Meta has launched Muse, a personal AI agent that can browse the web, fill forms, book appointments, connect to email and calendars, send messages and complete purchases. The product moves Meta from answering questions into delegated action, where the scarce input is no longer only model capability. It is permission to use credentials, personal context and money.
Meta has put a visible price on the higher-usage version of that proposition. Muse is available free with limits, while the Power tier costs $20 a month and Maximum costs $100. Meta told TechCrunch it expects most people to remain on the free tier. Paid conversion is only one part of the commercial case. The larger prize is action frequency: how often users allow Muse to move from research into booking, messaging and spending.
Trust is the constraint Meta has to move. A Visa-Harris Poll found that 72% of US consumers had used an AI assistant, while 23% trusted generative AI to handle payment transactions on their behalf; naming Visa as the payments brand lifted trust to 61%. For Meta, that 38-point uplift measures how payment-brand reassurance can widen the pool of users willing to delegate a transaction.
Muse turns trust into a permissions ladder
Meta's architecture acknowledges that the risk changes when an agent can act. Muse runs in a dedicated persistent virtual machine with its own browser. Connected credentials sit in a separate store, keeping their contents outside the model's readable context. Users can manage permissions, review an audit trail and approve sensitive actions such as purchases and emails. At checkout, Meta says Muse can use one-time card details through Stripe Link rather than exposing the underlying card credential to the agent.
Those controls are the product's commercial architecture. Requiring confirmation for every trivial action would erase much of the time saved by automation; allowing high-consequence actions without review would raise error and liability costs. The product has to expand authority selectively as users gain confidence in what it does.
The paid tiers make the first revenue path straightforward, although their economics remain largely private. Subscriber targets, expected conversion, merchant fees, payment revenue and the value of commerce actions completed through Muse remain private. If most users stay free as Meta expects, the $100 tier can still work as a high-value productivity product for a smaller population with frequent, expensive workflows. Mass-market economics would require a much broader pattern of delegated action.
Interest in assistance is already high. The harder conversion is from asking software for advice to allowing it to send, book and spend. Visa's result suggests trust can be layered: users may accept an autonomous workflow when a familiar institution remains responsible for the payment rail.
Muse gives Meta a credible product design for testing that transition. Approvals, isolated credentials and an audit trail let users increase authority in increments instead of making one all-or-nothing choice. The commercial hinge is the Visa survey's trust gap. Muse succeeds as a platform if its controls can produce a similar expansion in real delegated behaviour, turning free assistance into paid workflow and commerce.