Yesterday produced the clearest two-document snapshot of where agentic commerce actually stands in November 2026, and neither document was a product launch.

The first was a Reuters analysis by Juveria Tabassum, Nicholas P. Brown and Arriana McLymore, published November 4, that names the question every major retailer is now quietly working on: how do you let AI agents shop your website without sawing off the advertising revenue and the customer data that fund the entire operation? The second was Riskified’s Agentic Commerce Pulse, a survey of more than 5,100 consumers globally, which found that shoppers have already decided the terms under which they will accept agent-led buying. Riskified calls it conditional autonomy: consumers will hand AI real authority, but only if they set the limits, keep the approval rights, and know whom to blame.

Put the two side by side and the strategic picture comes into focus. The retailers are designing the machine-readable storefront around their ad businesses. The consumers are designing their acceptance around evidence and accountability. And the connective tissue between the two, the actual product data an agent reads before it buys, is the one layer of the stack that nobody is verifying. That gap is the subject of this piece.

The 74.4 billion dollar market machines cannot see

Start with the money, because the money explains everything else. Per Emarketer projections cited by Reuters, US retail media spending is estimated at 74.4 billion dollars in 2026, up 22.6 percent from 60.7 billion dollars last year, and Amazon and Walmart together hold a combined 88 percent of that market. These are the world’s two largest retail media businesses, and the segment is not a side project. Amazon’s advertising arm consistently accounts for about 10 percent of total net sales, and Walmart CFO John David Rainey told analysts and investors in August that advertising, including the Walmart Connect retail media business, along with membership and marketplace operations, accounted for 45 percent of quarterly operating income growth.

That is the asset the agents are walking toward. Morgan Stanley researchers said in July that AI agents could account for 10 to 20 percent of US e-commerce spending by 2030, and the early curve is already bending: agent-completed orders hit a record 1.52 percent of US e-commerce orders in October, per FirstPageSage’s panel data, a 5.6x jump in a single quarter.

The threat is structural, not cyclical. Retail media sells human attention: the sponsored tile, the banner, the badge a shopper’s eye lands on while scrolling. A shopping agent does not scroll, does not impulse-click, and does not experience visual flair. It ingests structured attributes, compares them against a query, and returns a shortlist. Every dollar of retail media built on capturing attention in a browser is a dollar aimed at a machine that does not have attention to sell.

For now, the effect is officially invisible. Amazon CEO Andy Jassy said last week that there has been no meaningful impact on Amazon’s ads from third-party agents. The honest backdrop: the company’s 11 percent year-over-year advertising revenue growth in the third quarter was its slowest pace in nearly four years, per LSEG data. Cause and effect are unproven. Direction is not in doubt.

Two doors, one landlord

The Reuters piece frames the two superpowers’ responses as opposite strategies, and on the surface they are. Amazon banned Meta’s Muse from its retail site in September, saying the agent appeared to have accessed its online store without authorization, capturing and storing customer credentials; Meta declined to comment. Amazon also won a court order blocking Perplexity’s shopping agent Comet in March, a ban the Ninth US Circuit Court of Appeals overturned in August in a decision Amazon said it disagreed with; the case is ongoing. The company’s stated position to Reuters is that third-party agents “should operate openly and respect service providers’ decisions about whether or not to participate.”

Walmart went the other way. On October 21 it rolled out infrastructure for third-party agentic commerce: when a shopper uses an agent to browse, Walmart creates a profile for that agent linked to the shopper’s account, with purchase approvals and spending limits. It is officially partnering with Meta’s Muse rather than litigating it.

But read past the tactics and both are protecting the same balance sheet. Amazon’s door keeps sessions inside Alexa for Shopping, where it controls monetization end to end. Walmart’s door admits outside agents through a gateway it controls, with the shopper relationship, the data, and the ad surfaces still on Walmart property. One blocks the agents it cannot monetize and promotes the one it can. The other charges tolls at the bridge. Neither is contemplating the one future that would actually hurt: agents shopping a neutral, unsponsored product graph.

The answer taking shape: advertising becomes structured data

Here is the part of the Reuters analysis that deserves the most attention, because it is where retail media and agent infrastructure converge. Walmart has started testing ads inside Sparky, its AI shopping chatbot, just as Amazon has embedded advertising in Alexa for Shopping. Advertising aimed at bots, Reuters reports, “may rely less on visual flair or emotional resonance and more on specific product data” that matches customer queries. Procter & Gamble, Nestle and L’Oreal are already working to make product information digestible for AI agents. Mondelez’s vice president of global digital commerce, Andrew Lederman, said on a podcast in April that agents are bringing brands closer to consumers, while noting that retail media sites still drive the vast majority of the company’s volume.

