Welcome to the latest edition of Unpacking Commerce, our newsletter about emerging trends in retail, brands, and new commerce.
👉 TL;DR - three things in this edition:
Muse is Meta’s new AI agent. It does tasks for you, including buying things, and it has been the number one app in the US since 18 September. Amazon blocked it. Shopify opened all its stores to it. Since 29 September, it also runs a small business’s ads. Meta now gets paid twice on the same sale: on the ad, and with a cut of the order.
Who gets paid when an assistant shops: the merchant pays the same, to someone else. OpenAI tried a fee per order and switched to ads within six months; Amazon put ads inside Rufus; Google sells ads inside AI Mode. The ad moves from the retailer’s page into the assistant’s answer. The losers are retailers that sell ads on their own pages, a €13.7bn business in Europe, and have no assistant of their own.
Amazon opens the back door: sellers can now run their eBay, Walmart, TikTok Shop and Shopify business from the Amazon dashboard, for free. Every one of those sales can still use Amazon’s warehouses, delivery, and software, each one billed: seller services made Amazon $172bn in 2025. And Amazon now sees what each seller earns on every rival, which is not allowed in Europe.
Muse, explained
1. What it is
🤖 Muse is a new app from Meta, launched in the US on 8 September. It is an “agent”: an AI that does tasks for you, where a chatbot only answers questions. You tell Muse what you want. It writes the email, books the trip, fills in the form, and asks you before anything final, such as sending a message or paying. Three things are worth knowing. It runs on Meta’s own model, called Muse Spark. Each user gets a private computer in Meta’s cloud, where the agent works and where the user’s logins are stored, so the model itself never sees a password or a card number. And it is free for most uses, with paid plans for heavy users; it is available in the US and Canada only. It was an instant hit: more than 900,000 downloads in the first six days, 3.4m by 25 September, and first place on the US App Store since the 18th. Meta’s own ads accounted for only 6% of the exposure behind those downloads; most of the growth came on its own.
2. How it shops
🛒 Muse can buy things. It pays through Link, the digital wallet run by Stripe. Link creates a single-use card number for each purchase, so the shop never sees the real card, and Link’s buyer protection applies: returns, and a refund if the price drops after you buy. On 23 September, at its Connect conference, Meta named the retailers that agreed to be bought from through Muse: Walmart, Best Buy, Sephora, Ulta, Gap, American Eagle, Dick’s, Fanatics, Michael Kors and Wayfair, plus Instacart for groceries, Shop Pay and PayPal as other wallets, and Expedia “coming soon”. More than 1,500 developers applied to connect their own service in the first week. Zuckerberg also explained how Meta will make money. Muse stays free, and Meta will “profit by taking a small fee from transactions”. In plain terms: when Muse buys something for you, the merchant pays Meta a cut of the order. The rate has not been published.
3. Who let it in, and who shut the door
🚪 This is the heart of the story, so here it is in order. On 8 September, the day Muse launched, Shopify, the software behind more than a million online stores, added Meta as an “AI channel”. From that day, the products of every eligible Shopify store were shared with Meta by default, so Muse could find and recommend them. A merchant can switch this off; nobody was asked to switch it on. On Sunday 20 September, Amazon blocked Muse. A shopper who asked Muse to buy on Amazon saw a notice: “Continued access by an unauthorized AI agent violates Amazon’s Conditions of Use.” Amazon said it had asked Meta to remove Amazon from the app, and Meta had refused. Amazon’s argument: an app that buys on your behalf “should operate openly and respect service provider decisions about whether or not to participate”. Amazon already blocks the shopping agents of Google and OpenAI. On 21 September, Shopify went the other way. Its CEO Tobi Lütke announced “agentic checkout with Shop Pay on all Shopify stores”: Muse can now pay in any Shopify store using Shop Pay, Shopify’s own wallet. Shopify’s shares rose 7% that day.
🧭 Why the opposite choices? Follow the money. Amazon earns when you browse its pages: $68.7bn of advertising revenue in 2025, mostly from brands that pay to appear in its search results. An agent that buys without browsing never sees those ads. Amazon also has its own assistant, Rufus, used by 300m customers in 2025, and wants the shopping conversation to happen there, not in Meta’s app. Shopify earns on the sale itself, a software subscription plus a fee on each payment, so an extra order placed by Muse is pure gain. Two more facts complete the picture. Amazon’s own agent, “Buy for Me”, buys from other brands’ websites without asking those brands first; a brand that objects has to opt out by email. And in August, in Amazon’s lawsuit against Perplexity’s shopping agent, a US appeals court ruled that it is the user who “accesses” Amazon’s site, not the company that built the agent. So Amazon cannot use anti-hacking law against agents. What it has left is its terms of service and a pop-up.
