By This Hour AI Desk
Walmart is reportedly preparing to let approved artificial-intelligence agents shop on behalf of customers, placing the retailer on the more permissive side of an emerging fight over whether autonomous software should be allowed to browse, select and buy goods online.
The reported plan centers on an Agent Gateway that would connect approved agents to Walmart’s online systems. Those agents could search inventory and make purchases for a user, while customers would receive tools intended to define what each agent may spend and whether every order requires approval. If implemented as described, the arrangement would make Walmart and Sam’s Club available through a controlled route for agent-led shopping rather than treating such software as unwanted traffic.
The contrast matters because Amazon has reportedly begun blocking AI agents from making purchases for online shoppers. Other online services, including sites in travel and booking, have also been described as restricting agent access. Walmart’s reported approach does not amount to unrestricted entry: it would invite approved agents into a system with stated limits, customer permissions and the possibility of being blocked for noncompliance.
A gateway intended to distinguish approved agents from ordinary automated traffic
The centerpiece of the reported rollout is Walmart’s Agent Gateway, described as a server using the Model Context Protocol, or MCP. The gateway would give approved AI agents a designated connection to Walmart, enabling a defined set of shopping tasks rather than leaving agents to navigate the retailer’s website without an agreed technical arrangement.
In practical terms, the reported capability is straightforward but consequential. An agent could look through Walmart’s online inventory and complete a purchase for the person using it. The technology is also intended to support shopping at Sam’s Club, extending the model beyond Walmart’s main retail operation.
The report characterizes MCP as a protocol for connecting AI systems with external data, tools and workflows. Here, the relevant point is not the protocol’s broader technical role but the boundary it could establish. Walmart would be offering an approved means for an agent to interact with its shopping systems, while retaining authority over which agents can use that route and under what conditions.
That distinction is central to the retail strategy described in the report. A company can choose to resist automated purchasing altogether, or it can construct an access point designed for software acting on a customer’s instructions. Walmart appears to be pursuing the latter course, but with a framework intended to limit both the information agents can obtain and the actions they can take.
The reported gateway also aims to let customers shop for familiar products or brands without directly sharing their personal information or purchase history with an agent. That is an important design claim, because an agent helping with repeat shopping might otherwise need substantial knowledge of a person’s past behavior. The account does not spell out the technical method by which this separation would work, nor does it specify precisely what information would be available to an agent during a transaction.
Profiles would put permissions and spending controls in the customer account
Walmart is also reported to be introducing Agent Profiles, a feature designed to separate purchases made by AI agents from purchases a customer initiates directly. Each agent would receive its own profile within the customer’s Walmart account, allowing the shopper to identify activity associated with that particular agent.
Under the reported model, a customer would first authorize and link the profile when using an agent through the Agent Gateway. The shopper could then decide how purchases should be approved. One option would require approval for every transaction. Another would allow the agent to make purchases within a spending limit set by the customer.
Those choices address a basic tension in agent-led commerce. A system that asks the customer to approve every order preserves close oversight, but it reduces the convenience of delegating routine shopping. A preset limit may make the agent more useful for repeated tasks, yet it requires the customer to trust that the agent will make selections consistent with the instructions it has received. The proposed controls do not remove that trade-off; they give the customer a way to decide where to place it.
The separate record of agent purchases could matter just as much as the spending limit. If an order is placed by software rather than by the account holder, the customer needs to be able to recognize that distinction when reviewing activity. The report says Walmart has developed technology to differentiate these transactions. It does not provide details on how that identification would appear, how quickly it would be updated, or whether it would cover every type of purchase available through the gateway.
Customers would reportedly be able to delete an Agent Profile at any time. The report further says access would be available to anyone with a Walmart account and would not require a Walmart+ membership. That suggests the feature is being positioned as an account capability rather than a benefit reserved for the retailer’s membership program, although the supplied account does not describe any rollout sequence, geographic scope or eligible agents.
Walmart’s welcome comes with rules on access, data and conduct
Calling the approach an opening to AI agents should not obscure its restrictions. Walmart is reported to have created terms specifically for agents, setting out how they may operate and what rights they have when transacting on the retailer’s site. Agents that fail to comply could be blocked from accessing Walmart.
