Speaking to analysts on Oct. 21, 2026, Mark Zuckerberg said Meta expects to “build a model where we get a percent of the value that we are helping to create for people” with its Muse agent. That single sentence is a choice every founder building an agent faces: charge the user a subscription, or take a cut of what the agent makes happen. This guide uses Meta’s AI agent business model as the worked example, because it is trying both at once.
Where it stands: As of Andreessen Horowitz’s Oct. 5 consumer AI report, Muse had “described plans to earn a fee from transactions over time, though neither is doing this yet,” referring to Muse and a rival agent. Subscriptions are live. The fee side is still a plan.
What AI agent business model did Meta pick for Muse?
A hybrid with three layers, in Zuckerberg’s own description on the third-quarter call.
The base is free: “For most people, the product should be free, and we will build commerce tools so we get paid a fee based on the transactions that we help facilitate. We expect this will be paid for by businesses.”
The top layer is a subscription: “For people who want the most advanced models or to use extra tokens, they subscribe directly and pay.” Reuters reported in September that “the basic version is free, with subscriptions priced at $20 and $100 a month for heavier use.”
The money layer in between is the take rate, a percentage the platform keeps from each sale it helps close. Fortune summarized the shift this way: “Rather than just selling ads to businesses hoping to reach potential customers, Meta plans to charge businesses a fee when Muse helps them complete a sale.”
| Revenue line | Who pays | Status |
|---|---|---|
| Transaction fee | Businesses | Planned, not yet charged (a16z) |
| Subscriptions | Heavy users | Live: $20 and $100 a month |
| Free tier | Nobody directly | Default for most people |
When does a take rate beat a subscription?
A take rate works when the agent sits at the moment money changes hands. If your agent books flights, buys groceries or reserves tables, the value it creates is visible as a transaction, and a fee scales with that value rather than with how often the user opens the app.
A subscription works when the value is the usage itself: research, writing, coding, planning. There is no checkout to clip, so the user pays for access and capacity.
The two also behave differently on cost. A subscription caps revenue per user while compute rises with every request. A take rate ties revenue to outcomes, but a user who chats a lot and buys nothing pays nothing. Meta’s hybrid tries to cover both gaps: the fee catches shoppers, and the paid plans catch the heaviest non-buyers. For a broader look at seat-based plans under pressure from agents, see our piece on per-seat SaaS pricing after AI agents.
Why does Citi expect transactions to dominate Muse revenue?
Because Citi’s model leans on commerce. PYMNTS, relaying a Reuters report on Citi’s note, wrote that Muse “could reportedly generate revenues of more than $27 billion by 2030,” and that “Citi’s revenue estimates are based on Muse accounting for approximately $23 billion from transaction-related revenue and about $4.5 billion from subscriptions.”
That is an analyst estimate for 2030, not a disclosed number. It is still instructive for founders: Citi’s model assumes the fee layer, which is not yet switched on, will be much larger than the subscription layer.
There is a consumer-side reason too. PYMNTS cited Bank of America Institute data showing “just 3% of households paying for any AI service, at a median of $20 a month.” If few consumers pay directly, the business that captures the merchant side may matter more. Our sibling article on monetizing after the network forms looks at the timing question.
What plumbing does a transaction-fee agent need?
Payments and permission. Meta’s launch post said “when it comes time to pay, Muse can checkout with Link built by Stripe, and it is the first AI agent covered by Link’s purchase protections.” It added that “Shop Pay is coming soon as another way to pay.”
PYMNTS listed more partners in September: “Muse launched with Stripe’s Link wallet, and PayPal added its checkout across its merchant network,” and travel and dining partners plugged in early, with “Flight bookings run through Duffel, a travel API provider, and Ticketmaster and OpenTable plugged in at launch, Skift reported.”
Meta is also pushing a standard. Fortune reported that “Meta CFO Susan Li said the company is working with partners including Shopify and Walmart to develop a standard called the Personal Agent Protocol to define how consumers’ agents can securely interact with a business’s agents.”