4. What it does for a business
📣 On 29 September, Meta added Muse for Small Business. A business connects its Instagram account, its Facebook Page and its Meta ad account, plus the tools it already uses: Shopify, Stripe, Klaviyo, QuickBooks, Canva, Slack and nine others. Muse then does the work of a small marketing team. It drafts ad campaigns, tells the owner which posts worked, flags the emails that need an answer and the expenses that look wrong. The rule: “nothing publishes, sends, or spends without your approval”. The change is from advice to action. The business tools Meta added to Meta AI in August could read an ad account and make suggestions; Muse can make the change, once the owner says yes. The scale behind it is large. Advantage+, Meta’s automated ad product, is at a $75bn annual run rate; 9m small businesses already use at least one of Meta’s AI tools to make ads. Zuckerberg’s stated goal, possibly as soon as the end of 2026, is advertising “as simple as inputting a credit card number and a business goal”.
5. Where the money goes
🪙 Follow one sale, in three versions. One: the shopper tells Muse "buy me the same running shoes as last year". The brand pays Meta a cut on a customer it had already paid for with Instagram ads. Two: the shopper asks "find me good running shoes under €150". Muse picks a brand whose ad the shopper saw last week. Ad and agent both claim the sale; the brand pays twice. Three: same question, but the shopper has never seen an ad from the brand Muse picks. The cut replaces the ad, and if it is small, this is the cheapest customer the brand has ever bought. Which version a brand pays for depends on two things it can't control (for now at least): the rate, which Meta hasn't published, and the ranking. Meta has not said how Muse chooses between two products (by price, by reviews, by past purchases, or by whether the merchant advertises with Meta); its only statement is that Muse's conversations are not used by its ad systems. Attribution, deciding which channel gets the credit for a sale, was already hard with ads alone. Now ad and agent sit under the same roof, Meta holds both sets of data, and the brand sees the invoice.
6. What remains open
❓ Four questions. How many people will let an agent buy for them? An NIQ survey this spring found that 42% of US consumers had used an AI tool while shopping in the past month, and 5% had let an agent place the order. What will the cut be? Meta has published no rate, and the retailers above signed without knowing it. Does the merchant get a say? On Shopify the default is in; on Amazon the default is out. And Europe. Muse is a US and Canada product, its shopping partners are American chains, and Meta has given no EU date. When it arrives, a European retailer will have to decide whether to be on the list, and whether the orders it brings are new ones or the same orders with a fee attached.
Who gets paid when the agent shops
When an AI assistant does the shopping, the merchant does not pay less to win a customer. It pays someone else. The ad moves from the retailer’s web page into the assistant’s answer, and the retailers that sell ads on their own pages are the ones who lose.
💸 Start with how a sale is paid for today. A brand pays to be seen: an ad on Google, an ad on Instagram, or a “sponsored” slot at the top of Amazon’s search results. That last kind is called retail media: ads a retailer sells on its own site and app, to the brands it stocks, billed per click or per thousand views. In Europe retail media was worth €13.7bn in 2024, up 21%, and should reach €20.8bn in 2026. Most of it is “on-site”, meaning the shopper has to be on the retailer’s page to see it. For scale, Amazon alone sold $68.7bn of advertising worldwide in 2025, more than four times the whole European retail media market.
🤖 Now add an assistant. It does not browse the retailer’s pages. It reads the “feed”, the list of products, prices and stock that the merchant sends directly, and it recommends from that. The sponsored slot is never seen. So who gets paid instead? Three experiments answer. OpenAI launched Instant Checkout in September 2025: buying inside ChatGPT, with “a small fee” paid by the merchant per order and results “organic and unsponsored”. It dropped the checkout in March 2026, and by 31 August ChatGPT was showing ads, at a $1bn annual run rate. Amazon put the ads inside its own assistant: Rufus, used by 300m customers in 2025, now carries “Sponsored Product and Brand prompts”, and nearly 20% of shoppers who interact with one keep talking about that brand. Google sells Shopping ads inside AI Mode, its AI answer page. Meta is the only one promising a cut per order, on top of the ads it already sells. The pattern is clear: the ad does not disappear but moves into the assistant’s answer, and the company that owns the assistant collects it.
📉 Who loses, then? The retailer that sells ads on its own pages but has no assistant of its own. Amazon can defend itself: ads inside Rufus, other agents blocked. Walmart, with a $6.4bn ad business and an assistant, Sparky, that only lives inside its own app, chose to be everywhere: Muse, ChatGPT and Google. A mid-sized European retailer has neither a Rufus nor the weight to say no. For now the volumes are small: AI assistants send a low single-digit share of most sites’ visits. But those visits more than doubled in a year to US retail sites, and over Prime Day they converted 40% better than visits from search, email, or social.
