Written by Jeremy Souffir Founder, JTS Tech Services
The short version. In the fourteen days after Meta launched its personal agent Muse, the agent acquired three separate ways to pay. It shipped on 8 September with Stripe's Link wallet. On 21 September Shopify's chief executive announced agentic checkout with Shop Pay across Shopify stores. On 22 September PayPal joined as the third. Most coverage framed this as a payments race, which it is. But if you run a Shopify store, something quieter happened to you specifically: you became reachable by a new buyer, over a new order path, using a wallet you did not pick, without anyone at your company approving it. That is not a scandal, and it is probably even good for you. It is simply a change to how a sale can reach your business, and right now most merchants could not say what their own store is set to do about it.
What actually happened, in order?
- 8 September: Meta launched Muse in the US, a personal agent that browses and acts for you. It launched with Stripe's Link wallet, which per Shopifreaks checks out directly at more than a million businesses and generates a one-time virtual card everywhere else.
- 20 to 21 September: Amazon began blocking Muse from Amazon.com, objecting that the agent does not identify itself and that it moves through customer accounts. We wrote about that yesterday, and it is the other half of this story rather than the same half.
- 21 September: Shopify's chief executive Tobi Lutke announced the company is "partnering deeply with Muse to enable agentic checkout with Shop Pay on all Shopify stores." Meta's chief AI officer posted the same line from the other side.
- 22 September: PayPal announced its checkout inside Muse, letting its customers buy through the agent at merchants that accept PayPal. Third rail in fourteen days.
- Scale, for context: PYMNTS reports Shop Pay passed $400 billion in lifetime accelerated gross merchandise volume by June 2026. This is not a pilot integration on a side channel.
Whose claims these are, and where they stop
The "all Shopify stores" line is the chief executive's own phrasing in a public post, echoed by his counterpart at Meta, and repeated by the trade press from those posts. It is a strong claim from someone entitled to make it, but it is a sentence, not documentation. As of the reporting we can read, several things a merchant would actually want are simply not published anywhere: how a Muse order is attributed, whether the usual payment fees apply unchanged, how returns and customer service work when the buyer was software, and what data moves between Meta and Shopify. We are not going to guess at any of those, and you should be wary of anyone who does this week. Notably, one outlet that went looking for a Muse-specific opt-out reported it could not find one clearly documented at the time it checked, while a specialist Shopify outlet describes two toggles that already exist in the admin. Both were written within days of each other. That disagreement is the most useful thing in this article, and we come back to it below.
Why is three wallets in two weeks the story, rather than one more AI channel?
Because of what each rail removes. For most of e-commerce history the checkout was the one part of the journey a merchant fully controlled. You chose the payment methods, you designed the page, you decided what to upsell on it, you owned the confirmation screen and the email after it, and you learned something about the customer along the way. Discovery could happen anywhere, on a search engine, a marketplace, a social feed, but the sale itself came home.
An agent with its own wallet inverts that. The customer never sees your checkout, because the point of the agent is that they do not have to. The wallet that completes the sale was chosen by the agent's maker and the customer's prior habits, not by you. And with three rails live in a fortnight, the agent is not even committed to one: it can route a given purchase through whichever of Link, Shop Pay or PayPal fits that merchant and that buyer. The competition happening here is not for your business. It is for the position directly in front of your business, and it is being settled quickly, by companies negotiating with each other.
So what does this look like inside a Shopify store?
Here is where the honest labelling matters. According to a specialist Shopify outlet that walked the admin, the plumbing already existed before this announcement: your products reach agents through Shopify Catalog, which was populated by default rather than by opt-in; the transaction runs over the Universal Commerce Protocol; and the resulting order lands in your admin tagged to Meta the way any other sales channel's orders are. Meta joins ChatGPT, Google's AI surfaces and Microsoft Copilot inside an Agentic Storefronts sales channel that most merchants have never opened. The same outlet describes two switches under that channel: catalog access, on by default, which you can turn off although Meta may still find your products by crawling the open web; and direct checkout, on by default for eligible stores, which you can turn off to stay discoverable while sending the buyer to your own site to finish.
We are reporting that as one outlet's reading, deliberately, because another outlet checking around the same time said no Muse-specific opt-out was clearly documented. We cannot reconcile those from the outside and we are not going to pretend otherwise. What we can tell you is the part that is true regardless of which account is more current: there is a sales channel in your Shopify admin that governs how AI agents see and buy from your store, its defaults were set by somebody other than you, and you can go and read them in about two minutes. Your own admin is the authority here. Not a CEO's post, not the trade press, and not us.
