Amazon says its own shopping agent, Alexa for Shopping, “was used by 400 million+ customers in the last 12 months, with customer interactions up 570% year-over-year”, according to its third-quarter earnings release of 29 October 2026. Walmart, in Agentic Access Terms marked as its 21 October 2026 version, requires a human to approve every checkout by an outside AI agent unless that agent shops through Walmart’s own agent, Sparky.
Up 570% means customer interactions with Amazon’s agent ran at about 6.7 times the level of a year earlier, by our arithmetic.
Business Engineer Course · The FDE Playbook
An evergreen Business Engineer course on forward-deployed engineering, how AI gets put to work inside organisations. It teaches the general idea only and does not discuss Amazon’s or Walmart’s disclosures.
The key insight:Amazon reports that 400 million+ customers used its own shopping agent, Alexa for Shopping, in 12 months, with interactions at about 6.7 times a year earlier by our arithmetic, while Walmart’s 21 October terms require a human consent gate before every checkout by an outside agent unless it shops through Walmart’s agent, Sparky.
What happened next (30 Oct): DoorDash on outside agents.TechCrunch reported on 30 October that DoorDash wants to learn more about agentic use cases and “develop a healthy relationship with third-party agent makers”, and that chief executive Tony Xu said there could in future be an opportunity for DoorDash to advertise within these agents. Users still predominantly order through DoorDash’s app, TechCrunch reported.
On the earnings call, Andy Jassy said, per a machine transcript published by Gloom, that US customers using Alexa for Shopping “are spending 50% more per order than those who don’t”. He also said Amazon anticipates working with horizontal agents but believes Alexa for Shopping “is going to be the primary shopping agent customers use to shop Amazon”.
On DoorDash’s own third-quarter call the same day, chief executive Tony Xu said, according to the call transcripts, that the company is “seeing maybe several hundred orders in total from some of these third-party agents”. We covered both calls in our DoorDash and Amazon piece.
Alexa for Shopping as Amazon reported it for Q3 2026: customer interactions up 570% year over year (about 6.7 times, our arithmetic) and, per the call, US users spending 50% more per order than non-users.
What Was Posted on X
Amazon made two things plain in its earnings call last night:
1) It expects its on-site agent, Alexa for Shopping, to be “the primary shopping agent customers use to shop Amazon,” meaning it seems unlikely to meaningfully open its catalog to third-party agents for agentic shopping tasks, and 2) Alexa for Shopping has seen significant adoption by Amazon customers, with over 400MM customers having used it in the trailing 12 months.
To my mind, the notion that (third-party) Agentic Commerce is a Mirage has already been demonstrated. For instance, Tony Xu noted in the DoorDash call this week that third-party agents had completed “several hundred orders in total” on its platform. There is no “there” there with Agentic Commerce; platforms have an inherent, insurmountable data and capabilities advantage with shopping tasks and their own agents will outperform the third-party agents that are even given access.
And those third-party agents are increasingly starved of access. Walmart recently published its Agentic Access Terms, and the company forces third-party agents through Sparky for autonomous checkout, meaning they effectively just remit Sparky output. Jassy noted on the Amazon call that the company does “anticipate working with horizontal agents,” but my sense is this will look either like *advertising* deals (Amazon struck such a deal with ChatGPT last month) or Walmart’s access terms, broadcasting Alexa for Shopping into third-party surfaces.
I’ve seen analysis attributing the deceleration in Amazon’s advertising business (26% in Q2 to 11% in Q3) to third-party agentic commerce. This is absurd on its face. First, Jassy said as much on the call. Second, Amazon *blocks* external agents, so by definition, endemic brands would not see their ad spend displaced by third-party agentic discovery. And third, and most importantly: Prime Day moved from Q3 to Q2 in 2026.
Third-party agents will fulfill the discovery role in commerce, which means that advertising is the most natural business model for them. Everything is an ad network.
