More than 10 million people activated Meta’s Muse agents in their first six weeks, AMD chief executive Lisa Su said on AMD’s third-quarter call on 3 November, according to our machine transcript of the webcast. Five days earlier, DoorDash chief executive Tony Xu told analysts that DoorDash was seeing “maybe several hundred orders in total from some of these third-party agents”.
By our arithmetic, 1,000 orders, an upper bound for several hundred, equals about 8 seconds of DoorDash’s average order flow in the third quarter, when its release reports 993 million total orders, or about 125 a second.
Olivia Moore, a partner on Andreessen Horowitz’s investing team who focuses on AI, according to a16z’s site, wrote on X on 4 November that agentic commerce is “undeniably overhyped based on current usage”. She added that she would bet it is “massively underhyped based on volume agents will drive in ~3-6 months”.
Her post quotes one by @maiab, who wrote of never having heard “so much hype and discussion” about a product experience that “basically does not work at all” for the poster, adding: “I feel like I’m using theranos”.
Business Pill · Agentic Commerce
An evergreen Business Engineer explainer on agentic commerce: what changes when the shopper is an AI agent. It explains the general idea only and does not discuss Muse, DoorDash or this week’s launches.
The key insight: More than 10 million people activated Meta’s Muse agents in their first six weeks, AMD’s chief executive said, while DoorDash’s chief executive counted several hundred orders in total from third-party agents, about 8 seconds of DoorDash’s average third-quarter order flow by our arithmetic.
Update, 4 Nov, 09:57Z: Meta says Muse bug bounties have paid out about $1.2 million. In a 2 November post on the Muse blog, Meta’s David Singleton wrote that the program, private since April and public since the 8 September launch, has received more than 630 submissions, identified 64 valid findings and awarded about $1.2 million to date. The largest single award, $300,000, was paid four months before launch for a bug that gave a researcher access to conversations and tokens in pre-release VMs used by staff testing the product; Meta says it shut the service down and fixed it. Other paid reports included prompt injection disguised as Muse’s own tools and a gap in one-time-passcode redaction, at $90,000 each. SecurityWeek reported the post on 4 November.
By our arithmetic, about one submission in ten was valid, the average payout per valid finding was roughly $19,000, and the single pre-launch award was about a quarter of the total. Meta says Muse’s top award is several times higher than other major AI labs’ published maximums, without naming them; against the up to $172,000 including bonuses that OpenAI announced on 2 November, it is about 1.7 times, by our arithmetic. The post follows Meta’s 20 October pause of Muse chat sharing.
Xu was answering an analyst who asked how much DoorDash would cede to “more generic agents like Muse”. He added that “the predominant use case of that is actually through our app”, and did not say which agents sent the orders. Later on the call he said “AI agents are generating several hundred orders across the time that they’ve been up and running”.
Amazon’s third-quarter release says its own agent, Alexa for Shopping, “was used by 400 million+ customers in the last 12 months, with customer interactions up 570% year-over-year”. On outside shopping agents, chief executive Andy Jassy said “few customers are doing much shopping this way yet”, according to TheFly’s report of the call.
Mastercard chief executive Michael Miebach said in prepared remarks on 29 October: “It is still early days, and the numbers are not yet significant in the grand scheme.” He added that “agentic transactions are ramping up at impressive weekly growth rates”, and gave no rate in the transcript we read.
Su’s Muse figure counts people activating agents, not purchases. None of the three companies gave a dollar value for purchases made through outside agents in the passages we read.
Launches for AI-agent shopping (teal) and what executives said about usage (red), 26 October to 3 November 2026, from Instinct’s X posts, xAI’s Grok Bot changelog, Shopify’s Fall ’26 Edition, opencommerce.sh, the DoorDash and Mastercard call transcripts, our machine transcript of AMD’s call, and, for Muse carts, a post on X by Marketplace Pulse’s founder. The 8-second comparison is our arithmetic, with 1,000 orders as an upper bound for several hundred.
What Was Posted on X
Agentic commerce is undeniably overhyped based on current usage
…but, I’d be willing to bet it’s massively underhyped based on volume agents will drive in ~3-6 months.
And, if you’re not optimizing your product (software, service, or physical good) for agents, you’ll be left behind.
Why
1. As Maia’s demo shows – agents are much improved, but still nowhere near perfect. 99% of consumers will churn off an agent if there are any issues or delays. Browser use is still slow / buggy and the ecosystem of connectors / plug-ins available on these agents are nascent.
Similarly – saving down credentials, handling payments, etc. all have too much friction and uncertainty now. This will all be solved over time.
2. The VAST majority of consumers still are not using agents. Pick a person off the street in any city other than SF, and they have likely heard of neither Muse nor Instinct.
These products have (so far) done a fantastic job optimizing for the workflows / use cases of high income, tech-forward SV users. They are still early in doing things that are valuable for the mainstream user on a repeatable basis (retention), especially while being able to offer the product for free.
This will likely involve unlocking network effects (yes, some consumer agents now are multiplayer but usage is fairly low) and giving people ways to make money / spend time vs. save money / save time (@ekuyda’s thesis!) Model costs coming down will help here too.
3. “Agents” right now generally refer to standalone harnesses built for consumers to use (Instinct, Muse, Fo, Tomo, Pally, Folk, Tab, Daimon and many others).
