---
title: "Meta's Muse - Has the Demand Side Arrived to Agentic Commerce?"
url: "https://lumosbusiness.com/notebook/metas-muse-has-the-demand-side-arrived-to-agentic-commerce/"
date_published: "2026-09-26"
format: "essay"
author: "Joel Finningley"
thesis: "Muse is the first demand-side catalyst for agentic commerce at mass-market speed — and whatever happens to Muse itself, the winners are the merchants with legible storefronts and structured catalogs."
---

# Meta's Muse - Has the Demand Side Arrived to Agentic Commerce?

**Thesis.** Muse is the first demand-side catalyst for agentic commerce at mass-market speed — and whatever happens to Muse itself, the winners are the merchants with legible storefronts and structured catalogs.

---

In a post at the end of August, I looked at [Whitespaces & Opportunities in the Agentic Economy](/notebook/agentic-websites-whitespace-and-opportunities/). At that time, there were no clear demand-side catalysts for the consumer side of agentic commerce specifically. Small-scale experiments had been conducted in OpenAI (Instant Checkout), but nothing had stuck yet when it came to consumers and agents.

Meta changed everything on the demand side of the equation on September 8th when they dropped Muse. What is Muse?

A personal agent on a dedicated cloud - one VM (virtual machine) per user. The agent browses on behalf of the user, books them in services or buys them products, and presents the transaction to them for their final approval. It's free for users up to 100M tokens per week, with $20 and $100 tiers for more usage. Checkout is via Stripe Link, Shop Pay, or PayPal; the merchant stays the merchant of record.

| Field | Detail |
|---|---|
| Product | Muse — a personal AI agent from Meta Platforms |
| Launched | September 8, 2026; commerce partnerships and business model announced at Meta Connect, September 23, 2026 |
| Availability | US only, ages 18+; iOS, Android, web (muse.ai), and inside WhatsApp; Mac computer control from September 17 |
| What it does | Browses, books, and buys on the user's behalf; keeps working after the app closes; asks for approval before purchases or sending messages |
| Architecture | One dedicated cloud virtual machine per user; a separate "Sentinel" agent approves network access (company statement) |
| Model | Muse Spark 1.3, from Meta Superintelligence Labs; closed weights |
| Price | Free up to 100M tokens per week; paid tiers at $20 and $100 per month |
| Payments | Stripe Link, Shop Pay, PayPal; the user approves each total in chat; the merchant remains merchant of record |
| Commerce partners | Shopify, Walmart, Best Buy, Gap, Sephora, Wayfair, American Eagle, GameStop, Instacart, Expedia |
| Blocked by | Amazon (from September 20, 2026) |
| Business model | A future merchant-paid transaction fee; rate undisclosed; no fee on Shopify today; analysts expect it from 2027 |
| Estimated cost to Meta | ~$0.25 per task and ~$11 per user per month at launch (Cantor Fitzgerald estimate) |
| Early usage | ~557,000–642,000 daily users in the first two weeks (Sensor Tower, Apptopia estimates; Meta has disclosed none) |

*As of September 2026. Company statements, analyst estimates, and third-party data are labelled as such; figures will change.*

We commonly associate Meta with social platforms such as Facebook, Instagram, and WhatsApp; the nearly $2 Trillion company makes their money on ads. Meta has dabbled in commerce before with Marketplace, and the [live shopping](/notebook/the-hsn-reboot-a-new-live-shopping-golden-era/) on Instagram it shut down in 2023; Muse puts them in commerce's transaction layer for the first time.

## The Muse Business Model - A Take Rate with No Rate

Mark Zuckerberg has said that the company will make money off transaction fees (typically called the 'take rate') on the merchant side of the transaction, but there are currently no fees being charged. Wall Street Analysts aren't expecting the company to implement the fee model until 2027.

What makes this fascinating in some ways - given the way agents are restructuring the way the internet is navigated - is that Meta is potentially hedging against agents eating their ads model. In other words, as agents and agentic traffic scales up across the web - acting on behalf of their users - human attention on platforms gets cut. An agent that picks the merchant bypasses the ad auction that sells the merchant the customer's attention.

Businesses like Amazon, whose eCommerce platform monetizes both transactions and ads, have taken the approach of essentially banning third-party agents from crawling and transacting on the platform. And in the case with Meta's Muse agent, Amazon has already responded by banning it from their platform.