For a startup, the lesson is that a fee model depends on integrations you do not control. Payment processors and merchant platforms decide whether your agent can check out at all.
What can go wrong with a take-rate model for agents?
Merchants can say no. a16z noted that “Amazon shut down access for Meta’s Muse in under two weeks, while Shopify, Instacart, OpenTable, Expedia, Ticketmaster, Plaid, and others have signed official integrations.” A fee model only earns where sellers let the agent in.
The plan can also take time to settle. At launch on Sept. 8, CNBC reported that Meta “hasn’t settled on any concrete plan” for taking a cut of shopping transactions, according to Wang. Zuckerberg laid out the fee model in detail on the Oct. 21 call, but as a16z noted, collection had not started by early October.
And the numbers have to close. Zuckerberg himself said on the call that “it is gonna be very important to kind of get the math on all of this to work out.” Li, for her part, said “we see multiple paths to monetization beginning with subscriptions” and that new AI businesses “can begin contributing incremental revenue in 2027.”
How should a founder choose between the two?
Start with where your agent creates value you can measure. If it completes purchases, a take rate aligns your revenue with the user’s outcome and lets merchants carry the cost, which is the bet Meta is making. If it produces work rather than sales, a subscription with a clear usage allowance is simpler and can be turned on from day one.
Many products will end up hybrid, like Muse. The order matters, though. Meta’s CFO said monetization would begin with subscriptions, and the fee depends on partners such as Stripe and Shopify. A startup without those partners in place can price access now and add a take rate once checkout integrations exist. Our guide to SaaS pricing strategies covers the subscription side in more depth.
Citi’s split, roughly $23 billion from commerce versus roughly $4.5 billion from paid plans in its 2030 estimate, is the clearest public sign of which layer analysts think will carry an agent like Muse.
Frequently Asked Questions
What is the best business model for an AI agent?
There is no single answer. Meta plans to keep Muse free for most people, charge businesses a fee on transactions the agent helps complete, and sell subscriptions to people who want extra tokens or advanced models. A take rate suits agents that close sales; subscriptions suit agents whose value is usage itself.
Does Meta charge a transaction fee on Muse purchases yet?
Not as of a16z’s Oct. 5, 2026 report, which said Muse planned to collect transaction fees eventually but had not started charging them. Zuckerberg described the fee model again on the Oct. 21 earnings call.
How much of Muse revenue does Citi expect to come from transactions?
Citi estimated, as reported by Reuters and PYMNTS, that Muse could generate more than $27 billion of revenue by 2030, with roughly $23 billion of that tied to transactions and roughly $4.5 billion to subscriptions. These are analyst estimates, not Meta figures.
Who pays the fee in an agentic commerce model?
In Meta’s stated plan, businesses do. Zuckerberg said on Oct. 21, 2026, that the commerce fee is expected to be paid for by businesses, while consumers who want more pay through a subscription.
Sources
- Introducing Muse: The World’s First Personal AI Agent Built for Everyone — Meta Newsroom, September 8, 2026
- Meta (META) Q3 2026 Earnings Call Transcript — The Motley Fool, October 21, 2026
- Meta pushes into personal AI agents in Muse Spark family — CNBC, September 8, 2026
- Mark Zuckerberg said demand for Meta’s Muse was so high, he had to get Stripe to raise credit limits — Fortune, October 22, 2026
- The Top 100 Gen AI Consumer Apps — 7th Edition — Andreessen Horowitz (Olivia Moore), October 5, 2026
- Meta’s Muse Could Become $27 Billion Business, Citi Says — PYMNTS, October 6, 2026
- Is Muse Really the Answer to Meta’s AI Woes? — PYMNTS, September 23, 2026
- Wall Street expects Meta’s AI agent to shape into a new revenue engine — Reuters, via The Star, September 22, 2026
This article is general information for founders and is not accounting, legal or investment advice. Figures are as reported by the sources cited on the dates shown; analyst and third-party estimates are not company disclosures. Found Founders is not affiliated with Meta.