AI Shopping Agents Split Amazon and Walmart Over $74B Ad Market

The bots are coming for the sponsored listings

For ten years, retail media has grown on a simple idea: shoppers scroll through a retailer's pages, and brands pay to be at the top of the list. AI shopping agents could break that idea. These are bots that research, compare and buy products for a person. They don't scroll, and sponsored placements mean little to them. Amazon and Walmart now have to decide how much access to give these agents, and they have made opposite choices.

The two companies dominate this market. Emarketer projects US retail media spending of about $74.4 billion, and Amazon and Walmart together hold 88% of it.11 In July, Morgan Stanley researchers estimated that AI agents could handle 10% to 20% of US e-commerce spending by 2030.11 Ads that only work when a human is looking lose value if even a small share of that spending moves to software buyers. Finimize puts the problem plainly: when an agent does the browsing, the people looking at sponsored slots are gone, and auction-based ad prices could weaken.19

Amazon shuts the door

Amazon has taken the most defensive position of any major retailer. In September it banned Meta's new shopping agent, Muse, saying the tool had accessed its store without authorization and captured and stored customer login credentials.11 Amazon said Meta never told it Muse would access the store and that the agent did not identify itself. Users who tried it saw a pop-up saying that access by an unauthorized AI agent violated Amazon's Conditions of Use.20 In mid-October, users of Grok Bot reported the same error message.20 Amazon had already blocked Perplexity's Comet browser and agents from Google and OpenAI, while running its own Buy for Me and Alexa shopping agents.7

The legal side is unsettled. Amazon won a court order against Perplexity's Comet agent in March, but the 9th US Circuit Court of Appeals reversed that ban in August. Amazon says it disagrees with the ruling, and the case continues.11 Amazon's stated position is that outside agents should operate openly and respect a platform's decision about whether to take part.13

Publicly, Amazon says the threat is small. CEO Andy Jassy said last week that third-party agents have had no meaningful effect on its ad business.11 He also argued that Amazon's own Alexa for Shopping will be the main agent customers use on Amazon, and said more than 400 million people used it in the past year.20 The numbers make that reassurance harder to accept. According to LSEG data, Amazon's ad revenue grew 11% year over year in the third quarter, its slowest growth in nearly four years.11 The Wall Street Journal reported that Amazon earned nearly $70 billion from advertising last year, much of it from sponsored placements shown to shoppers.20

The slowdown can't be pinned on agents yet. But Amazon is blocking agents at the same moment its most profitable growth business is slowing, so its caution looks like protecting revenue, not just policing security. Both explanations can be true. The credential-capture allegation against Muse is serious. Still, an Amazon that profits when people browse its pages has every financial reason to keep the browsing on those pages.

Walmart opens it, on its own terms

Walmart has gone the other way. On October 21 it launched infrastructure that lets third-party agents shop its site. Each agent gets a profile linked to the shopper's Walmart account, and customers can approve each purchase or set spending limits.11 Walmart's example limits are $50, $100 and $200, and it says it may still ask for approval below the limit. Agents never receive card numbers, cannot add or change payment methods, and can be blocked if they break the rules.20 Walmart has also partnered with Meta's Muse, the same tool Amazon banned.11

Calling this the "open" approach overstates it. Walmart decides who gets in, what data each agent sees and what happens when an agent breaks the rules. PYMNTS CEO Karen Webster said the two giants are closer than they look, because in both cases agents come in on the owner's terms.20 The real difference is business model. Walmart wants extra orders and is willing to manage agents to get them. Amazon makes a lot of money from browsing on its pages.20

Walmart also has a lot to lose. CFO John David Rainey said in August that advertising, which includes the Walmart Connect retail media business, together with membership and marketplace, drove 45% of quarterly operating income growth.11 Walmart doesn't report Walmart Connect separately, so the size of its exposure isn't disclosed.17 By giving agents a sanctioned way in, Walmart is betting it can earn trust with AI developers while keeping control of the customer account. BigGo Finance notes the downside: partners get a role in a shopping journey Walmart used to control on its own.12