The reported terms would prohibit agents from taking action beyond the shopper’s intent. They would also bar them from accessing a customer’s payment information and limit them to information needed for a particular task. Together, those limits describe an attempt to keep the agent’s role narrow: it may assist with a defined shopping job, but it should not gain a broader mandate over the account or a wider view of customer information.
Such commitments will be tested in the ordinary ambiguities of shopping. A customer might ask an agent to find a usual item, but a product could be unavailable, offered in different sizes, or sold in a range of versions. The supplied report does not explain how the system would determine whether a substitution, a different quantity or a different price still falls within the shopper’s intended instruction. Nor does it describe how disputes over those judgments would be resolved.
The returns policy is one answer to the risk of an unwanted outcome. Walmart reportedly says its existing returns process would apply when an agent does not act as intended. That gives customers a stated route after an order has been made, rather than making them bear the full cost of a mistaken purchase simply because software initiated it. But the report does not detail whether any special process would apply to agent-caused errors or how Walmart would assess whether an agent had departed from a customer’s instructions.
The most important unanswered operational question is therefore not whether an agent can submit an order. It is how responsibility will work when the result is unsatisfactory. Customer approval settings, transaction records and returns may each help, but the supplied information does not establish how those elements will function in difficult cases.
A different response to the battle over who controls online transactions
The reported Walmart initiative arrives amid a broader argument over access to commercial websites. An AI agent can be framed as a customer’s delegate, carrying out searches and purchases that the customer could have performed personally. A retailer can also view it as automated software requiring safeguards, especially when it seeks access to inventory, account functions and transactions.
Amazon’s reported move to block AI agents from making purchases illustrates the opposing choice. The available account does not provide Amazon’s explanation, the mechanics of its restrictions or the exact scope of agents affected. It supports only the narrower conclusion that Amazon and Walmart are being presented as taking different paths: one restricting agent purchases, the other building a controlled channel for approved agents.
That disagreement reaches beyond a question of convenience. If AI agents become a regular interface between shoppers and retailers, the company that owns the storefront will still decide the terms on which those agents enter, what they can see and which transactions they can complete. Walmart’s proposed gateway would preserve that retailer control even while enabling agent participation.
Walmart had already announced a partnership with Meta’s Muse and integrations with AI applications including ChatGPT and Gemini, the report says. The new gateway and profiles would extend that direction from integration toward a more specific purchase-and-permission framework. Yet the supplied material does not say which agents will be approved, when all features will be available, or how broadly customers can use them.
The report has not been independently corroborated. It is based on a single supplied account, and Walmart’s detailed implementation, eligibility rules, technical safeguards and handling of disputes cannot be confirmed from the information available here.
The consumer promise depends on controls working in practice
For shoppers, the reported appeal is delegation without surrendering the account. An agent could handle a defined shopping task, while the customer retains a visible record, transaction approval or a spending ceiling, and an option to remove the agent’s profile. For Walmart, the appeal is a way to accommodate agentic shopping without offering open-ended access to its systems.
Whether that balance holds will depend on details absent from the report. The account does not establish how agents will verify a user’s intentions, how a customer changes instructions after an agent has begun a task, how limits interact with shipping or other transaction components, or what happens when an agent encounters incomplete or conflicting information. It also does not establish whether the features are available immediately or whether any announced timing applies to the whole package.
What is clear from the report is the direction of travel: Walmart is described as treating AI shopping agents as participants that can be admitted through rules, rather than as software to be excluded categorically. The success of that approach will hinge less on the existence of a gateway than on whether its permissions, spending boundaries, data limits and remedies give customers confidence that an agent is acting for them—and no further.
For further context on this subject, see Websites put new limits on AI agents as access disputes widen.
Reporting notes
What is confirmed: The report describes separate agent profiles, purchase approval settings, spending limits and terms that can lead to blocked access.
Why this matters: The reported plan offers a controlled alternative to Amazon’s reported blocking of agent-led purchases.
What remains unclear: Availability, eligible agents, technical safeguards and procedures for disputed agent actions are not established by the supplied account. This report is based on one source and has not been independently corroborated.
