Meta’s Muse, its general-purpose AI agent for consumers, may never carry an advertisement, and MBI Deep Dives argues in an analysis published on 1 October that this is a shrewd strategy rather than a sacrifice. The thesis: Meta can leave ads out of the agent itself and still collect advertising revenue from what the agent does elsewhere.
The argument starts with a fight over who owns the shopping conversation. Amazon has already blocked Muse, and two app companies answered with agents of their own. MBI Deep Dives reports that DoorDash launched an agent inside Apple’s iMessage, reportedly live for 20,000 US iPhone users, and that Airbnb rolled out its first AI search. Airbnb’s Brian Chesky said at the Skift Global Forum, “Next year we will launch our agent.”
The early DoorDash numbers are mixed, and they come from the company’s side. MBI Deep Dives cites reporting from Implicator: grocery orders placed through Ask DoorDash, the in-app chatbot, carried baskets nearly 50 percent larger than the same customers’ usual orders from August to early September. DoorDash has released no equivalent figures for the iMessage pilot. A Citizens analyst note on comments from DoorDash chief marketing officer Tim Castree described early agent traffic as producing baskets about two-thirds smaller.
The author’s main objection to store-built agents is that they cannot play fair. The author expects a DoorDash agent to stay silent when Uber Eats sells an identical burrito bowl for less, and an Airbnb agent to skip Vrbo entirely. The author says he gave Muse a standing instruction to look for a cheaper way to book anything, and it complied without being asked twice. That is one writer’s test, not an independent evaluation.
On money, Axios reported that Muse launched without advertising, and Alexandr Wang said Meta was exploring commerce instead. At Meta’s Connect event, Mark Zuckerberg reportedly never said the word “advertising” in his keynote. He did say Muse will stay free up to a generous volume of tokens, and that “we will profit by taking a small fee from transactions.” The fee would reportedly fall on merchants, but its size and mechanics are undisclosed. Whether that fee shades which merchants Muse favors is the open question, though the author compares it to a card network that is indifferent to who sells the goods.
The sharper claim concerns advertising that never appears in the agent. Meta’s own post, “How We Built Safety Into Muse,” says conversations and data inside the agent’s virtual machine are not shared with Meta’s ad systems. It then concedes that Muse’s browsing looks like the user’s own activity. If Muse buys a shirt from a designer’s site, that designer could show the user an Instagram ad afterward. A restaurant booking or a Facebook Marketplace search could shape ads the same way.
Read that way, every Muse errand hands Meta a signal of real intent, which the company has historically had to guess. The author sees few rivals who can copy the loop. It needs a tag on the merchant’s site and a large owned surface for the resulting ad. Startups have neither. ChatGPT sells ads, but only inside ChatGPT, using the conversation itself, which is the very trade-off Meta is positioning against. Google has the reach, yet once an agent closes the sale itself, the shopping search that Google earns the most from never gets typed. These are the author’s inferences. Meta has not said it designed Muse around this loop.
Access is the weak point. Stores that refuse an agent entry are closed to it, and Amazon’s block is the biggest piece of leverage app companies still hold. Meta has added Walmart, Best Buy, Gap, Sephora, and Wayfair, plus Shopify’s catalog. The author reasons that Amazon’s choice is rational for now, since admitting agents threatens its lucrative ad business immediately while staying out bleeds volume slowly. He adds that few companies command Amazon’s position, so most holdouts would lose demand to rivals who cooperate.
The author is candid about limits. Most investors he speaks with expect ads in Muse eventually, Meta carries a trust deficit, and he still finds the apps better than asking an agent today.
For DoorDash and Airbnb, the test is whether shoppers will trust an agent that browses only its owner’s shelves. For every other agent maker, a Meta pitch of “no ads” would force a choice between a subscription price and a trust penalty.
Reported by MBI Deep Dives on 1 October 2026.