Conversely, Shopify is all-in agents. They have previously enabled the agents.md web-root standard for agent crawlability on all their merchant's stores, making it easy for agentic commerce. Muse's integration runs through Shopify Catalog, Shop Pay, and checkout via UCP; Shopify is an official partner that gives Muse a large distribution partner for sophisticated commerce brands of all shapes and sizes.

## The Bulls, The Skeptics, and The Ripple Effects

As this is a new innovation in the market, many different views exist as to how this will play out in the market:

-** The Bulls -** Deutsche Bank's bullish view on Muse relative to Shopify; since Shopify is really good at organizing a storefront's product catalog for agents to parse and index, Muse comes in and executes the transaction. In this case, Shopify is the rails and shares the transaction economics with Meta.
- **The Skeptics -** A slew of others, including prominent university professor [Scott Galloway](https://finance.yahoo.com/technology/ai/articles/no-one-violates-privacy-meta-094515687.html), have stated bluntly that Meta has a major trust problem, presenting a major roadblock to broad-scale consumer adoption. An Oppenheimer Study found that only 8% of users would trust Meta with their password, compared to 16% for ChatGPT and 30% for Google.
- **The Ripple Effects -** Many companies in travel (Expedia), hospitality (Airbnb), and other related consumer sectors immediately saw their stock prices decrease as the Muse hype ramped up. The TLDR is that if a consumer agent like Muse becomes popular for everyday transactions, it routes traffic away from some of the traditional traffic powerhouses for these various consumer segments. Muse is looking for the best deal for the consumer, so it has the ability to search out information at far greater capacity and speed than the typical search process.

Despite only being a couple of weeks old, there are anecdotal reports across the internet about how people are using Muse to save or make money; after all, Zuckerberg has stated he believes 'Muse will make you money:'

- Crawling through an email account for unused subscriptions and cancelling them
- Connecting it to Meta Ads and optimizing campaigns
- Finding the best deal on a trip or for event tickets

The hard part of the equation to speculate on right now is that most people have no idea what they would do with a 'personal agent.' So while many are downloading the app (Muse is #1 on the U.S. app store) and playing around with what Muse can do for them, the big tbd is going to be whether it really sticks for consumers as a whole.

## How Much Does it Cost to Run a Personal Agent?

It certainly appears that Muse is the mainstream version of [OpenClaw](https://openclaw.ai/), which took the AI side of the internet by storm earlier this year as a technical blueprint for running a personal agent, really the first of its kind. The downside of it was that it was seen by many as technically complex and cumbersome to configure to personal preferences.

[Hermes](https://hermes-agent.nousresearch.com/) has since come out and built a more efficient, easier-to-configure architecture for users to build personal agents. But despite the relative ease compared to OpenClaw, the downside in many ways is still the cost of running an agent.

Now comes Muse, which operates on a dedicated cloud per user and offers 100M tokens a week for free to every user. This is a significant amount of compute per user, in addition to the inference to actually run the agent:

- **Tokens (Inference) - $0.125 per task x 44 tasks per month = $5.50 per month -** [Cantor Fitzgerald](https://www.benzinga.com/analyst-stock-ratings/price-target/26/09/61954918/meta-platforms-4-reasons-why-this-analyst-is-bullish-on-muse) estimates that Meta incurs a cost of $0.25 per task for the agent at launch - as agentic tasks consume 10 - 100X more compute than basic chat. This is in reference to what would typically be the API tokens that a user consumes to run an agent. But since Meta has its own chips, data centers, and engineering team to optimize the agent's performance, it is safe to assume the cost is less than that, maybe 50% below that estimate ($0.125 per task)
- **Compute (Per-User Machine) - $3 - $5 of VM per month - **[Stark Insider](https://www.starkinsider.com/2026/09/meta-muse-specs-what-it-runs-on.html) estimates that each user gets a small VM: 2 vCPUs, ~7.7GB of RAM, no GPU, and browsers leased from a shared pool. For always-on rented machines, that would cost about $30 - $60 per user per month, but since the agent is not always active, the cost can be estimated to around $3 -5 per month

Cantor Fitzgerald's estimate is that Meta's launch cost is $11 per user per month for Muse, while these estimates come in at somewhere between $8.50 and $10.50 per user per month, so we can assume the range is somewhere around those numbers.