Eric Seufert wrote on X on 30 October that third-party agentic commerce is a Mirage and that platforms have an inherent, insurmountable data and capabilities advantage with shopping tasks. That is his view.
TechCrunch’s post on its report: it says DoorDash said that it is open to working with third-party agent makers.
What Walmart’s Terms Require
Walmart’s terms define an agent as any software that “autonomously or semi-autonomously accesses or interacts with the Walmart Sites” for a customer, including AI assistants, browser-automation tools and API-connected services. They state: “Except for transactions utilizing the Walmart Agent Gateway, Walmart will require a Human Consent Gate to confirm the Authorizing Customer’s approval of Agent actions before every checkout”.
The Walmart Agent Gateway is, in the terms’ words, an MCP or similar connection through which an agent engages in “agent-to-agent shopping with Walmart’s agent, Sparky”. The terms add that “Agents may not check out autonomously unless they comply with the requirements for the Walmart Agent Gateway” and that Walmart may block agents that do. Inside the Gateway, the terms say a User-Defined Guardrail “may allow the Authorizing Customer to configure per-transaction spend limits, category or department restrictions, transaction frequency”, and each session needs its own access token.
Eric Seufert wrote on X on 30 October that third-party agentic commerce is a “Mirage” and that platforms have “an inherent, insurmountable data and capabilities advantage with shopping tasks”. That is his view; on Jassy’s comment about horizontal agents, he wrote that his sense is it will look either like advertising deals, such as the one Amazon struck with ChatGPT, or like Walmart’s access terms. Related coverage: banks blocking AI agents and Shopify’s Shop Pay for agents.
The Structural Read
As we read the two disclosures, Amazon reported usage of its own agent while Walmart set out the rules for everyone else’s: its terms name a single route, the Walmart Agent Gateway to Sparky, for outside agents to check out without a human step.
As we read the numbers, the contrast is in units as well as scale: Amazon counts 400 million+ customers of its own agent over 12 months, while DoorDash counted several hundred orders from outside agents in total, so the two are not a ratio.
As we read Seufert’s post, his argument that third-party agentic commerce is a mirage is an opinion; the filings and terms show what the platforms built and allow, not how outside agents will fare.
Walmart Agentic Access Terms, 21 October 2026 version
“Agents may not check out autonomously unless they comply with the requirements for the Walmart Agent Gateway set forth in Addendum 1.”
Three Implications
FOR AGENT BUILDERS Walmart’s terms say an outside agent needs a human consent gate before every checkout unless it uses the Walmart Agent Gateway, where each session needs its own access token.
FOR AMAZONAmazon reports US customers using Alexa for Shopping spend 50% more per order than those who don’t, per the call machine transcript, a comparison of users with non-users.
FOR ADVERTISERSAmazon Ads said on 10 September 2026 that select US advertisers are testing ChatGPT Ads as part of a pilot through its OpenAI partnership.
The framework says “The agentic surface โ the harness โ is becoming where value pools in the entire AI stack.”
The framework is the owner’s argument about AI agents, not evidence about this story; it does not discuss Amazon’s or Walmart’s disclosures.
What Is Not Established
Amazon’s 400 million counts customers who used its agent over 12 months; DoorDash’s several hundred counts orders placed through outside agents in total. The two measure different things and are not comparable as a ratio. Walmart’s terms set rules for access; they do not say how many outside agents shop on Walmart today. The 50% spend figure compares users with non-users and does not by itself show that the agent caused the difference.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Amazon reports 400 million+ users of its own shopping agent, with interactions at about 6.7 times a year earlier by our arithmetic, while Walmart’s 21 October terms make outside agents either get a human’s approval at every checkout or shop through Sparky. On its own earnings call, DoorDash counted several hundred orders from outside agents in total.
A note on sourcing. Amazon figures come from its third-quarter earnings release and, where stated, a machine transcript of its earnings call; Walmart quotes come from its Agentic Access Terms (21 October 2026 version). Figures labelled our arithmetic are our own calculations.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
