🔍 Read the full analysis: AI Shopping Agents Challenge Retailers’ Hold On Customer Relationships on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Meta says more than 10 million people in the United States and Canada set up its Muse shopping agent within about six weeks, while Walmart has announced tools for approved agents and Amazon has blocked Muse. The competing approaches put retailers’ control of checkout, customer data and advertising revenue at the center of a fast-moving dispute; reported usage figures and commercial effects remain largely company claims or early data.
Walmart has announced a way for approved AI shopping agents to search its inventory and make purchases, while Amazon has blocked Meta’s Muse agent, setting up a dispute over who controls the shopping relationship when software shops for consumers. The split comes as Meta says more than 10 million people in the United States and Canada set up Muse within about six weeks, though those adoption figures come from the company.
Walmart’s October 21 announcement introduced an Agent Gateway, built around the Model Context Protocol, and Agent Profiles that can give an agent permission to shop under customer-defined limits. The gateway is intended to let approved agents access product inventory and make purchases through structured connections rather than acting only by clicking through web pages. The announcement does not establish how widely the system is operating or how many shoppers or agents are using it.
Meta launched Muse on September 8. The company describes it as a personal agent that can browse, fill out forms and act on a user’s behalf. In an October 21 earnings call, Meta said more than 10 million users in the United States and Canada had set it up in roughly six weeks, shopping was its most common use and the average purchase exceeded $200. These are Meta’s figures; independent confirmation was not provided in the source material.
Amazon blocked Muse on September 20. The Verge reported that the agent could not check Amazon prices when asked to compare the cost of granola bars, while returning results from Sam’s Club, Walmart and Costco. Amazon said Muse’s access violated its Conditions of Use, arguing the agent did not identify itself and appeared to capture and store customer credentials. Meta disputed the account, saying Muse had no visibility into users’ passwords or payment methods. The available information does not resolve that disagreement.
Agentic shopping
Three ways an AI agent buys for you
Every fight between shops and AI agents comes down to which of these three routes the agent uses, and whether the shop agreed to it.
The browser, like a human
The agent opens pages built for people, signs in with your login and clicks through checkout. Works almost anywhere, which is why shops object: they can't tell the agent from you.
Examples: Meta's Muse (from a virtual machine in Meta's cloud), Perplexity's Comet (from your own computer)
The shop's own agent door
Approved agents search and buy through a structured interface instead of clicking through pages. The shop sets the rules.
Examples: Walmart Agent Gateway (MCP, beta), Shopify MCP server and WebMCP checkout tools
Shared protocols
Common rules for who the agent is, what you allowed, and how the checkout and payment are confirmed.
Examples: Personal Agent Protocol (draft), Universal Commerce Protocol, Agentic Commerce Protocol
The protocols competing to be the standard
| Protocol | Backers | Covers | Status |
|---|---|---|---|
| Personal Agent Protocol (Poppy) | Sierra, Meta, Walmart, Shopify, Stripe + 35 design partners | Agent identity, customer permission, one session across site, API and company agent | Draft 0.1 (9 Oct 2026) |
| Universal Commerce Protocol (UCP) | Google, Shopify; site also lists Etsy, Wayfair, Target, Walmart, Amazon, Microsoft | Discovery, cart, checkout, orders; payment mandates via AP2 | Published |
| Agentic Commerce Protocol (ACP) | OpenAI, Stripe | Checkout between agent and merchant | Open source; ChatGPT's own checkout scaled back in March 2026 |
What a careful agent purchase looks like
- You give the task and a limit, for example a $50 cap.
- The agent identifies itself to the shop and gets only the access you approved.
- It prepares the cart and shows you an approval card.
- Payment goes through a token or one-time card, so the agent never sees your card.
- The shop’s normal returns apply if the agent got it wrong.
Sources: Meta (8 Sep 2026), Walmart (21 Oct 2026), TechCrunch on Shopify (28 Sep 2026), Sierra (9 Oct 2026), ucp.dev, agenticcommerce.dev, OpenAI (24 Mar 2026). Purchase steps: Walmart’s agent terms and Meta’s Muse description.
Powered by Thorsten Meyer AI
Retailers’ Control of Agent Purchases
The conflict is about more than whether an automated tool can load a product page. Retailers have traditionally controlled the search, checkout and service experience on their own sites, including the data generated along the way. When a separate agent chooses a product and handles payment, the retailer may lose direct contact with the shopper and have less influence over what is recommended.
Advertising and referral income are also at stake. The source material says Amazon generated more than $68 billion in advertising revenue last year, citing GeekWire, and that Meta expects to earn a small transaction fee from Muse. Those businesses are not equivalent, but both depend on companies holding a position in the path between customers and products. Affiliate publishers may also lose credit when an agent reads a review and completes a purchase without following a tracked link. The size of any resulting losses has not been established.
