Two of the largest commerce platforms in the world took opposite decisions about the same piece of software this month, and the seller’s catalogue is the thing being argued over. Amazon asked Meta to remove Amazon from its new shopping agent and enforced that request on its own site. Shopify went the other way and switched agentic checkout on by default for eligible merchants.

The reason this happened in September, rather than at any point in the previous two years, is a court ruling that almost nobody has connected to it.

What Meta launched

Meta introduced Muse on 8 September, describing it as “a secure, private personal AI agent that proactively helps with people’s goals and suggests ideas.” The capability that matters commercially is blunt: “It can open a browser, fill out forms, and negotiate on their behalf.” Where a retailer offers no interface for an agent, Muse uses the ordinary web the way a person would.

It is rolling out in the United States on iOS, Android and the web. At launch it pays with Link, built by Stripe, and Meta says Shop Pay is coming as a second option. Meta also says Muse “has no visibility into people’s passwords or payment methods.”

From the publisher

Everything we know about running Amazon, in one course.

The commercial strategy, taught by operators who run our own brands.

Amazon’s position, in Amazon’s words

Amazon says it had no advance notice that Muse would shop its site and did not authorise it. Its spokesperson statement is the whole argument in one sentence: “We think it’s fairly straightforward that third-party applications that offer to make purchases on behalf of customers from other businesses should operate openly and respect service provider decisions about whether or not to participate.”

The verb Amazon uses is that it asked Meta to remove Amazon from the experience. That request was then backed by enforcement on Amazon’s side. It is worth keeping those two things distinct, because the gap between asking and enforcing is where the legal interest sits.

Amazon published nothing about any of this in its newsroom. A dispute over who may transact on the largest store in the West was conducted entirely through emailed statements.

The word doing the most work in Amazon’s statement is “openly”. Read charitably, it asks agents to identify themselves and honour a retailer’s decision about participation, which is a reasonable governance position and one most platforms will eventually want. Read commercially, it describes a world in which Amazon decides which agents may buy, on what terms, and therefore keeps the ability to charge for the privilege.

There is a structural reason Amazon would rather this happened on its own terms. An agent that completes a purchase without ever rendering a search results page shows no sponsored placements. Advertising is among the most profitable things Amazon does, and it is attached to a page that agentic buying is designed to skip. Nobody involved has said this, and no figure has been published for it, but it is the shape of the incentive underneath the principle.

No number of any kind attaches to this story. Not in Amazon’s statements, not in Meta’s announcement, not in the reporting. Anyone offering you a market size for agentic commerce this week has made it up.

Tool comparison · FastMoss vs Kalodata

Agent-driven buying removes the browsing step, which is where most product discovery data comes from today. Sellers who already track demand at product level on social commerce surfaces will notice the shift first. FastMoss and Kalodata both measure product and creator performance on TikTok Shop.

Comparison coming soon

FTC disclosure: we may earn a commission if you buy through links on this page. It costs you nothing and does not change what we recommend.

The ruling that changed the weapon

On 4 August the Ninth Circuit vacated the preliminary injunction Amazon had won against Perplexity’s shopping agent, in Amazon.com Services LLC v. Perplexity AI Inc., No. 26-1444.

The reasoning is the part that travels. The court held that “It is the user who ‘accesses’ Amazon’s computers, with the help of the Assistant to carry out specific acts on Amazon.com”, and that “However advanced the Assistant currently is, it is a tool, not a person for statutory purposes.” On that reading, Amazon was unlikely to succeed on its claim under the federal computer-misuse statute, because the agent’s maker never accessed anything.

The court was careful about what it had not decided. It expressly left other theories alive, saying it did not address whether “in other contexts, including tort claims”, the agent’s maker could avoid liability, and it noted that Amazon retains the ability to enforce access restrictions through its terms of service.

Seven weeks later Amazon blocks a different agent and grounds the block in its Conditions of Use. That is a contract theory, which is precisely the door the court left open, rather than the hacking theory the court had just closed. Amazon did not change its mind about agents. It changed its instrument, because a court took the first one away.

