AI Shopping Agents Like Meta's Muse Split Retailers Over Checkout

An agent that shops, and a market that reacted fast

Meta launched Muse, its personal AI agent, on September 8. It can search retail sites, compare prices and complete purchases for a user. In about a month it passed 5 million downloads and 3 million weekly active users, and Meta's stock rose more than 17% after the launch.15 Muse also overtook ChatGPT as the top free app on Apple's and Google's U.S. stores.14 A Meta spokesperson said shopping quickly became one of the main reasons people use it.12

The checkout process is built around human approval. A user describes an item, including brand, price or availability if they want. The agent scans retail sites and returns listings. If the user chooses to buy through Muse, it prepares the order and shows an "approval card" that the user must confirm.12 Payments currently go through Link by Stripe and Shopify's Shop Pay. PayPal has been announced as a partner but is not live yet. Mark Zuckerberg has said Meta will take a small fee on each transaction.13

Investors started pricing in the idea almost immediately. From Muse's launch to October 8, Planet Fitness fell almost 12%, Booking Holdings 11%, Charles Schwab 9% and SiriusXM 8.6%. Traders lumped these companies into a "consumer inertia" trade, meaning businesses that depend on customers not bothering to switch or cancel.11 Goldman Sachs describes the shift as moving from "search and shop" to "delegate and approve." It notes that today's AI tools are still mostly used to discover products, though it expects them to handle more of the purchase over time.11

The retail split: who lets the agent in

The biggest question Muse raised is not technical. It is about who controls the customer. Meta announced a set of partner "connectors" on September 23, including Walmart, Gap, Sephora, Expedia, Wayfair and Best Buy, though not all were active at launch.12 Shopify CEO Tobi Lütke confirmed a deal that lets Muse complete purchases across Shopify-powered stores through Shop Pay, and Shopify's stock rose over two trading sessions afterward.14

Amazon went the other way. On September 20, twelve days after launch, it began showing Muse users a pop-up saying the agent was not authorized.22 Reports give somewhat different reasons. One account says Amazon called Muse a violation of its terms of service.13 Another says Amazon objected that the agent did not identify itself and appeared to store customer credentials, which Meta denies.23 Amazon also said it had not authorized Muse to access customer accounts, scrape data or process transactions.29 It has since blocked OpenAI's new Dots agents too.22

Target sits in between. Its terms of use, updated September 10, allow four named agents: its own assistant, OpenAI, Google and Microsoft. All others are prohibited.22

The best explanation for these different positions is how much each retailer earns from advertising. One analysis puts Amazon's ad revenue at about twice its operating income outside AWS. The same measure is about 21% for Walmart and 18% for Target.22 Estimates of Amazon's ad business vary: $68.6 billion in one analysis22, $76 billion over the last twelve months in another17, and an expected $83 billion this year according to FactSet.11 Every estimate leads to the same conclusion, though. An outside agent that buys without ever seeing a sponsored listing undercuts Amazon's most profitable business.17

This is not a refusal to use agents. Amazon renamed its Rufus chatbot Alexa for Shopping in May. Andy Jassy says more than 350 million customers used it over the past year, and that shoppers who click a sponsored prompt inside it convert 48% more often and spend 21% more.22 Amazon wants agentic shopping on its own terms, with ads included.

Walmart is betting it has more to gain than lose. Its U.S. CEO, David Guggina, says shoppers arriving from OpenAI and Gemini convert at very high rates and come back often. Outside models mostly send them to Walmart's own checkout.23 Walmart learned this the hard way. When OpenAI ran checkout inside ChatGPT, purchases there converted at about a third of the rate of shoppers sent to Walmart's site. Walmart now places its own assistant, Sparky, inside ChatGPT and Gemini, so the outside agent helps find products and Walmart handles the cart.2327

Three retail stocks, three different defenses

Analysts mostly argue that the threat depends on the business model, not on whether a company sells online. Rosenblatt's Scott Devitt says retailers with strong value propositions like Amazon are less exposed. Goldman adds that companies with good prices, broad selection, fast fulfillment and strong first-party data are best positioned.11 For Amazon, the risk is losing customer data and ad revenue, not losing the sale itself. Telsey's Joe Feldman notes that Amazon could still fulfill an order while giving up direct contact with the shopper.11

Consumer behavior so far favors Amazon. A PYMNTS survey found that 59.7% of respondents' most recent AI-assisted purchases were completed on Amazon, compared with 19.5% at big-box and department stores and 6.3% on brands' own sites.26 Agents may influence which brand a shopper picks, but the sale still tends to happen at the biggest retailer.

Costco may benefit. Its CFO says sales coming from searches on large language models are growing at triple-digit rates, though from a very small base, and are increasing membership activity. That matters for a company that does little paid digital advertising.11 The risks are that agents cut down on impulse buying in the warehouse and make it easier for members to question whether the fee is worth it.11

TJX may be the best protected because its products are hard for an agent to find. Wells Fargo's Ike Boruchow says an agent cannot hunt for treasure. Guggenheim's Simeon Siegel argues that off-price retailers succeed precisely because they lack e-commerce, and that people will hand off "needs" to agents long before "wants."1115 From Muse's launch to October 8, TJX rose nearly 8%, Costco about 4%, and Amazon slipped 1%.11

The payments land grab

Card networks are moving as quickly as retailers. One executive described their effort as a "land grab for infrastructure standards." In September, Mastercard introduced an option that gives an AI agent a virtual card with limits on spending and merchants.5 On September 30 it added a score estimating how likely it is that a transaction was started by an agent, along with partnerships with Cloudflare and Skyfire to verify agent identity.6 Visa announced Agent Score and an Agentic Registry earlier, and also partnered with OpenAI to let agents pay with tokenized Visa credentials within limits the user sets.1 On the protocol side, Google's Universal Commerce Protocol, OpenAI and Stripe's Agentic Commerce Protocol, and Stripe and Tempo's Machine Payments Protocol all compete to become the common standard.41

The payments infrastructure is ahead of consumers. At a Goldman conference on September 8, the same day Muse launched, Visa CEO Ryan McInerney said people are using AI for shopping "but not yet for autonomous payments": they compare products with assistants and then pay on the seller's site.9 OpenAI shut down ChatGPT Instant Checkout in March, about six months after launching it.4 BrightEdge estimates that agent-driven purchases are growing 43% a month, but only 3% of the merchants it tracks let agents complete a checkout.25 One analysis points out that those two BrightEdge figures do not fully match, and that growth is coming from a very small base.22

Liability is still unresolved. Under every published protocol, the merchant remains the merchant of record and absorbs chargebacks. Neither Visa nor Mastercard has published liability rules for agent transactions.9 When asked who pays if an agent buys something outside a user's instructions, Mastercard pointed to its Verifiable Intent record and did not give a direct answer.5 Technical problems also continue. Some users report trouble checking out at Walmart despite its partnership with Meta.18

The takeaway

The coverage agrees on the direction but not the timing. AI agents already influence what people buy, but fully delegated purchasing is still a longer-term prospect.11 This analysis reads the current stock moves as reasonable for subscription-based businesses but early for most retailers. The real question for the next year is who controls the purchase and its data, not whether agents replace stores. In that fight, the companies with the strongest position are those that own their own agent and checkout, like Amazon and Walmart, or the payment infrastructure, like Shopify. The numbers to watch are completed agent checkouts and merchant sign-ups, which will show whether agentic shopping actually takes market share or stays a novelty.14