Two questions for the next twelve months. Will the assistant’s cut replace the ad budget, or come on top of it? And, as a16z’s Alex Immerman and Santiago Rodriguez ask, how many of the orders an assistant brings are new, and how many are orders the merchant would have had anyway, now with a fee attached?
Amazon opens the back door
Amazon now helps its sellers sell on eBay, Walmart, TikTok Shop and Shopify, from the Amazon dashboard, for free. This is not generosity. A sale made elsewhere can still run on Amazon’s warehouses, Amazon’s delivery and Amazon’s software, each one billed, and Amazon now sees what each seller earns on every rival.
🔧 The facts first. Most of what is bought on Amazon is not sold by Amazon: more than 60% of it comes from independent sellers. Those sellers pay Amazon for three things: a commission on each sale, storage in its warehouses, and delivery. In 2025, these seller fees brought Amazon $171.4bn worldwide, two and a half times what it made from advertising. Selling services to sellers is Amazon’s biggest business after its own shop. On 24 September, at its Accelerate conference, Amazon announced that Seller Central, the seller’s dashboard, now manages a seller’s business on eBay, Shopify, TikTok Shop and Walmart as well. The seller writes a product listing once, and Amazon adapts it for each site. Orders from all channels appear on one screen. One click sends them to Amazon’s warehouses for delivery, a paid service called Multi-Channel Fulfillment. And a profit page shows what the seller earns on each channel. It is free, and rolling out to US sellers first.
🏗️ Why would Amazon do this? Because more than 95% of its sellers already sell somewhere else, and Amazon has decided to make money on those sales too. For years it tried to stop them: a seller who used Amazon’s warehouses to deliver an order from another site paid a higher fee for it. Since 2023, Amazon has done the opposite. It now offers to do the work for those other sales, and charges for it. It will ship the goods from the factory, store them, and deliver them, whether the order comes from Amazon or not: that is Supply Chain by Amazon, launched in 2023. Its delivery service for orders from other sites, Multi-Channel Fulfillment, served 200,000 US merchants in 2024, with orders up 70%, and since 2025 it delivers orders placed on Walmart, Shopify, and Shein. By the end of this year, one shipment from a factory into an Amazon warehouse will be able to serve eight countries, including the UK, Germany, France, Italy and Spain. The new dashboard is the last piece: after the warehouse and the delivery van, the software. The logic is simple. If a seller is going to sell on Walmart anyway, Amazon would rather store, ship, and manage that Walmart order itself, and be paid for each step. The seller sells less on Amazon’s site, but every sale still runs through Amazon’s system. And the profit page gives Amazon something it never had: a view of what each seller earns on every competing site.
🇪🇺 Three consequences.
First, sellers are less independent than the word “multichannel” suggests. Among sellers who started on Amazon and now also sell on other marketplaces, 71% still make three quarters or more of their revenue on Amazon. With the new dashboard, the remaining quarter is managed, stored and shipped by Amazon too.
Second, a whole category of software just became free. Tools that let a seller manage several marketplaces from one screen were a business: Shopify sells one, Marketplace Connect, and Europe has specialist vendors such as ChannelEngine, which raised €45.5m in 2022 to build exactly this.
Third, Europe. The profit page tells Amazon how much each seller makes on Walmart and TikTok. In Europe, Amazon cannot use that information: in December 2022 it promised the European Commission to stop using sellers' private sales data to help its own retail business, and since September 2023 the Digital Markets Act forbids it. The tools are US-only for now. If Amazon brings them to Europe, it can offer the dashboard, but it will have to prove it never looks at the page.
🔗 Ecommerce Europe and EuroCommerce: European e-commerce grew 7% to €911bn in 2025, and slows to 5% in 2026 - Only 4% after inflation. The growth is now in the South and the East.
Marketplace Pulse: Amazon’s top sellers aren’t TikTok Shop’s - Only one in five of TikTok Shop’s top 100 sellers make Amazon’s top 10,000. Two different games.
Shein’s first results as a listed company - Quarterly profit down 67%, European sales down 14% in the quarter before the €3 duty. The shares have lost more than a quarter of their value since the Hong Kong listing on 1 September.
Adobe: US online holiday sales to reach $275bn, with AI-driven traffic up 130% - The first holiday season with a forecast line for shoppers who arrive from a chatbot.
H&M: profit up 23% on flat sales - Gross margin at 54%, helped by tariff refunds, and the Persson family now above 68% of the shares. Watch for a take-private.
👍 If you like Unpacking Commerce and want to help it grow, please share this newsletter with your colleagues, followers, and friends. Have a great day!
About us
Spring Invest is a European investment fund dedicated to companies that are shaping the future of commerce. We champion doers who build innovative companies making commerce better, from enabling technologies to new commerce models and everything in between. More info about our investment thesis 👉 here.