Walmart Opens Its Shelves to AI Agents While Amazon Builds Walls: The E-Commerce Agent Battle

On October 21, Walmart announced the launch of an "agent gateway" that allows AI agents to search its products and complete checkout. This marks the first time a major retailer has systematically opened its doors to AI shopping agents. Amazon, by contrast, has already confirmed to media outlets that it chooses to block such agents. These two starkly different stances are turning "who presses the buy button on the user's behalf" into the next watershed moment for the e-commerce industry.

Walmart's Approach: Putting Agents on a Leash, Not Locking the Door

Walmart's core design isn't complicated: every AI agent must create a profile, similar to an authorized supplementary card. In the default mode, each purchase requires the user's individual approval; users can also set a spending limit, allowing the agent to shop autonomously within that budget.

The key to this mechanism is keeping control in the user's hands—you retain the final decision, while the agent does the legwork. Walmart explicitly stated in its announcement that a "tethered agent" is better than an "agent locked outside the door." In other words, rather than outright rejecting agents like Amazon, it's better to bring them into your own transaction system through rules.

From a business logic perspective, this is actually a form of defensive openness. If AI agents become the primary entry point for users' shopping and Walmart is absent, it stands to lose an entire generation of traffic. By proactively setting the rules, it can at least participate in defining how the game is played.

Agent Gateway is an intermediary layer between AI agents and e-commerce platform backend systems. It handles agent identity authentication, request authorization and rate limiting, and logs every search and transaction. Unlike traditional open APIs, an agent gateway typically requires agents to pre-register and bind to a user identity, making "who is shopping on whose behalf" traceable. Walmart's agent gateway does exactly this by creating an authorization profile for each agent, linking the user, the agent, and the payment credential together, enabling the platform to retain audit and risk control capabilities while allowing agent traffic through. This design borrows from the "tokenization" mechanism used in the payments industry—the platform doesn't directly expose the user's payment information to the agent but instead issues a virtual authorization card with a limit that can be revoked.

Amazon's Choice: No Confirmation, No Cooperation, No Openness

Compared to Walmart's high-profile approach, Amazon's attitude is a silent blockade. According to The Verge, Amazon confirmed to reporters that it has blocked such AI agents but offered no further explanation.

There is a legitimate business concern here: automated agent ordering brings a range of operational issues such as higher return rates, inventory fluctuations, and fraud risks, and the platform may also lose control over product recommendations and pricing displays. But the cost of blocking is equally obvious—when users are accustomed to having agents compare prices and place orders for them, blocked stores will simply disappear from the candidate list.

The divergence between the two giants is essentially two answers to the same question: Is agent shopping a threat, or a new channel that must be secured? Walmart is betting on the latter; Amazon is temporarily betting on the former.

AI shopping agent blocking is typically implemented through various technical means: identifying agent traffic via User-Agent signatures, enabling CAPTCHAs or JavaScript challenges for non-human browsing behavior, declaring prohibitions on automated purchasing in robots.txt and terms of service, and requiring device fingerprints or secondary verification at checkout. For platforms, blocking agents is not just a stance but also involves practical operational cost calculations: agent orders drive up return rates, disrupt inventory forecasting, increase payment fraud detection difficulty, and may bypass the platform's recommendation slots and advertising display logic. However, blocking also carries compliance and competitive risks—for example, in some regions it could touch on antitrust or interoperability regulatory discussions. Amazon's choice not to publicly explain is partly because any explicit rule could be specifically circumvented by agent developers.

Meta's 100,000-User Base Makes "Blocking" More Costly

There is another easily overlooked variable in this game: the supply side of agents is rapidly expanding. Meta says 10 million people have created Muse agents. This number doesn't directly equate to shopping demand, but it signals that the user habit of "letting an agent make decisions for me" is taking shape.

Once these agents begin executing purchases, blocked stores will face structural attrition rather than isolated user complaints. Walmart's early opening effectively pre-claims the position of default shopping destination for this cohort of agents; Amazon's blockade is a bet on time—wagering that agent shopping won't grow large enough to ignore anytime soon.

Muse agents are a Meta feature for creating personalized AI agents. Users can describe goals and preferences in natural language to generate agent instances capable of executing specific tasks. Unlike general-purpose chatbots, these agents are designed to run long-term with certain memory and tool-calling capabilities. Ten million creations doesn't directly equal ten million shopping agents, but it reflects a key trend: ordinary users are already getting comfortable with the interaction paradigm of "configure an agent and let it act on your behalf." Once these agents connect to payment and e-commerce interfaces, they will become a new form of traffic gateway—no longer relying on users actively opening an app to browse, but having agents complete price comparisons, filtering, and ordering in the background. For retailers, this means the value of traditional search rankings and homepage recommendation slots may be redistributed.

What Agent Shopping Really Needs to Solve Isn't "Can It" But "Dare You"

In Walmart's solution, authorization profiles and spending limits are two notable details. They show that retailers have realized the bottleneck of agent shopping isn't technology—it's trust. Users need to know that agents won't overspend, won't buy the wrong things, and that there's a paper trail if something goes wrong.

The per-transaction approval mode is the safest but feels close to manual shopping; setting a limit partially delegates judgment to the agent in exchange for convenience. These two tiers are essentially probing how much control users are willing to hand over. Whoever can better calibrate this trust curve will have a better chance of becoming the default checkout channel in the agent era.

As for whether Amazon will eventually follow suit, there are no public signals yet. But one thing is certain: once Walmart turns "agents can enter the store" into a formal service, blocking goes from being the default option to an exception that needs explaining. The rule-making competition among e-commerce platforms around AI agents has only just begun.

One-line Summary

Walmart opens agent checkout while Amazon chooses to block, kicking off the rules battle over AI shopping agents in e-commerce.

Paragraph Summary

On October 21, Walmart launched its "agent gateway," allowing AI agents to search products and complete checkout—making it the first major retailer to systematically open up to AI shopping agents. Its mechanism requires each agent to create an authorization profile, with per-transaction approval as the default and optional spending limits for autonomous shopping, keeping control in users' hands. Amazon, meanwhile, confirmed to media that it has blocked such agents, taking a silent but firm stance. The article argues that the divergence between the two giants is essentially two answers to whether agent shopping is a threat or a new channel. Meta says 10 million people have created Muse agents, and the expansion of agent supply will amplify structural attrition for blocked stores. The real bottleneck of agent shopping isn't technology but trust—authorization and limit designs are probing how much control users are willing to cede.