🔍 Read the full analysis: Who Owns The Customer In An AI-Driven Shopping Journey? on ThorstenMeyerAI.com
Get smart everyday buys 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 agent within about six weeks, with shopping its most common use. Walmart is building tools for approved agents, while Amazon has blocked Muse, exposing a fight over access, customer relationships and who receives credit when an agent shops.
Walmart is opening shopping tools to AI agents while Amazon has blocked Meta’s Muse, a dispute that puts control of the customer relationship at the center of a fast-developing retail shift. 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 about six weeks, and that shopping was its most common use; those figures are the company’s own, not an independently verified count.
A demonstration of the access problem came when a Verge reporter asked Muse where granola bars cost less per piece. The agent named Sam’s Club, Walmart and Costco, then said it could not check Amazon because the retailer had blocked it. The incident illustrates a practical limit of shopping agents: they can only compare stores they can access, and access can depend on each retailer’s rules and technical controls.
Walmart announced Agent Profiles and a Walmart Agent Gateway on October 21. The company’s gateway uses the Model Context Protocol, or MCP, to let approved agents search inventory and make purchases through structured access rather than navigating a store website as a person would. Agent Profiles are designed to set permissions and spending limits. The announcement does not mean that every agent can use the system automatically.
Amazon’s position is different. The Verge reported that users attempting to use Muse on Amazon saw a notice saying access by an unauthorized AI agent violated the company’s Conditions of Use. Amazon said Muse had not identified itself as an agent and alleged it appeared to capture and store customer credentials. Meta’s launch post said Muse had no visibility into people’s passwords or payment methods. Those accounts are disputed, and the supplied material does not independently resolve how Muse handled credentials.
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 Compete for Agent Access
The dispute matters because shopping agents may place a new service between consumers and stores. An agent can compare products, narrow choices and, in some cases, complete a purchase. Retailers then face a choice: permit outside agents, build a controlled route for them, or direct shoppers toward their own assistant. That choice can affect which products appear in recommendations, what information an agent can retrieve and how a transaction is completed.
It also affects the economics of online shopping. The source article says Amazon earned more than $68 billion from advertising last year, citing GeekWire. Meta, meanwhile, has said it expects to profit by taking a small fee from transactions. These are different business interests: a retailer may want to keep shoppers within its own shopping and advertising systems, while an agent provider may seek payment for generating or completing sales. The reported ad figure and Meta’s stated plan do not, by themselves, establish how the companies will earn from agent-driven purchases in practice.
Publishers and affiliate businesses could also lose attribution when an agent reads a review or product listing and completes a purchase without using a tracked link. The source material describes Amazon Associates commissions as tied to customer clicks on tracked links. If an agent makes a purchase through another route, the publisher may not receive the credit it would get from a conventional referral. The size of that effect is not established by the material.
For consumers, the issue is not only convenience. An agent that can spend money raises questions about purchase approval, spending limits, payment security and returns. A retailer’s agent gateway and a third-party browser agent may operate under different permissions and terms. Users need to know who is placing the order and which merchant’s policies govern it.
Popular right now
TCL 85-Inch QM6K Series QLED Mini LED 4K HDR Smart Google TV
Vitamix 5200 Professional-Grade Blender, 64 oz. Container, Black
Ninja Foodi Air Fryer 10 Qt DualZone 6-in-1 Black DZ302
As an affiliate, we earn on qualifying purchases.
From Browser Bots to Retail Gateways
The current dispute follows a rapid series of announcements described in the source material. Meta launched Muse on September 8, presenting it as a personal agent that can open a browser and complete tasks. The system runs in a virtual machine in Meta’s cloud and uses Link by Stripe for payments, which the source says issues a one-time-use card. On September 20, Amazon blocked Muse. Meta later announced retail connections involving Best Buy, Gap, Sephora, Walmart, Wayfair and Expedia.