## Can Personal Agents Become a Commerce Revenue Machine?

Imagining a projected future where let's say 100 Million active users were on Muse, that would be $13B a year in costs at current rates. Cantor estimates that task cost will come down by 80% in the years ahead, but for argument's sake let's hold on to the $11 per user per month number for now:

- **Subscription Revenue:** the $20 and $100 tiers help drive revenue to offset costs, but they also ensure that power users pay for their share of compute and keep overall costs down
- **The Take Rate:** presumably there will be a whole calculation on Take Rate, whereby: intent volume x conversion x basket size x take rate will create the revenue flywheel for Muse's whole business model

*Intent Volume* is inherently hard to estimate and calculate, but its relevance is tied to the way Meta's ads model currently works; it has multiple layers of data about billions of people around the world, and it can measure a user's intent to serve an ad. Compared to Google, on the other hand, whose ads model is based around keywords. Muse could certainly have a large influence on web traffic, bypassing several layers of the traditional search process.

*Conversion* is where the rubber meets the road because intent to purchase by the agent only converts into the actual purchase if the user approves the transaction. Given that Muse is working from the user's personal email, with access to calendar and other attributes, it will be interesting to watch if conversion is high for Muse overall compared to other traffic sources. If conversion is high, then merchants will be optimizing as much about their whole web experience as they can.

*Basket Size* is the great unknown at this point. Will people trust Muse on higher ticket items or stick to small stuff? If early, anecdotal indications are anything to go off of, the nature of the ask is about saving money or finding the best deal, so Muse could process higher-ticket items. The savings are largest for the Muse consumer on big-ticket items.

*Take Rate* is likely a byproduct by the next two or three quarters of data relative to all of the above. They want to ideally keep the Take Rate as low as possible for their strategy, which would require large volumes of Muse-based searches, high conversion rates, and a strong breadth of basket sizes that includes higher-ticket items. This would let Meta charge smaller transaction fees to merchants and muscle into the commerce arena.

## The Thesis - Legible Merchants are the Winners

Whether or not Muse takes off and becomes the big winner as the de-facto personal agent for consumers is secondary to the fact of what this moment signals overall for commerce on the web; Muse is a big signal that the demand side of agentic commerce is here. Shopify has been positioning for the supply side by integrating agentic commerce into the core of merchant's storefronts, making them easy to navigate and transact on. The big winner is actually the merchants who have a legible storefront and the structured product catalog to support transactions.

Playing all of the above forward, what should the average Shopify merchant do today?

- Turn Shopify's Meta channel on (free)
- Fix and optimize structured product data, as the agent's ability to search and execute on any transaction is bounded by it
- Fix attribution - set up server-side event tracking so Muse orders are measured, browser pixels don't see them
- The free era for Muse will only last to sometime in 2027, so now is the time to test the setup

For Muse to work for Meta, they need merchants to stay in. Amazon has blocked it outright, Shopify merchants can switch the Meta channel off with one toggle. Merchants have to adjust their storefront's setup, including attribution and pixels for measurement, but beyond that the no-fee grace period offers a window into the mechanics of agentic commerce. ChatGPT had proposed a 4% fee for such transactions, as an example, and that fee alone could change a lot of the margin economics for the average store; with an open period of testing, a merchant can determine whether the juice is worth the squeeze.

If a Shopify merchant is heavily leveraged to Meta now for Ads, Muse could change that entire strategy in a matter of months; not because the Ads on Meta will be any more or less effective, but because the average consumer may start spending less time on those social platforms and more time on Muse.

There are many what-ifs and ripple effects that will likely alter the agentic commerce landscape now that Muse has fired the first shot. Anthropic recently launched [Claude Commerce Agents](/notebook/claude-commerce-agents-measured-cpct/) for merchants, will they now attempt to launch something on the consumer side? What about OpenAI, will they jump back into the ring? Could another big tech company like Apple come out of left field and attempt to make a splash?

All are unknown at this point, but competition is heating up and for now, Muse is the buyer. The aggregate activity favors the merchants who make their stores' agents readable, and gives consumers more intelligent and actionable commerce insight than ever before.


---

Written 100% by hand by [Joel](/about/) — no AI in the prose. Research and contextual support from Claude; shipped on Claude Code.