At Connect, Mark Zuckerberg also described new Mac control for the agent, extending a product that already works across files and native apps.
By [Ryan Merket](https://runtimewire.com/author/ryan-merket)
· Published
Primary source: [ChatGPT shared conversation](https://chatgpt.com/share/6ab4611a-fd2c-83ea-82a4-c5578cd5bbe6?ogimg=plain)
Why it matters #
Meta is tying Muse's economics to transactions the agent helps complete, a model that depends on retailer access and user trust as the agent gains permission to act across Mac apps.
At Meta Connect on September 23rd, Mark Zuckerberg (@finkd) said Meta expects Muse to earn money by taking a small fee from transactions its personal AI agent helps complete. The plan makes shopping and other paid actions part of the product's business model, rather than treating Muse solely as a chatbot or a subscription. In the same keynote excerpt, Zuckerberg said Muse was adding the ability, with user permission, to operate any app on a Mac. The remarks appear in a shared transcript of the keynote; the excerpt does not specify the fee, which transactions would qualify, or when the computer-use feature will be available.
Zuckerberg framed the pricing idea around a claim that Muse can save users money. The transcript describes Meta offering the agent free for a large number of tokens and expecting to profit over time from a small share of transactions. That is a different commercial bet from charging only for access or usage: Meta would need Muse to help users complete purchases or other paid tasks, then collect a fee when those actions generate revenue. The transcript gives no take rate, payment mechanics, or estimate of how much activity would be needed to make that model work.
The Mac announcement extends a feature set that was already in users' hands. Meta introduced Muse on September 8th, describing it as an agent running on a dedicated cloud virtual machine and able to work across connected services. On September 17th, Alexandr Wang (@alexandr_wang), Meta's chief AI officer, announced Muse for Mac. TechCrunch reported that the Mac app could interact with files, messages, calendars, notes and mail in their native applications. The new claim in the Connect excerpt is broader: Muse can drive any Mac app, with permission. The excerpt supplies no technical explanation of how that differs from the existing integrations or what controls apply app by app.
That gap matters because local computer access changes what users are being asked to trust. Meta's launch description says Muse runs in its own secure virtual machine, with users choosing connected services and approving sensitive actions. A Mac agent that can operate across any app reaches beyond that cloud environment and the connectors Meta described at launch. The keynote excerpt promises permission controls but does not spell out whether the feature acts on the user's desktop, uses a separate environment, or what the audit and approval process looks like.
Meta's transaction-fee thesis also runs into a practical constraint: an agent can only shop where retailers let it. On September 21st, Axios reported that Amazon had blocked Muse from browsing and buying on its marketplace. Amazon argued third-party services acting on customers' behalf should respect a retailer's choice to participate. That dispute puts pressure on the idea that a general-purpose agent can reliably turn user intent into purchases across the web. A fee model depends on access to merchants and checkout flows that Meta does not control.
Zuckerberg has cast Muse as a step toward personal superintelligence, and the Connect excerpt gives that ambition a commercial form: let the agent do more work, make it useful enough to influence real decisions, and take a small share when those decisions become transactions. The Mac feature extends the agent's reach; the proposed fee makes its success depend on being present at the point of purchase. Meta has not, in the excerpt, explained how it will calculate or collect that fee, or how retailers will respond to an intermediary taking part in their checkout relationship. Those are central operating details, not fine print, for the model Zuckerberg described.