Morning Edition · Thursday, September 10, 2026Published at 2:24 AM EDT · New York
The agent runs on Meta's Muse Spark model family, keeps working after a user closes the app, and is priced with a free tier plus monthly plans at 20 and 100 dollars.
Meta has moved its consumer artificial intelligence product from answering questions to executing transactions. Muse, released to United States users this week, reads a user's email, books travel, fills in forms, negotiates bills, makes purchases and continues running after the app is closed. Mark Zuckerberg, Meta's chief executive, described his own use as a persistent worker rather than a chatbot, including watching for mountain climbing permits to become available and securing them.
The commercial structure matters as much as the capability. Meta is selling Muse on a free tier with monthly plans at 20 and 100 dollars, which puts a subscription line item on top of an advertising business. Zuckerberg has said Meta will add encryption modelled on WhatsApp so that Meta itself cannot read certain data the agent stores, a claim no external party has yet audited.
Muse runs on Muse Spark, the model family that Meta Superintelligence Labs began shipping this year and opened to developers through the Meta Model API. Muse Spark 1.1 has a context window of one million tokens and actively compacts it, and Meta made the endpoint compatible with the OpenAI software development kit, so existing code can be redirected to it with an API key.
What no one has published is an independent success rate. Booking a flight, disputing a bill and completing a checkout are long-horizon tasks where a 90 percent per-step success rate compounds into frequent failure, and Meta has released no third-party evaluation of end-to-end completion. The near-term evidence will come from refund and chargeback volumes, not from demonstrations.
Meta, which converts an advertising-funded assistant into a recurring subscription and positions itself at the point where a purchase begins, taking discovery traffic and transaction data from travel, ticketing and retail intermediaries.
Part of a tracked trend
Consumer Agents Gain Payment Authority
Personal AI agents increasingly hold payment credentials and complete purchases directly, moving competition from chat quality to control of the transaction, and forcing merchants, payment networks and regulators to build rules for machine-initiated commerce.
Start a discussion in Townsquare.
More from this edition
The launch, the United States availability and the free plus 20 and 100 dollar tiers check out, but two qualifiers are load-bearing: Muse requests approval before sensitive actions such as purchases or outgoing email, so it is not an unattended spender, and the WhatsApp-style encryption is a roadmap item rather than a shipped one, since Meta's own Superintelligence Labs leadership says engineers can still access data inside the Secure VM today while the key-custody feature has no confirmed date.
An open-source-intelligence read of how likely this story is true with its real nuance, not a judgment of any outlet. It assesses the claim, weighing independent and adversarial reporting. How we label confidence.
What this means
An agent with payment authority, operating at Meta's distribution scale, threatens the discovery layer that travel, ticketing and retail intermediaries monetize. If Muse becomes the place a purchase starts, booking platforms and comparison sites lose the traffic they resell to merchants, while card networks and Meta gain a transaction stream instead. The same authority is also the liability. When the agent executes a purchase incorrectly, or acts on an injected instruction, the resulting support and dispute costs fall on Meta rather than on the user.
What to watch
Observations to monitor, not financial advice.
Synthesized from: Polylog editors · Meta AI
Comments
0No comments yet.