Walmart’s October 21 announcement puts a boundary question ahead of the usual promise of frictionless shopping: how much should an AI agent be allowed to do before a customer must take the decision back? The retailer says its planned Agent Profiles will separate agent-made purchases from a shopper’s other orders, while letting the customer approve each purchase or set a spending limit such as $50, $100 or $200. A company beta of its Agent Gateway is also planned, but the announcement does not say which agents will support profiles at launch or when the beta will open broadly. For O2O retail, the issue is not just whether an agent can find a product; the store, listing, price and customer approval must still describe the same offer. Digital Commerce 360: Walmart’s AI-agent shopping controls
Permission is a product feature, not a checkout footnote
Walmart’s official description says customers will authorize a dedicated profile for each agent, review those purchases separately and delete a profile when they choose. Purchases require approval by default; a shopper may instead allow transactions within a chosen limit, and Walmart may still request additional approval. The rules also prohibit agents from changing payment methods or receiving card details, and restrict access to information needed for the task. Those details make permission part of the shopping proposition itself: an agent that saves time but obscures what it selected or how much it may spend has not removed friction so much as moved it to the moment of trust. Walmart: Agent Profiles and customer controls
Consumer evidence suggests that assistance and delegation should not be treated as synonyms. NMI’s survey of 1,000 U.S. adults found that 11% had used AI to complete a transaction, 70% wanted the ability to review or override an AI decision before purchase, and 45% said they would not trust AI to buy on their behalf at all. These figures describe that survey’s U.S. sample and August 2026 research, not every market or every shopping category. They nevertheless identify a useful design test: can a customer understand the recommendation, correct a mismatch and retain a clear final say? NMI: consumer research on AI shopping and control
O2O reliability depends on the offer surviving a channel handoff
For a brand operating across ecommerce and stores, an assisted journey can fail even when the recommendation is sensible. A product title may refer to one pack size while the store carries another; a listed price may not match the offer shown at checkout; or a location may appear able to fulfil an order when the item is not available. These are not claims about Walmart’s implementation. They are operational questions for retail trend analysis: keep product identity, current price, store availability and substitution rules consistent, then make the approval point visible whenever the agent’s choice changes one of them. Price monitoring and on-shelf availability checks answer different questions and should not be collapsed into a single readiness score.
A physical store also retains roles that are not captured by a digital permission screen. Northgate’s newly converted La Habra grocery store combines everyday groceries with prepared-food counters, artisan goods and gathering spaces, and the report notes that more than 340 associates work there. That is a distinct retailer’s format, not evidence about AI shopping adoption. It is a reminder that service, local assortment and the ability to answer a customer’s question remain part of the O2O promise when discovery begins elsewhere. A pilot should specify which store task an agent is meant to support and how staff can correct a wrong item, price or fulfilment assumption. The Shelby Report: Northgate’s La Habra store format
Measure trust and execution separately
A practical test can begin with a narrow group of repeat-purchase products and a fixed set of customer requests. Record whether the agent identifies the correct SKU, presents the applicable price and fees, finds a store that can actually fulfil the item, asks for approval at the promised threshold, and makes the resulting order distinguishable from a customer-placed purchase. Track correction rates, unavailable-item substitutions, approval abandonment and customer complaints alongside completed orders; none of these indicators alone proves that agent shopping is preferable. Compare the same products through a direct digital route and a store-supported route, while keeping the observation window and customer permissions consistent.
The announcement establishes a set of Walmart controls and a proposed gateway, not a measured lift in conversion, customer satisfaction or store productivity. The survey identifies consumer reservations, not a forecast of how quickly they will disappear. As an industry trend analysis, the announcement is a prompt to test agentic commerce as a product and operating hypothesis, not to treat it as an inevitable replacement for search or human choice. For BXT-style O2O industry trend analysis, the decision is whether accurate product data, channel availability and customer review points can work together in a measurable pilot; the evidence should decide whether expansion is justified.
FAQ
What can a Walmart customer control in an Agent Profile?
Walmart says customers can authorize an agent, review its purchases separately, require approval for each purchase or set a spending limit, and delete the profile. The launch timing is described as within the next month, not as a completed general release.
Does the NMI survey show that most shoppers reject AI assistance?
No. It separates assistance from purchase authority: the survey reports interest in AI shopping help alongside low transaction completion and strong demand for review or override rights. Its 1,000 respondents were U.S. adults.
Which O2O checks should a retailer run before a pilot?
Check SKU identity, displayed price and fees, store-level availability, substitution handling and the customer’s approval step across the chosen digital and physical routes. The sources do not establish that these checks have already improved Walmart’s results.