Why the payments rails matter more than the press releases

The argument over agent access sits on top of a payments system that is being built quickly. In June, Visa announced a deal with OpenAI that lets ChatGPT agents pay with tokenized Visa credentials, limited by spending caps, merchant categories and user approvals.1 Mastercard has expanded Agent Pay with a score that estimates whether an AI agent started a transaction. It is also working with Cloudflare and Skyfire on agent verification.1 Visa, American Express, Stripe and PayPal are all building competing trust layers. The contest is now about who can prove an agent did what its buyer asked, and who is responsible when it didn't.2

Coverage of how far along this is varies. One industry guide describes Visa Intelligent Commerce as live in the US and Europe, with agent purchases running through more than 30 European issuers.6 An October review was more cautious. It found that most live transactions were still pilots or controlled tests, and a fraud vendor counted only "hundreds" of completed agent-initiated transactions worldwide by September.5 The Visa–OpenAI checkout had no consumer launch date as of October.10 Consumer trust is also low: Visa's own Trust Index found only 23% of US consumers trust generative AI to handle payments for them.10

The cautious view is the more accurate one: the infrastructure is ahead of actual use. OpenAI has already pulled back once, retiring ChatGPT Instant Checkout in March, about six months after launch, and sending checkout back to merchants' own sites.7 Liability for wrong purchases is still unresolved. Under Visa's rules, the cardholder is responsible for the agent's actions, so "my agent did it" is not a fraud claim on its own.5 Walmart's guardrails, including spending caps, no card numbers and normal returns, look built to handle exactly these open questions.

Ads will follow the agents, not die

Retail media isn't going away. Morningstar analyst Brett Husslein called the moment "an evolution rather than an existential threat." Emarketer forecasts that brands will spend 22.6% more on retail media in 2026 than the $60.7 billion they spent last year.11 Both retailers are already putting ads into their own agents. Walmart is testing ads in its Sparky chatbot, and Amazon has placed them in Alexa for Shopping.11

What changes is the ad itself. Practitioners expect ads aimed at bots to depend less on visuals or emotional appeal and more on exact product data that matches a shopper's request. Procter & Gamble, Nestle and L'Oreal say they are rewriting product information so agents can process it.11 Footprints AI CEO Dan Marc expects the big retailers to sign their largest advertisers to multiyear deals and to cut prices on digital placements.11 Finimize notes the catch: long contracts make revenue steadier but limit upside compared with a strong auction market.19 Kearney's Katherine Black describes a bigger change, with retail media becoming a bidding system where agents look for products at a shopper's price and retailers set price floors and ceilings.11 That scenario would put retailers in a similar position to the search engines that brands have long bid against.

The robots behind the bots

An agent can place an order, but someone still has to deliver it, and both companies are spending heavily on fulfillment. On October 29, Walmart expanded its deal with Symbotic, committing about $6 billion over several years to automate 25 US regional distribution centers.25 The expansion comes as Walmart works through a hard retrofit. The Journal's reporting describes boxes too large for the machines, wheeled robots that break down, dust that can halt systems in a chain reaction, and power bills that can top $800,000 a month at automated grocery warehouses.28 Symbotic says the share of its robots under repair at any one time has fallen from under 10% to about 5%.24

Amazon's starting point is stronger, although reported robot counts differ. One account cites more than 750,000 robots.23 Other reporting says Amazon passed one million in 2026.2526 At its Tracy, California site, robots move pods, sort parcels and pull carts. Headcount has still grown from about 1,000 to more than 3,000 since 2023, and humans still pack orders, load trailers and fix problems when the automation fails.21

The two fights are connected. If agents make shopping a contest of price and product data, with less room for sponsored placement, retailers will compete more on cost and delivery speed. That puts more weight on warehouses. Walmart's strategy is consistent: let agents in, keep control of the account, and spend heavily on robots so it can win on fulfillment. Amazon is betting that its own agent, its scale and its legal fights can keep shoppers on its pages. For now the ad money keeps growing. What's uncertain is whether brands keep paying for placements once a growing share of buying is done by software.