The two conclusions that both get this wrong
The first wrong conclusion is to switch it off. The reflex is understandable, because nobody enjoys being enrolled in something by default. But the customer on the other end is a real customer with real intent who has asked a machine to go and buy something, and the store that makes that errand fail is not defending its brand, it is losing an order to whoever made it easy. Discoverability without checkout is a reasonable middle setting; blanket refusal mostly is not. The second wrong conclusion is the opposite reflex: it is on, it works, Shopify handled it, nothing to do. That one ignores that you now have a sales channel with no owner, unknown attribution, an untested returns path and no monitoring, which is exactly the profile of the integrations that quietly break and are discovered a quarter later in a revenue report. The mistake is not choosing on or off. It is not knowing which one you are on, and having nobody responsible for the answer.
What would we actually check this week?
None of this is a project. It is an afternoon of reading settings and writing down what you find, which is the cheapest work in this entire category and almost nobody does it.
- Open Sales channels in your Shopify admin and read the agentic section end to end. Note what is on, what is off, and which of those you or anyone at your company chose. That written note is worth more than any prediction about agentic commerce, because it is about your store.
- Decide the discoverability question and the checkout question separately. They are different settings and they deserve different answers. Plenty of merchants will sensibly want to be found by every agent while still finishing the sale on their own site, at least until the attribution and returns mechanics are published.
- Place a test order through the path if you can, and follow it all the way. Does it appear in your admin tagged correctly? Does it enter your fulfilment flow like any other order? Does the customer get your usual emails, or does the confirmation happen somewhere you cannot see? Does your analytics count it, or does it land in direct traffic and quietly distort everything else?
- Ask what happens when this order needs to be returned. The buyer was software, the wallet might be a one-time virtual card, and your returns process assumes a human with an order number and the card they paid on. Somebody in support should know the answer before the first one arrives, not during it.
- Check whether your product data is ready to be read rather than browsed. An agent choosing between you and a competitor is working from titles, attributes, price and availability as your catalog holds them, with no photography and no persuasion. This is the same work as being visible in AI answers, which is why it pays for itself twice.
- Give the channel an owner and a monthly glance. One named person, one recurring five-minute check that orders are still arriving and settings have not changed under a platform update. Defaults move. Ours moved this month.
The genuinely useful part
Three wallets competing in fourteen days is a sign of a market that wants your inventory badly, and that is a better position to be in than the alternative. Shopify merchants in particular got handed a working agentic checkout without building anything, which two years ago would have been a serious integration project with a serious invoice attached. The platform did the hard part. What the platform cannot do is decide how your business wants to be bought from, test that the order path works end to end in your operation, or notice six weeks from now when something changes. Those are small, cheap, human decisions, and they are entirely yours. A merchant who spends one afternoon this week reading their own settings is ahead of most of their competitors, who will read about this, agree it sounds important, and never open the menu.
Where we fit
This is the shape of problem our AI Ops Automation Sprint exists for, and it is worth saying plainly why hiring someone for it beats leaving it on a list. A new sales channel turned itself on inside your store. It has no owner, no documented settings, no tested order path, no returns answer and no monitoring, and it sits alongside a dozen other integrations and scheduled jobs that touch your product and order data in ways nobody has inventoried since the person who set them up left. In a Sprint we go through all of them: name every system that reads or writes your catalog and your orders, record what each one is set to and who owns it, walk the new agentic path end to end with a real test order so you know what your team will actually see, get your product data into the state an agent can read, and leave monitoring behind on the paths that carry revenue. The result is not a strategy document about agentic commerce. It is that the next default someone flips on your behalf shows up as an alert you get, rather than a gap in a report three months later.
Sources
- PYMNTS — Shopify brings Shop Pay checkout solution to Meta's Muse AI agent (21 September; the Lutke and Wang quotes, and the Shop Pay lifetime accelerated GMV figure)
- Shopifreaks — Shopify partners with Meta on agentic commerce, letting the Muse AI agent check out with Shop Pay on all Shopify stores (the Shopify deal as reported at the time)
- Shopifreaks — PayPal partners with Meta to add its checkout to the Muse AI agent, two weeks after Muse launched with Stripe's Link wallet (22 September; the three-rail timeline and the Stripe Link launch detail)
- Yahoo Finance — PayPal joins Meta's Muse checkout ecosystem (the PayPal announcement and its merchant reach)
- ContentGrip — Shopify brings Shop Pay checkout to Meta Muse (the announcement, and the note that a Muse-specific opt-out was not clearly documented at the time of checking, along with the attribution, fee, returns and data questions still open)
- Naughton & Bird — Meta Muse on Shopify: what you need to know (one specialist outlet's walkthrough of Shopify Catalog, the Universal Commerce Protocol, the Agentic Storefronts channel and the two default-on toggles)
- Shopify — The agentic commerce platform (Shopify's own framing of connecting merchants to AI conversations)
- Pulse 2 — Shopify integrates Shop Pay with Meta Muse for AI-powered checkout (independent confirmation of the integration and its date)
- JTS Tech Services — Amazon just locked Meta's shopping agent out (22 September; the same agent seen from the login and fraud side, which is the other half of this)
- JTS Tech Services — The checkout is leaving your website: ACP, UCP, and what to actually do (24 July; the protocols explained before they started shipping on by default)
