As of November 5, 2026, the fight over agentic shopping has produced two opposite answers in plain sight. Amazon blocks Meta's Muse from buying on its site, arguing the agent does not identify itself and appears to store customer logins; Walmart opened an Agent Gateway on October 21 that lets approved assistants including Muse, ChatGPT and Gemini search and check out, with spend limits of 50, 100 or 200 dollars set by the shopper. Shopify says Muse will be able to complete purchases through Shop Pay. The split matters because Amazon's advertising business, roughly 70 billion dollars last year, depends on a human scrolling past a sponsored slot.
An agent does not scroll past a sponsored slot
A sponsored listing works because a shopper sees it while browsing. An agent does not browse; it queries, filters and compares attributes, then buys. That changes what the product page is for. Price, availability and specifications are read as data, and reviews are read as evidence, which means the listing that wins is the one whose claims are verifiable rather than the one with the largest ad budget. Amazon's own position is that few customers shop this way yet: chief executive Andy Jassy told analysts that almost nobody does, and pointed to more than 400 million people who used Alexa for Shopping in the past year.
Reviews become the shelf
For brands, the practical consequence is that review content moves from a marketing asset to an input the agent parses. An agent comparing two serums will weigh the number of reviews, the language of complaints and the specificity of claims, because that is the only evidence available in the data. A listing with fifty detailed consumer review entries describing texture and absorption is easier to reason about than one with two thousand star ratings and no text. Sellers who treated reviews as a volume game now have a reason to treat them as structured evidence.
That makes review analysis an operational function rather than a marketing one. A negative review that names a specific product pain point, such as pilling under makeup or a pump that jams, is readable by an agent and will be weighed against a competitor whose reviews do not mention it. Sentiment scores compress that detail into a number and lose the part an agent actually uses. The brands that will hold up in agent-mediated comparison are the ones whose consumer feedback is specific, answered, and consistent across markets, because consistency is what makes a claim verifiable.
Two doors: a gateway and a lock
Amazon's calculation is straightforward. Sponsored products are estimated to make up about two-thirds of its ad business, and they only exist if the shopping journey starts inside Amazon. Walmart's calculation is different: its global advertising business was about 6.4 billion dollars in the fiscal year ended January 2026, roughly a tenth of Amazon's, so opening the door costs it less and may divert sales from a larger rival. Shopify sits underneath both, having committed to checkout through Shop Pay and to the Personal Agent Protocol published with Meta, Walmart, Sierra and Stripe in October.
| Player | Stance | Mechanism | What brands must do |
|---|---|---|---|
| Amazon | Blocked external agents | Site conditions of use; own assistant promoted | Keep listings accurate and review-rich inside Amazon |
| Walmart | Opened to approved agents | Agent Gateway with Model Context Protocol; Agent Profiles | Decide whether agents may buy, and at what limit |
| Shopify | Checkout for agents | Shop Pay plus protocol signatory status | Treat the product feed as machine-readable copy |
| Meta | Owns the agent | Muse with Stripe Link payments | Assume discovery can start outside any retailer |
Spend limits are the new checkout rules
Walmart's design gives the shopper a veto: an agent can be required to ask before each purchase, or operate inside a cap. Those caps are the practical version of a merchant's terms, because they decide which baskets an agent can assemble without a human in the loop. A 50 dollar limit allows a grocery top-up and blocks a headphone purchase; the same agent behaves like two different customers depending on a setting most shoppers will never revisit. For brands selling above the common caps, agent-mediated demand will be thin until the defaults change.
The volume is still small, which is why both companies can afford to disagree. JPMorgan called Muse the most widely used consumer AI application since ChatGPT, but Morgan Stanley's estimate that 10 to 20 percent of US e-commerce could run through such agents by 2030 is a forecast, not a measurement. The nearer signal is advertising: Amazon's ad revenue grew 11 percent in the third quarter, the slowest pace in nearly four years, on LSEG data. If agents were already redirecting shopping, that number would look worse; that it is merely decelerating suggests the shift is real but early.
Four checks for an agent-readable listing
None of this removes advertising; it relocates it. When discovery happens inside an agent, the budget that used to buy a sponsored slot partly moves to the inputs an agent reads: structured attributes, review depth, and third-party evidence such as test reports. The brands that treat their catalogue as data rather than as a brochure will be legible to both a shopper and a machine, and that legibility is the only durable advantage in a channel where the buyer may never see the page.
- Ads need a humanA sponsored slot works when someone scrolls; an agent queries and compares attributes instead.
- Reviews are evidenceText-rich reviews give an agent something to reason about; star counts alone do not.
- Limits decide basketsA 50 dollar cap allows groceries and blocks electronics, shaping demand more than any ad.
For brands selling through marketplaces this quarter, the useful question is not whether agents will shop but which of your listings an agent can actually evaluate. Running that test now — attributes, review text, price per unit, availability — produces a clear list of pages that would lose a comparison they never see. Fixing those pages is cheaper than bidding for a slot that fewer humans will scroll past.
FAQ
Why did Amazon block Meta's Muse?
Amazon says the agent does not identify itself while browsing and appears to store customer logins, which it treats as a violation of its conditions of use.
How is Walmart's approach different?
Walmart opened an Agent Gateway on October 21 for approved assistants, pairing it with agent profiles and shopper-set spend limits.
Do shoppers use agents to buy today?
Volume is still small; Amazon's Andy Jassy said almost no customers shop that way yet, while JPMorgan sees Muse as the fastest-growing consumer AI app since ChatGPT.
What should brands fix first?
Machine-readable attributes, review text with substance, and a clear stance on whether agents may complete purchases.
