrong order | Substitutions and damage land back in store operations |
Shoppers want a brake, not a chauffeur
The 11 / 70 / 10 split
Walmart's caution is easy to explain. In survey results published in August, payments technology provider NMI asked 1,000 U.S. adults about AI shopping. Only 11% had used AI to complete a transaction, 70% wanted the ability to review or override an AI purchasing decision before checkout, and just 10% would hand over full control. Per-purchase approval as the default is a direct answer to that distribution.
For brands, the practical reading is that agent traffic will arrive in small, high-intent bursts rather than as a flood. Orders that do arrive carry a full instruction set behind them, which makes them unusually clean evidence about which SKUs shoppers genuinely prefer once they delegate the choice. That is a different measurement problem from browsing behaviour, and it rewards teams that already keep product content, stock signals and price in syncDigital Commerce 360.
FAQ
Does an agent purchase change how stockouts should be measured?
Yes. An agent converts a data error into an order within seconds, so out of stock has to be measured on the item file rather than on the shelf. Teams that only audit physical availability will keep discovering stockouts through cancellations, which is the most expensive possible detection method.
Why does price parity matter more once agents buy?
Because an agent compares what the system returns, not what a shelf label says. A price gap between app and store stops being an occasional customer complaint and becomes a systematic error that scales with every agent order. Price monitoring therefore has to run on the same feed the agent reads.
What should a store network check first?
Item-file accuracy for the SKUs that carry repeat purchase. Those are the items an agent is most likely to look for on a customer's behalf, and they are also the items where a wrong availability flag produces the fastest substitution, the fastest refund and the least recoverable margin.
