MBI Deep Dives, a subscription newsletter on business and technology, published an analysis on October 1, 2026 contending that Meta's Muse personal agent may never carry advertising, because the agent's browsing of merchant websites can still inform ads that run on Facebook and Instagram.

In Short

On October 1, 2026, a newsletter called MBI Deep Dives argued that Meta's new Muse assistant, which can shop and run errands for people, may never show advertising inside the app, because Meta can still earn from ads elsewhere when Muse visits online shops. Shops, delivery apps and advertisers are affected, since DoorDash and Airbnb are building assistants of their own while Amazon has blocked Muse from its store. Nothing about Meta's rules has changed: the post is one analyst's view, and the fee Meta plans to charge merchants has no published size.

Who is behind the analysis

The post, titled Why Muse May Never Need Ads, appeared on October 1, 2026 under the newsletter's Meta category. It is a daily entry rather than one of the monthly company studies, and it closes with a disclosure that the author's personal portfolio may colour the write-up and that nothing in it amounts to a recommendation to buy or sell securities. The holdings table sits behind the subscriber paywall in the copy reviewed.

MBI stands for "Mostly Borrowed Ideas", a pseudonym first used on Twitter, according to the newsletter's About page. The writer is Abdullah Al Rezwan, who majored in finance at the University of Dhaka, spent three and a half years in sell-side equity research in Bangladesh, completed an MBA at Cornell University and then worked as a generalist on the US large-cap team at Madison Investments. He holds CFA and FRM charters. The site was started in September 2020, sends one email a day and publishes one detailed company study each month, according to the page.

The app layer starts to respond

According to MBI Deep Dives, application companies began answering Muse within weeks of its release. Amazon moved first. PPC Land's weekly edition for the week to September 27 recorded that Amazon blocked Muse from its store after Meta declined to remove Amazon from the agent's reach, and that Amazon complained the agent captured shoppers' credentials without identifying itself. The complaint tracks Amazon's own Agent Policy, which took effect on March 4, 2026 and requires AI agents to identify themselves as automated systems.

Two more responses, from DoorDash and Airbnb, arrived just before the post was published, the author writes.

DoorDash tests ordering by text

DoorDash released an AI agent that takes orders inside Apple's iMessage, according to the post. A customer texts DoorDash to have a usual order delivered to the office without opening the app. The agent matches the phone number to the customer's DoorDash profile, reads the order history and preferences, and completes checkout with the stored payment method inside the thread. The pilot is reportedly live with 20,000 US users on iOS.

The format has an evident limit: it suits customers who already know what they want. DoorDash chief marketing officer Tim Castree said half of customers arrive at the platform without knowing what to order, according to the post. MBI reads that as a glass half full, since the other half could place an order with a quick prompt to DoorDash's agent or a Muse-like agent, never opening the app. DoorDash co-founder Andy Fang had separately said that users of the company's assistants favour practical functions such as reordering usual items over the standard chat window.

The advertising side of the business assumes browsing. DoorDash reached 400,000 advertisers in June 2026 and added Spotlight, a homepage placement meant to capture purchase intent before an explicit search. A text thread has no homepage, and the post does not address where comparable inventory would sit.

Early usage data is thin and mixed. According to Implicator, whose figures the post reproduces, grocery orders built with Ask DoorDash, the in-app chatbot, had nearly 50% higher basket value than the same consumers' traditional orders from August through early September 2026. Nearly half of the chatbot's restaurant orders from June through August went to places the customer had never tried. Those are DoorDash's own usage figures, and the company has released no comparable numbers for the Messages pilot. A Citizens analyst note summarising Castree's remarks said people who clicked through to DoorDash from AI chatbots converted at higher rates and filled bigger baskets than standard traffic, while orders from early agent traffic carried baskets two-thirds smaller.

DoorDash is also talking to Meta. Fang said the company is in talks about an official link with Muse, according to the post, which notes that Tony Xu sits on Meta's board. MBI's reading is that DoorDash is running both sides of the trade, building its own agent where users already spend time while staying open to acting as a connector inside someone else's. Google had taken a similar route, naming DoorDash among its food delivery partners when it extended the Universal Commerce Protocol beyond retail.

Airbnb treats agents as a channel

Airbnb took a more defiant posture, the post says. A toggle switches on AI search, in which users describe what they want by text or voice and the interface generates dynamic filters from the query. Brian Chesky has deliberately steered away from a chatbot-style design. At Skift Global Forum the week before, he said "Next year we will launch our agent", an interface he described as much richer than a chatbot. In an X post last edited at 8:06 PM on September 30, 2026, Chesky called the new search a "baby step" towards agentic AI.