Other companies are trying to make agent shopping work through structured connections rather than ordinary web browsing. Shopify introduced tools for browser agents to interact with checkout, while Walmart’s gateway offers a route for approved agents to access store functions. Protocol efforts are also underway. The Personal Agent Protocol, associated in the source with Meta, Sierra, Walmart, Shopify and Stripe, had a draft version 0.1 published on October 9. Google and Shopify’s Universal Commerce Protocol and OpenAI and Stripe’s Agentic Commerce Protocol pursue related parts of the shopping process.
These efforts remain at different stages and do not yet amount to one agreed system. OpenAI’s earlier Instant Checkout rollout offers a cautionary example: the company said in March 2026 that the initial version did not offer the flexibility it wanted and that it would let merchants use their own checkout experiences while focusing on product discovery. Building a connection is not the same as establishing a widely used way to buy.
“Continued access by an unauthorized AI agent violates Amazon’s Conditions of Use, to which our customers have agreed.”
— Amazon, in the message shown to Muse users, as reported by The Verge
Access, Safety and Sales Attribution
Several parts of the emerging market remain unresolved. The source provides no independent audit of Meta’s claim that more than 10 million people set up Muse, or of its reported average purchase value of more than $200. It also does not specify the time period or methodology behind that average. The figures should be treated as Meta’s account of usage, not as an independently confirmed measure of completed purchases.
It is also unclear how often agents actually complete checkout rather than research products. The source cites HUMAN Security data from September showing that 81% of agent requests went to product and search pages and 2.4% to checkout and payment. Those are shares of agent requests in that dataset, not shares of shoppers or purchases, and the material does not provide a broader comparison baseline. Retail access may also vary: the source reports that Muse encountered Target’s bot check one day and got through the next, while users had difficulty with services including Delta and Amtrak.
The parties’ competing statements about credentials remain unresolved in the supplied reporting. Nor is there enough information to establish how widespread agent purchases are, whether retailers will accept common protocols, or how sales and affiliate commissions will be attributed when an agent mediates discovery and checkout. Protocols, policies and commercial terms may change as companies test different approaches.
Retail Tools Face Real-World Tests
The immediate test is whether Walmart’s gateway and agent profiles lead to reliable purchases while giving retailers enough control over permissions, identity and spending. Meta’s retail connectors and the draft Personal Agent Protocol will also need adoption from merchants and agent providers before they can offer broad coverage. Announcements and protocol drafts are not evidence of widespread use; usage, completed transactions and consumer protections will be more informative measures.
For shoppers, the practical next step is to check an agent’s permissions before linking accounts or authorizing purchases. The source recommends requiring approval for each purchase, setting a spending cap and considering a one-time-use card. Consumers should also verify the merchant, payment route and applicable return policy. For retailers, publishers and payment companies, the next developments to watch are access rules, protocol support and how purchases are credited. The material does not identify a date for a common standard or a resolution between Meta and Amazon.
Key Questions
What is the dispute between Meta and Amazon about?
Amazon blocked Meta’s Muse agent from accessing its shopping site, saying the agent was unauthorized and raising concerns about identification and credentials. Meta said Muse could not see users’ passwords or payment methods. The supplied reporting does not independently settle those conflicting claims.
What is Walmart’s Agent Gateway?
Walmart announced a gateway using the Model Context Protocol to let approved agents access shopping functions such as inventory search and purchasing through structured connections. Agent Profiles are intended to set permissions, including spending limits.
How many people use Meta Muse?
Meta said on October 21 that more than 10 million people in the United States and Canada set up Muse in about six weeks. The figure is company-reported; the source material does not provide independent verification or define what counts as an active user.
Are AI agents already making most online purchases?
The source material does not support that conclusion. It cites September data from HUMAN Security in which 81% of agent requests went to product and search pages and 2.4% to checkout and payment. Those numbers describe requests in that dataset, not the share of online purchases made by agents.
What should consumers check before letting an agent buy?
Consumers should review purchase approvals, spending caps, payment methods and return terms before authorizing an agent. They should confirm which retailer will process the order and whether the agent is acting through that retailer’s own system or a third-party service.
Source: ThorstenMeyerAI.com