That sequence is worth holding onto, because it predicts what comes next. A contract claim binds whoever agreed to the terms, and the party clicking through a retailer’s conditions in an agentic purchase is the shopper, not the software company. Enforcement of this kind points at the customer rather than at the agent’s maker, which is an uncomfortable place for a retailer to end up and a reason to expect the terms themselves to be rewritten before anyone is sued.

A person working at a laptop
Where no interface exists for an agent, Muse uses the ordinary web the way a person would. Photo: Mikael Blomkvist / Pexels.

Shopify took the opposite decision, and documented it

Shopify’s own help centre is unambiguous: “When your store is eligible, purchasing through Meta’s direct checkout is active by default in your Shopify admin, and it has to remain active for your products to be directly bought by customers on Meta surfaces.”

The scope on that sentence is doing heavy lifting and is routinely dropped. Shopify’s stated eligibility is “You must sell to customers in the United States, Canada, or Mexico.” A merchant selling only into European markets is not enrolled in this and has nothing to switch off today. That distinction is the difference between a task and a watching brief, and most of the coverage has blurred it.

For merchants who are eligible, the terms are mild. Shopify says “There are no fees associated with selling in Meta’s direct checkout. You pay only your standard payment processing fees”, and “You remain the merchant of record, and you continue to manage fulfillment, returns, and customer service for these orders.”

The switch exists, and it is four steps down. Sales channels, then Agentic, then deactivate “Allow Shopify to manage for me”, then Meta, then deactivate Direct checkout, then save.

Tool comparison · FastMoss vs Kalodata

If a buying agent picks products on price and specification alone, the listing data underneath becomes the whole battleground. Knowing how your titles and prices compare against the category is the first defensible step. FastMoss and Kalodata both surface competitor listing and pricing data.

Comparison coming soon

FTC disclosure: we may earn a commission if you buy through links on this page. It costs you nothing and does not change what we recommend.

For the business operator

If you run a Shopify store selling into the United States, Canada or Mexico, open Sales channels and look at the Agentic section this week. The point is not that the default is wrong. The point is that it is a decision you have not yet made, and you should make it deliberately rather than discover it.

Pay particular attention to the setting that lets Shopify manage agentic channels for you, because it enrols the store in future channels automatically. That is a standing consent, not a one-off.

If you sell only into Europe, do nothing operationally and keep this on a watching brief. Your useful work is different: ask whoever owns your customer data what happens to consent when the buyer is an agent. An agent that holds its own email address becomes the address of record on the order, and no marketing permission attaches to it.

From the consumer’s side

The shopper in this arrangement is doing something they have not done before, which is handing purchase authority to software and keeping only a veto. That changes which signals matter. A shopper comparing products reads reviews, photographs and brand; an agent comparing the same products reads structured fields.

It also quietly changes who the seller’s customer is. When the agent carries its own address and its own payment instrument, the relationship the brand thought it was building with a person is mediated by a piece of software that has no loyalty to it. Shoppers rarely notice this at the point of purchase. They notice it later, when the follow-up they expected never arrives.

The three-line version

If you run the store or the marketplace channel, check your Shopify agentic settings if you sell into North America, and make the default a decision rather than an inheritance. The line to take upward is that the platforms have taken opposite positions, so there is no industry default to follow and the choice is genuinely ours.

If you have just taken on digital or e-commerce, the gap this exposes is consent. Nothing in the current agentic checkout arrangements captures marketing permission, so an agent-placed order arrives with an address you cannot lawfully add to a list. That is a data protection question sitting inside a channel decision, and it has no owner in most organisations.

If you sign the budget, refuse any spend framed as agentic commerce readiness. There is no European agentic checkout to be ready for, no published market size worth the name, and the only work that needs doing is a settings review and a consent question, both of which are free.

Sources

Meta Newsroom, Introducing Muse — https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/

Shopify Help Center, Meta agentic storefronts — https://help.shopify.com/en/manual/online-sales-channels/agentic-storefronts/meta

Amazon.com Services LLC v. Perplexity AI Inc., No. 26-1444 (9th Cir. 4 August 2026) — https://law.justia.com/cases/federal/appellate-courts/ca9/26-1444/26-1444-2026-08-04.html

Retail Dive — https://www.retaildive.com/news/amazon-meta-muse-ai-agent-shopping-experience/831018/

Leave a Reply

Your email address will not be published. Required fields are marked *