Meanwhile Airbnb treats ChatGPT and Muse as distribution channels that send it traffic, according to MBI. Chesky also called AI an existential risk in the same Skift conversation. The distribution role is not a large one yet: Comscore data covered by PPC Land showed Airbnb and Vrbo each appearing in 3% of lodging-related AI responses between December 2025 and May 2026, against 14% for Marriott.

What Meta has said about money

Axios reported at launch that Muse carries no advertising, with Alexandr Wang saying Meta is exploring commerce opportunities instead, according to the post. At Connect, the post adds, Zuckerberg reportedly did not say the word advertising once in his keynote. He did say that Muse would stay free for a large number of tokens and that over time "we will profit by taking a small fee from transactions". The fee would reportedly come from merchants rather than users. Its size and mechanics are undisclosed.

Meta's published commitments on data are narrower than a ban on advertising. Meta said on September 8, 2026 that Muse conversations and the data inside each person's virtual machine stay out of its ad systems. Meta AI works under a different rule, since conversations with it have fed content and ad personalisation from December 16, 2025. Zuckerberg's August letter on AI ran to 6,542 words without mentioning advertising. On September 28, Meta named the Muse agent among the first four products of its new enterprise division, with no pricing or dates.

A competing reading exists. Digiday's analysis of September 25 argued that advertising inside Muse looks likely, citing 730,000 US downloads in five days, revenue so far limited to $20 and $100 monthly tiers plus merchant fees, and 2026 AI capital expenditure of $135 billion to $145 billion that had absorbed 98% of operating cash flow by the second quarter. Against that sits the size of the existing business: Meta reported $59.36 billion in advertising revenue for the second quarter of 2026. Most investors MBI speaks to believe ads in Muse are almost inevitable. The author is not sure.

The argument: trust first, fees later

Trust as the product

The thesis rests on one claim: if trust is the decisive advantage over vertical-specific agents, advertising in Muse may never arrive. A person who hands over emails, logins and payment credentials to an agent must believe it works for them and not for the highest bidder, the post argues. "No ads" is, in its words, "a very strong counter-position", not least against Meta's own history.

The competitive logic runs further. If Meta sets the norm that a trustworthy personal agent carries no ads, rivals that need ads to fund their agent start to look compromised, and few have an ad engine elsewhere to subsidise a generous free tier. Meta can afford to wait. If Muse becomes the aggregator of aggregators within a couple of years, Meta would be positioned to ask for a take rate on much of the economic value flowing through it.

Trust is not a given. The post concedes a widespread lack of trust in Meta, and PPC Land's weekly edition noted an AdExchanger report that some of Muse's phone calls were placed by humans in a call centre, a detail that sits oddly with the description of the product as autonomous.

Objections the post addresses

The first objection is that a merchant-paid fee is itself an incentive, so Muse is not neutral. MBI does not find this troubling: if every merchant pays whenever a transaction passes through Muse, the agent should be largely indifferent to which merchant wins, much like a card network.

The second is access, and the post treats it as the strongest card the app layer holds. An agent can shop only where it is allowed to. The post says Meta has since added Walmart, Best Buy, Gap, Sephora and Wayfair among others, on top of access to Shopify's catalog; it cites no source for the list. The bet, as the author frames it, is that enough aggregated demand will make holdouts feel the loss.

Amazon's position is asymmetric in time. Letting agents in threatens its high-margin ad business at once, whereas staying out costs volume only gradually, and holding out is probably rational at present. Logistics, however, is a fixed-asset business in which utilisation matters, so the calculus in week 100 may differ from week one. Amazon is also likely an anomaly, the post argues, because few companies dominate their vertical as it does. Amazon's own conduct is not one-directional. PPC Land reported a job listing for an agentic commerce team meant to connect Amazon with external agent platforms, and Amazon Ads announced a ChatGPT advertising partnership on September 10, 2026, with Delta Vacations among the first testers.

The indirect route to ad revenue

The post's most distinctive point concerns what "no ads" would cost. It quotes Meta's blog post "How We Built Safety Into Muse", which says Muse does not share conversations or virtual-machine data with Meta's ad systems. The same post acknowledges that Muse's browsing appears as the person's own activity. If a person asks Muse to buy a shirt from a designer's website, the designer might use that visit to show an ad on Instagram, and a restaurant reservation or a Facebook Marketplace find may also influence the ads a person sees, according to the excerpt the post reproduces.

