When An AI Makes The Purchase, Who Is The Customer?

🔍 Read the full analysis: When An AI Makes The Purchase, Who Is The Customer? on ThorstenMeyerAI.com

Walmart is opening a route for AI agents to search and buy from its store, while Amazon has blocked Meta’s Muse, sharpening a dispute over who controls the customer relationship when software shops on a person’s behalf. Meta said on its October 21 earnings call that more than 10 million people in the United States and Canada had set up Muse in roughly six weeks, with shopping its most common use.

Walmart’s October 21 announcement introduced Agent Profiles, which let customers set permissions and spending limits, and an Agent Gateway built around the Model Context Protocol, or MCP. The gateway is intended to let approved agents access shopping functions through a structured interface rather than imitate a person’s clicks on a retail website. Meta’s reported adoption figures and average purchase value of more than $200 are company-reported figures; Meta has not supplied independent verification in the source material.

The disagreement is already visible in how agents reach stores. Muse uses a browser in a virtual machine hosted in Meta’s cloud. Amazon blocked the agent in September and said it had not been told Muse would shop on its site, did not identify itself, and appeared to capture and store customer credentials. Meta disputed the password and payment-method concern, saying Muse had no visibility into those details. A Verge reporter’s price comparison illustrated the access issue: Muse listed Sam’s Club, Walmart and Costco for granola bars, but said Amazon might have been cheaper and that it could not check because Amazon had blocked it.

Other companies are trying structured alternatives. Shopify has tools for browser-agent checkout, and Meta and Sierra have published a draft Personal Agent Protocol, also called Poppy, with Walmart, Shopify, Stripe and other design partners. Google and Shopify have also published the Universal Commerce Protocol, while OpenAI and Stripe developed the Agentic Commerce Protocol. These efforts address different parts of shopping and do not yet amount to one settled standard. OpenAI stepped back from its initial ChatGPT Instant Checkout approach in March 2026, saying it would let merchants use their own checkout experiences while it focused on product discovery.

Who Controls the AI Shopping Relationship

The immediate stakes are access, payment and accountability. If an agent selects a product and completes a purchase, retailers may have less direct interaction with the shopper and less control over how products are presented. An agent provider could become the main point of contact, influencing which stores and products are considered. Meta has said it expects to make money by taking a small fee from transactions; the source also cites Amazon advertising revenue of more than $68 billion last year, a business that may be affected if shopping discovery shifts to outside agents.

There are implications for publishers and affiliate businesses as well. Amazon Associates commissions generally rely on customers arriving through tracked links. If an agent reads a review and later buys through another route, the publisher may not receive credit for the sale. A September HUMAN Security analysis cited in the source found that 81% of agent requests went to product and search pages, while 2.4% went to checkout and payment. Those figures describe the requests measured in that analysis, not all agent shopping activity; they suggest agents were more often researching than completing purchases in the measured data.

For shoppers, handing purchases to software makes familiar consumer questions more complicated. Customers need to know which account and payment method an agent can use, whether a purchase requires approval, and which retailer’s return rules apply. Walmart’s spending controls address part of that problem, but a spending limit alone does not resolve disputes about mistaken orders, data handling or the agent’s responsibility when something goes wrong.

From Browser Bots to Retail Gateways

The current dispute developed over a short series of announcements. Meta launched Muse on September 8, 2026, describing it as a personal agent able to use a browser, fill out forms and negotiate for users. The agent runs in a cloud-hosted virtual machine and uses Link by Stripe for payment through a one-time-use card, according to the source. Amazon blocked Muse on September 20, citing its Conditions of Use and concerns about the agent’s conduct and handling of credentials.

Meta announced retail connections with companies including Walmart, Best Buy, Gap, Sephora and Wayfair on September 23. Shopify later introduced structured tools for agents to work with checkout, while Meta and Sierra announced Poppy in October. Walmart’s Agent Gateway announcement followed on October 21. In parallel, other retailers and financial institutions have raised concerns about outside agents operating in customer accounts; the source reports that JPMorgan Chase CEO Jamie Dimon described them as too risky.

The technical approaches reflect a basic tension. A browser agent can use pages designed for people, but a retailer may have difficulty distinguishing it from the account holder. A merchant’s own gateway can grant limited, defined access, but only where the retailer chooses to support it. Shared protocols aim to help agents and stores exchange information in a consistent way, yet their scope and adoption differ. The source describes Poppy as a draft and notes that payment functions are handled through extensions rather than included in its initial specification.

“Continued access by an unauthorized AI agent violates Amazon’s Conditions of Use, to which our customers have agreed.”

— Amazon, in a message shown to Muse users

Rules for Agent Purchases Remain Open

It is not yet clear how widely retailers will adopt agent gateways or whether competing protocols will become interoperable. The source describes some specifications as drafts, and the different systems cover different parts of a purchase. It also does not establish how many shoppers have completed transactions through Muse, how Meta calculated its average purchase figure, or how its adoption numbers compare with other shopping tools.

Responsibility for errors and disputes is also unsettled. The source does not specify a common policy for refunds, unauthorized purchases, product misrepresentation or the handling of shopper data across an agent, payment provider and retailer. Amazon’s allegations about Muse and Meta’s denial remain competing statements; the material does not independently establish whether Muse stored credentials or what access it had in practice. Nor is it clear how affiliate commissions will change if purchases increasingly happen without tracked links.

Retail Access and Protocol Adoption

The next developments to watch are whether Walmart’s gateway moves from announcement to broader use, whether more retailers permit outside agents, and whether Poppy, UCP or ACP gains wider support. Adoption will depend not just on technical compatibility but also on retailers’ decisions about identity checks, customer permission, payments and the commercial terms for access.

For shoppers, the near-term practical step is to review an agent’s permissions before enabling purchases. Spending caps, approval requirements and payment controls can limit exposure, but consumers should also check which retailer is taking the order and whose terms govern returns and disputes. The source materials do not provide a single industry-wide standard for those protections.

Key Questions

What is the dispute over AI shopping agents?

Retailers and agent companies differ over whether software should shop through ordinary browser pages or use a dedicated, approved connection. The dispute also concerns customer access, data, payment and revenue when an agent handles a purchase.

Why did Amazon block Meta’s Muse?

Amazon said Muse was an unauthorized agent, had not identified itself, and appeared to capture and store customer credentials. Meta said Muse had no visibility into people’s passwords or payment methods. The source material does not independently resolve those competing claims.

What is Walmart’s Agent Gateway?

Walmart announced an MCP-based interface intended to let approved agents access shopping functions through structured tools. Its related Agent Profiles let customers set permissions and spending limits, according to the announcement described in the source.

Are AI agents already completing most online purchases?

The source cites a September HUMAN Security analysis in which 2.4% of measured agent requests went to checkout and payment, compared with 81% to product and search pages. That dataset suggests research was more common than checkout in the measured traffic, but it does not establish the share of all online purchases made by agents.

What should shoppers check before letting an agent buy?

Review which purchases need approval, set a spending cap, and check the payment method the agent can use. Shoppers should also confirm which retailer is responsible for the order and what return and dispute terms apply.

Source: ThorstenMeyerAI.com