For shoppers, delegated purchasing introduces questions about permission, payment and returns. A purchase made by an agent could involve limits set by the user, a payment service, an agent provider and a retailer. Customers will need to know which party handles errors, cancellations and return requests, and how their information is shared.
Products Worth Considering
Sonos Era 100 Wireless Smart Speaker, Black
Bose TV Speaker (2nd Gen) Compact Soundbar with Bluetooth and HDMI Connectivity
Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Browser Bots to Retail Gateways
AI shopping agents can use different routes to reach a store. A browser-based agent interacts with pages designed for people and may sign in and check out on the user’s behalf. Retailers can object that this makes it difficult to distinguish an agent from its customer. Muse operates from a dedicated virtual machine in Meta’s cloud, according to the source material; the source says its payments use Link by Stripe and a one-time-use card.
Other systems offer structured connections between agents and merchants. Shopify has introduced tools for browser agents to complete checkout and provides a hosted Model Context Protocol server. Walmart’s gateway is another effort to provide controlled access rather than unrestricted automated browsing. Separately, Sierra and Meta announced the Personal Agent Protocol, also called Poppy, with Walmart, Shopify and Stripe, followed by a draft and additional design partners. Google and Shopify’s Universal Commerce Protocol and OpenAI and Stripe’s Agentic Commerce Protocol are among other efforts to set rules for transactions. The protocols differ in scope and maturity; Poppy was described as a draft, while the source characterizes UCP as published and ACP as open source.
The commercial model is still in flux. OpenAI introduced Instant Checkout in ChatGPT in September 2025, then said in March 2026 that the initial version lacked the flexibility it wanted and that it would let merchants use their own checkout experiences while focusing on product discovery. That change illustrates that connecting an agent to checkout is not the same as establishing a durable shopping system.
“Continued access by an unauthorized AI agent violates Amazon’s Conditions of Use, to which our customers have agreed.”
— Amazon, in a statement about Muse, as reported by The Verge
Adoption, Access and Accountability
The early figures do not show how often Muse users complete purchases, how many return to the service, or whether the reported average purchase value reflects typical behavior. Meta’s statements are company-reported, and no independent user or transaction data is supplied here. Separate September data from HUMAN Security, cited in the source material, found that 81% of observed agent requests went to product and search pages and 2.4% to checkout and payment. The dataset’s scope and how well it represents shopping agents overall are not specified in the material, so the figures should not be treated as a universal rate.
It is also unclear how many retailers will permit outside agents, which protocols will gain broad adoption, and whether customers will be able to move easily between them. Amazon’s allegations about Muse and Meta’s denial remain disputed. The longer-term effects on retail advertising, affiliate referrals and prices have not been quantified. Nor is it clear which company would be responsible when an agent makes an unauthorized purchase, uses incorrect information or fails to complete a return.
Rules for the Next Checkout
The next test is whether Walmart’s gateway and customer permission profiles move from announcement to regular use, and whether retailers and agent providers settle on practical ways to identify agents, set spending permissions and share responsibility for a transaction. Protocol drafts and design partnerships may show which technical approaches attract participants, but adoption by merchants and shoppers will matter more than the number of partners announced.
Users considering an agent should check its purchase permissions, set a spending cap or require approval for each order, and understand which payment method and return policy apply. For retailers and publishers, the unresolved issue is whether an agent can generate sales while preserving attribution and a direct relationship with the customer. There is not yet enough evidence to say which model will prevail.
Key Questions
What did Walmart announce?
Walmart announced an Agent Gateway for approved AI agents to access its inventory and make purchases through structured connections. It also described Agent Profiles that let customers set permissions and spending limits. The scale of deployment and actual usage have not been established.
Why did Amazon block Meta’s Muse?
Amazon said Muse violated its Conditions of Use, did not identify itself as an agent and appeared to capture and store customer credentials. Meta said Muse had no visibility into passwords or payment methods. The competing accounts have not been independently resolved in the source material.
How many people have set up Muse?
Meta reported that more than 10 million people in the United States and Canada set up Muse in roughly six weeks after its September 8 launch. The figure is Meta’s own report, not independently verified data.
Does the data show that shopping agents usually complete purchases?
No. HUMAN Security data cited in the source material says 81% of observed agent requests went to product and search pages, while 2.4% went to checkout and payment. The dataset’s scope is not specified, and the figures do not establish how often agents complete purchases across the market.
What should shoppers check before letting an agent buy?
Review what the agent is authorized to buy, set a spending limit or require approval, and check which payment method it will use. Shoppers should also confirm which retailer’s cancellation and return policy applies and what happens if an order is incorrect.
Source: ThorstenMeyerAI.com