From that, MBI builds a loop with two requirements: a site tag on the merchant's website that registers the visit, and a large owned surface elsewhere on which to serve the resulting ad. Startups have neither. ChatGPT has an ads business, but its only inventory is ChatGPT itself and its targeting draws on the conversation, the trade-off Meta is positioning against. Google could replicate the loop through Search, YouTube, Gmail and its display network, yet an agent that completes a purchase likely replaces the commercial query Google monetises best. Meta has always had to infer intent rather than observe it, the post notes, so every Muse errand can be incremental signal.

The conclusion follows: Meta can keep ads out of the Muse app and still be paid on Facebook and Instagram, which makes delaying direct monetisation far cheaper for Meta than for almost any rival. The author describes the indirect monetisation as already under way, with fees from transactions able to wait until enough demand has been gathered. As he puts it, Meta has a real shot at "aggregating the aggregators".

The caveat is stated plainly. As of the post's date, the author finds apps a better experience than asking an agent, and the near-term threat to vertical companies is not overstated. The open question is whether consumers want one agent per app or one agent for nearly everything.

Limits of the evidence

Several layers separate the reader from the underlying data. The DoorDash basket figures are DoorDash's own, relayed by Implicator and reproduced by MBI. The Citizens note reached the post through a summary of Castree's remarks. The 20,000-user pilot is described as reported. The additions to Muse's merchant roster carry no cited source, and the size of Meta's merchant fee is unknown.

The thesis also turns on a mechanism the post does not quantify. How often a Muse errand produces a retargeting event on a merchant's tag, and how many such events become Instagram ads, appears in neither the post nor the Meta material it excerpts. There is a technical gap as well. Code analysis by Juozas Kaziukenas, which PPC Land covered, found that Muse has three checkout modes: a browser mode that types card details into whatever form a merchant presents, a Shopify mode that runs over APIs and, in his assessment, most likely over the Universal Commerce Protocol, and a Stripe Link mode. A browser-side tag fires when a page loads in a browser. Whether API-based checkout would register on a merchant's tag at all is not addressed.

Finally, the post is an opinion from a single writer who discloses a personal portfolio, in a newsletter whose complete text is paywalled. Digiday's contrary reading rests on cash-flow arithmetic. Neither view is confirmed by anything Meta has published about advertising in Muse.

Why this matters for the marketing community

For advertisers, the first consequence is about inventory. If the thesis holds, Muse adds no new place to buy ads, and attention aimed at the agent's audience stays inside existing Meta surfaces. That contrasts with ChatGPT, where OpenAI said advertising passed a $1 billion annualised run rate on August 31, 2026, a number Digiday derived from roughly $83 million of monthly revenue that OpenAI has not itself published. Comscore's panel put ChatGPT at 50% of US AI prompt volume in June 2026, down from 70% in January, with Gemini at 30% and Claude at 11%. Whether an advertising-free agent and an advertising-funded one can both hold large audiences is a question the next quarters of data will begin to answer.

For measurement teams, the issue is labelling. Browser-mode agent traffic is, in PPC Land's earlier phrasing, a buyer merchants cannot tell apart from a bot at the checkout form. Meta's own description of agent browsing as the person's activity leaves little room for a separate flag in a merchant's analytics or retargeting pools, while Amazon's policy demands the opposite, that agents identify themselves. The two positions are not reconciled in any material reviewed here.

For delivery and travel brands, the post sets a timetable. Airbnb says its agent arrives next year, and DoorDash is building a text channel while negotiating a Muse link. Sales of homepage placements, built for people who browse, meet customers who prompt. How the economics move from one to the other has not been stated by either company.

What Meta will do is the open point. Its public record shows a data commitment, a plan for merchant fees and silence on advertising in Muse. A newsletter's argument fills that silence, and so far nothing else does.

Timeline

Summary

Who: MBI Deep Dives, a newsletter written by Abdullah Al Rezwan under the pseudonym "Mostly Borrowed Ideas", analysing Meta's Muse agent and the responses of DoorDash, Airbnb and Amazon.

What: An analysis arguing that Muse may never carry advertising because trust is a competitive advantage, merchant-paid transaction fees can follow later, and the agent's browsing of merchant sites can already inform ads on Facebook and Instagram.

When: October 1, 2026, with DoorDash and Airbnb developments dated to the preceding day and Amazon's block of Muse recorded in the week to September 27, 2026.

Where: Published on the MBI Deep Dives site, partly behind a paywall. Muse is available in the United States, and the DoorDash pilot is reportedly limited to US iOS users.

Why: Application companies are building their own agents while Meta decides how a general-purpose agent will earn money. The answer shapes where advertising inventory exists, how agent visits register on merchant tags, and whether vertical agents can claim neutrality.