Follow that thread to its end and you get the industry’s real answer to the conundrum. The ad does not disappear when the human stops looking at the shelf. It migrates into the data feed the agent reads. The sponsored signal stops being a tile and becomes an attribute: a bid expressed in price, delivery promise, specification match, or position inside the answer. Dan Marc, CEO of retail media platform Footprints AI, told Reuters he expects Amazon and Walmart to start signing their largest advertisers to multiyear deals to secure those revenues, and expects discounts on digital retail media as shopping shifts into large language models. Morningstar analyst Brett Husslein calls the moment “an evolution rather than an existential threat” for retail media. Amie Owen, chief client officer at digital commerce consultancy Flywheel, put the agency view plainly: “We are not recommending that clients fundamentally shift their retail media budgets overnight. But we are advising them to prepare for how discovery and purchase decisions may evolve.”

Kearney senior consultant Katherine Black goes furthest, envisioning retail media evolving into a “bidding system” where agents search for a product at a consumer’s price point and retailers set price ceilings and floors. Her closing line to Reuters deserves to be quoted in full: “All retailers need to be thinking about how to interact with consumers in new ways.”

Understand what that means. The attention auction becomes a parameter auction. The ad slot moves off the screen and into the agent’s objective function, and what is auctioned is no longer a human glance but a position in a machine’s reasoning. If that sounds abstract, recall that we have already watched this movie at smaller scale: Sponsored Prompts inside Amazon’s Alexa for Shopping, and Zalando planning to fund its cross-retailer agent layer with retail media sold against rival shops’ inventory. The referee selling ads inside the decision loop is becoming the default architecture of agentic commerce, not the exception.

The consumer has already set the terms: conditional autonomy

Against that machine, the Riskified data is the counterweight, and it is remarkably consistent for a market this young. The Pulse report polled more than 5,100 consumers globally as agents like Muse and Instinct attract mass attention, and the pattern it finds is a widening gap between using AI and letting AI transact:

  • 80 percent have used AI for some portion of the shopping process
  • 74 percent would allow agentic purchases in some form, and 68 percent are likely to use AI for holiday shopping
  • Yet only 11 percent have ever completed a purchase inside an AI tool, and only 7 percent are comfortable letting AI buy without their approval
  • Half of consumers say they are more comfortable than they were six months ago with an agent completing a purchase without immediate approval, so the direction of travel is toward autonomy
  • On control: 43 percent want to approve every purchase before it happens, while 29 percent would allow automatic purchasing within rules they set

Then comes the trust map, and it is bad news for the retailers building the gateways. One third of consumers would trust a general AI platform such as ChatGPT or Gemini to run a shopping agent, against 17 percent who would trust a retailer or brand they already shop with, and 9 percent who would trust a bank, credit card company or digital wallet. The retailers opening their doors to agents are the parties consumers trust least to run them.

And then comes the blame gap, which is the finding every agentic roadmap should be built around. When AI makes a mistake, 51 percent of consumers surveyed said the AI platform should be responsible for issuing refunds, even though Muse and Instinct both state in their terms and conditions that users are responsible for the actions their agents take. Riskified CMO Jeff Otto compressed the problem into one sentence: “Consumers are getting comfortable delegating shopping to AI, but they haven’t delegated the blame.” More than half expect the AI platform to refund a mistaken purchase, the platforms’ terms say the user is liable, and, as Otto notes, in practice that dispute lands on the merchant. His warning about where this goes next is worth quoting directly: “Agentic commerce only scales if merchants can tell a real customer using a legitimate agent from a hijacked account or malicious automation.”

Consumers also want disclosure: 68 percent said they would always want to know when they are interacting with an AI agent rather than a human. Keep that number in mind for the next section, because it applies with equal force to sponsored answers.

The field report: the door is still broken, which buys a window

For anyone tempted to think the ad question is imminent rather than structural, the week’s viral field report is the corrective. As i10x documented, a San Francisco tech worker’s November 3 post complaining that AI shopping agents “basically do not work at all” for her, that she felt “like I’m using theranos,” drew thousands of likes and a thread full of failure cases: a Sephora bot check that stopped the agent cold, an AutoZone order cancelled after roughly 17 hours of fraud review because it combined guest checkout, a one-time virtual card and mismatched names. A report from The Paypers with Riskified, published November 3, found that up to 34 percent of the fraud signals merchants normally rely on were missing in ChatGPT Instant Checkout orders. Visa’s own data, cited in the same coverage, puts the gap plainly: 54 percent of Americans have used AI to help them shop, but only 23 percent trust it with payments.

