The least interesting thing about this week AI-shopping news is which retailer likes artificial intelligence. The number that explains the industry split sits inside Amazon second-quarter report: $19.8 billion of advertising revenue in three months. That money is tied to the search results, sponsored listings and upsells on Amazon shopping pages - precisely the layer an outside AI agent is built to skip. Tell an agent to buy something and it goes straight to the item, past the ads.
That is the whole disagreement in one line, and it is why Amazon and Walmart chose opposite policies on the same news cycle. On October 21, at the Masters of Scale summit in San Francisco, Walmart CEO John Furner unveiled Agent Profiles, rolling out in beta within the next month. Every AI assistant a customer uses gets its own profile inside the Walmart account, so agent purchases do not mix with personal ones. Shoppers can require approval for every purchase or let the agent buy autonomously up to a $50, $100 or $200 cap, and delete the profile at any time. Payment-card data never reaches the agent, and Walmart says rule-breaking agents get banned. The company also launched the Walmart Agent Gateway, connecting OpenAI ChatGPT, Google Gemini and Meta Muse to Walmart and Sam Club systems.
Amazon wall, built in stages
Amazon line has been hard and consistent. It cut off Meta Muse in September, two weeks after the assistant launched. It sued Perplexity in November 2025 over Comet, won a preliminary injunction in March 2026, and watched a federal appeals court vacate it in August on the grounds that it was the customers - not the agent - doing the buying. In March 2026 Amazon rewrote its seller agreement to formally define an Agent, require automated systems to identify themselves, and cease access on request. Since July it has stripped item names from confirmation emails and expanded robots.txt to block 47 AI bots. None of this makes Amazon anti-agent: its own Alexa for Shopping reached 350 million users. The strategy is a controlled front door behind a walled storefront.
The market is splitting in two
The retail landscape has bifurcated. eBay updated its user agreement in February 2026 to prohibit unauthorized buy-for-me agents. Kohl hard-blocks agents; Tapestry allows browsing but not buying; Gap whitelists case by case. On the other side, Walmart, Target, Best Buy, Home Depot, Macy, Gap, Sephora and Etsy have chosen integration, and Shopify has opened every merchant to Muse. Forkast frames it as a new tollgate: on top of protocol integration costs ($5,000 to $500,000 per checkout protocol), retailers can simply refuse to serve an agent - a retailer access tax granted or denied at each company discretion.
The honest caveat
Adoption data should temper both strategies. A survey cited in coverage found only 16 percent of shoppers comfortable with an AI assistant buying for them, and an NMI poll put trust in agents completing purchases at just 3 percent of U.S. adults. None of this has reached an income statement. Mizuho analyst David Bellinger wrote that Walmart tools strengthen confidence in agentic purchasing, which he still calls experimental.
Editorial view
Do not read the split as Amazon being behind or Walmart being clever. Read it as two bets on where agentic shopping routes value. Amazon tries to keep the agent inside its own funnel, where it keeps visibility and a cut of discovery; Walmart tries to own the transaction an agent completes no matter where that agent learned to shop. Which side is right depends on whether consumers delegate the boring part of shopping to software, and which retailer becomes the one an agent cannot shop without. For B2B sellers, retail access is now a commercial negotiation, not a technical one, and integration channels are being drawn on both sides of a widening line.
Sources: AInvest, Forkast News, Nasdaq / Motley Fool, Amazon cease-and-desist letter, Amazon Seller Central, WSJ retailer survey.
What the Walmart mechanism actually does
The operational detail matters as much as the policy direction. Walmart's Agent Profiles keep every AI assistant in a separate account profile, so agent-initiated purchases are logged, capped and reversible in a way a human checkout is not. Requiring approval per purchase preserves human control at the final step, while the $50/$100/$200 autonomous caps let retailers test how much shopping behavior actually migrates to agents without surrendering payment control - card data never reaches the assistant, and any agent that breaks the rules is blocked. The seller side is watching just as closely. Amazon's decision to block outside agents protects the discovery and advertising layer that generated $19.8 billion in quarterly ad revenue, and seller-forum discussions in the weeks after the Muse cut-off suggest merchants are reading the Agent Policy changes as a sign that access to their own listings will increasingly be negotiated channel by channel. Whether Walmart's open-door approach converts into sales depends on adoption that is still thin: fewer than one in five U.S. shoppers say they are comfortable with an assistant buying for them, so the near-term commercial impact of either strategy is likely modest - which is exactly why the positioning matters more than the revenue today.