These products will continue to grow – and consumers will interact with agentic models more frequently via products like ChatGPT (now at 1.2B WAUs), that can now complete purchases instead of just recommending products.
As ChatGPT rolls out its wallet and more developers connect plug-ins, I could see a massive increase in % of LLM queries that lead to in-chat transactions. And, early data shows that AI-driven traffic is actually higher conversion and intent than non-AI commerce traffic.
Olivia Moore, whose X bio says she is a partner at a16z, wrote on 4 November that agentic commerce is “undeniably overhyped based on current usage”, but that she would bet it is “massively underhyped based on volume agents will drive in ~3-6 months”. Her reasons: agents are “still nowhere near perfect”, browser use is “slow / buggy”, saving credentials and handling payments carry “too much friction”, and the “VAST majority of consumers” are not using agents, with most people outside SF having heard of neither Muse nor Instinct.
re: agentic shopping I have NEVER in my life heard so much hype and discussion about a tech/product experience that basically does not work at all for me. I feel like I’m using theranos
@maiab wrote on 3 November, about agentic shopping, of having “NEVER” heard “so much hype and discussion” about a tech or product experience that “basically does not work at all” for the poster, adding: “I feel like I’m using theranos”. It is the post quoted above.
What Was Launched in the Same Nine Days
On 26 October Instinct posted that “Link balance is live in Instinct, built with @stripe and @link”, letting users set a per-purchase limit. The Grok Bot changelog for version 0.78.1, dated 27 October, says “Buy now starts an order you approve” and lets users connect Shop Pay.
Shopify’s Fall ’26 Edition of 28 October says shoppers “can buy from any brand on or off Shopify through Claude Duet, Grok Bot, and Instinct”. Instinct then expanded its 1Password integration and, on 30 October, announced a partnership with OpenTable.
On 31 October Agentcard launched opencommerce, which its site describes as “One integration to buy from most online stores”, listing 5,026 stores and a 1% fee on each completed order. On 3 November Juozas Kaziukėnas, founder of Marketplace Pulse, wrote on X that “Meta’s Muse is getting shopping carts powered by connectors” and that “Instacart is going to be first”.
Moore’s Reasons
Moore listed what holds usage back. She wrote that “99% of consumers will churn off an agent if there are any issues or delays”, and that “saving down credentials, handling payments, etc. all have too much friction and uncertainty now”.
She also wrote that “The VAST majority of consumers still are not using agents”, and that a person picked off the street in any city other than San Francisco has “likely heard of neither Muse nor Instinct”. She cited ChatGPT “now at 1.2B WAUs”, a figure we did not check, and said “early data shows that AI-driven traffic is actually higher conversion and intent than non-AI commerce traffic”, without naming the data.
As we read the nine days, the launches added ways for agents to pay, sign in, book and fill carts, while the usage statements counted activations, interactions and a few hundred orders.
As we read Amazon’s and DoorDash’s comments, both point shoppers to the retailer’s own app or agent: Amazon reports 400 million+ customers for Alexa for Shopping, and DoorDash said the predominant use case is through its app.
Michael Miebach, Mastercard chief executive, third-quarter call
“It is still early days, and the numbers are not yet significant in the grand scheme.”
Three Implications
FOR AGENT MAKERS Moore wrote that 99% of consumers will churn off an agent if there are any issues or delays, and named credential and payment friction. Several of the launches in this stretch target payment, sign-in and booking.
FOR RETAILERSAmazon reports 400 million+ customers in 12 months for its own agent, Alexa for Shopping, and DoorDash said the predominant use case is through its own app.
FOR PAYMENT NETWORKS Mastercard’s chief executive said agentic transactions are ramping up at impressive weekly growth rates and gave no rate in the transcript we read. Agentcard charges 1% of each completed order.
The framework puts it this way: “The stage of adoption looks completely different depending on which clock you read.”
As we read it, Muse’s activations and DoorDash’s count of agent orders are two such clocks. The framework does not discuss Muse, DoorDash or shopping agents.
The 10 million Muse activations and DoorDash’s several hundred agent orders come from different companies and measure different things. One is not a share of the other, and DoorDash did not say how many of its agent orders came through Muse.
Xu’s count covers the time the agents have been running, not a single quarter, so the 8-second comparison uses DoorDash’s quarterly average only as a yardstick. Su’s line comes from our machine transcript of the webcast, not an official transcript. Moore’s three-to-six-month view is her own, and we make no forecast.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
In the nine days to 3 November, Instinct, xAI, Shopify and Agentcard each added a way for AI agents to buy, and Marketplace Pulse’s founder reported that Muse is getting shopping carts. In the same stretch DoorDash said outside agents had sent several hundred orders in total since they began running, about 8 seconds of its average third-quarter order flow by our arithmetic, and Mastercard’s chief executive called the numbers “not yet significant in the grand scheme”.
A note on sourcing. Quotes come from the companies’ releases, pages, changelogs and posts, the Motley Fool transcripts of the DoorDash and Mastercard calls, TheFly’s report of Amazon’s call, and our machine transcript of AMD’s call webcast, read between 29 October and 4 November 2026. Moore’s and @maiab’s posts are their own views. The 8-second comparison and day counts are our arithmetic.
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.