So the honest state of play is: agents are winning research and losing checkout, retailers are rebuilding the front door, and the ad industry is redesigning its product for a machine reader that cannot yet reliably complete the purchase. That gap is a window. The standards being drafted now, while volume is still small, will decide what the machine-readable storefront looks like when Morgan Stanley’s 10 to 20 percent arrives.

What nobody is pricing: the difference between an ad and an answer

Here is the argument this publication exists to make, and this week’s documents finally make it concrete. The entire emerging retail media model assumes that agents consume structured product data as input, and that brands will pay to influence that data. Fine: advertising is legal, disclosure is the mechanism that keeps it honest. But sponsorship that rides inside a data feed without a machine-readable sponsored flag does not read as an ad to the shopper at the other end. It reads as an answer. The agent launders the placement into a confident recommendation, and the 68 percent of consumers who demand to know when they are dealing with an agent have never been offered the parallel right: knowing when the agent’s answer was bought.

The evidence layer beneath makes this worse, not better, because the corpus agents summarize is the least verified dataset in commerce. Amazon’s own enforcement disclosures put blocked suspected fake reviews above 275 million in 2024 alone, independent detection estimates of inauthentic reviews on the platform range from roughly 7 to 42 percent depending on method, and Pangram’s analysis found that 93 percent of first-page AI-generated reviews carried the Verified Purchase badge, which means the badge agents and shoppers lean on most certifies nothing about authenticity. Consumers, for their part, have already told the market what they want instead: 73 percent trust an AI recommendation more when it shows evidence such as reviews, specs or sources, per the Harris survey data assembled by SearchPanel.

Now assemble the failure case from this week’s numbers. An agent, optimizing inside a bidding system where retailers set price floors and brands pay for data placement, recommends a product whose rating was manufactured. The shopper, operating under conditional autonomy, approved it because the answer looked well-evidenced. The product disappoints. The shopper invokes the 51 percent expectation and demands a refund from the AI platform. The platform’s terms point back at the user. The merchant eats the chargeback. Conditional autonomy, the entire adoption curve Riskified measured, dies at the first bad purchase, and it dies because nobody verified the inputs.

What honest machine-readable commerce requires

The window while checkout is still broken is the time to build the three properties that separate a machine-readable storefront from a machine-exploitable one:

  1. Sponsorship that survives the pipeline. A sponsored flag is only honest if it travels with the data all the way into the agent’s answer, in a form the agent must represent to the shopper. Anything less converts advertising into invisible ranking manipulation. The 68 percent disclosure demand applies to answers, not just agents.
  2. Evidence with provenance. Reviews must be filtered for authenticity before they are counted, weighted, or summarized, and the shopper should see review quality, not just review quantity. A fluent summary of a contaminated corpus is confidence laundering, whatever the intent behind it.
  3. A referee that does not sell ads. No verification layer can be neutral if it is owned by a platform whose revenue depends on the outcome it verifies. The trust input to an agent’s reasoning has to come from outside the bidding system.

This is the design brief GoBuy was built against. The Smart Score from 0 to 100 is computed from review quality, not review volume and not ad spend, with fake reviews filtered out and authentic ones weighted up. GoBuy deliberately shows only the top seven products per category instead of thousands, because an agent fed a sponsored infinite shelf just inherits the manipulation at machine speed. The MCP server at gobuy.ai/api/mcp exists so that any shopping agent can consult an independent trust layer before it commits a purchase, and the GoBuy Verified badge marks products that have held a score of 80 or above across 90 days, a bar earned from sustained review integrity rather than a media budget. For shoppers who stay human, the Chrome extension injects the same trust panel directly onto Amazon product pages.

The standards being written this quarter

Kearney’s Katherine Black is right that all retailers need to be thinking about how to interact with consumers in new ways. The sharper version is this: the bidding system is coming either way, and the only open question is whether the machine-mediated shelf shows the best products or the best-funded ones. Walmart’s gateway, Amazon’s litigation, Sparky’s ad slots and the payment networks’ agent identity protocols are all being negotiated while agent orders sit at under 2 percent of e-commerce. By the time they reach Morgan Stanley’s range, the defaults will be locked in.

If you build or operate a shopping agent, wire it to a trust layer that has no stake in the answer: point it at gobuy.ai/api/mcp and read the integration docs at gobuy.ai/agent-docs. If you shop with your own eyes, install the GoBuy extension and check the Smart Score before the review section convinces you. Either way, the answer to “which one should I buy” should start with evidence, not with whoever paid to be the answer.