Skip to content

Product11 publishers3 min readPublished Updated

Meta's Muse is free to use, but a payment card is needed once usage crosses into paid tiers

Meta can put a personal agent in front of more people than any startup can reach, but the permission walk between installing Muse and getting one errand finished still has to happen, and nobody has published that number.

The Product Desk · Product desk

Photograph accompanying Meta's Muse is free to use, but a payment card is needed once usage crosses into paid tiers
Photo: theverge.com

What happened

  • Meta introduced Muse on Tuesday, a personal AI agent for U.S. users that TechCrunch describes as the company's biggest bet on consumer AI so far.
  • Meta says the agent can send emails, book travel, lower bills, fill out forms, turn recipe reels into grocery lists and make purchases, using Link by Stripe for checkout.
  • Muse is reachable at muse.ai, through iOS and Android apps and inside WhatsApp chats, with access from Meta's AI glasses promised soon.
  • Alongside a free tier, Meta launched two paid plans, Power at $20 a month and Maximum at $100, while saying it expects most people to stay free.
  • Meta says Muse runs inside a dedicated secure virtual machine with its own browser, alongside a separate Sentinel agent kept apart from Muse at the system level.

Compiled by The Product DeskSomething wrong?How this is made

Why it matters

  • decision A team shipping a rival consumer agent can no longer compete on where the product appears, so the fight moves to how few approvals sit between install and a finished errand.
  • cost A household that outgrows the free allowance jumps to $240 or $1,200 a year with no middle plan to absorb light paid use, and the usage meter decides when that happens.
  • exposure Users who connect email, financial, health and smart home accounts give Meta a wider view of themselves than its social apps ever had, protected by architecture Meta describes and no third party has checked.
  • contradiction SiliconANGLE calls the privacy pitch a tough sell on Meta's record, while TechCrunch weighs it against an $18 billion consumer-harm settlement agreed two weeks earlier, which leaves the trust judgement sitting with whoever taps the permission prompt.

A user installing Muse on Monday types in a payment card before the agent has completed a single task, because the paid plans start as usage rises [4]. Then comes the part that decides whether the product is worth anything: connecting email, then calendar, then payments, then whichever health, smart home, dining or shopping app that person actually lives in, one service at a time [5]. Meta presents the one-at-a-time flow as making the opt-in visible, and it does. It is also the gate. Where Meta ships no built-in connector, the user hands over credentials for a service with a public API, and where there is no API, Muse drives the service through a browser instead [6].

The reach is real, though the conversion is unestablished. Four of the five ways into Muse work now, the AI glasses being the exception [3], and WhatsApp is the one that skips the download for people who would never install an agent app [2]. Neither TechCrunch nor SiliconANGLE attaches a user figure to that reach, so the scale of the advantage is asserted rather than counted [4].

The price ladder says who the paid product is for. There is no middle rung: the upper plan costs five times the lower one, $1,200 a year against $240 [1][2]. The more revealing artifact is the usage meter, which shows what percentage of allowance is left and warns before usage tips into billing [4]. Value here is metered in errands rather than seats, which makes the honest measure of Muse how many tasks a household finishes in month two, not how often the app opens.

The security design carries its own drag. Sentinel watches what leaves the user's virtual machine, matches it against a policy that authorised it, and where no policy exists sends a notification asking the user to approve the action [11]. Alexandr Wang's launch post called Muse always-on [19]; an agent that stops to ask is the slower version of always-on, and that is the correct trade to make. Meta says Muse cannot see passwords or payment methods and that conversations are not fed to its ads systems [13], and Link by Stripe issues a single-use card number per purchase so the agent never holds the real one [8]. SiliconANGLE's reading is that Muse nonetheless asks for a wider view of a person than Facebook or Instagram get, covering email, calendar, financial, health and smart home apps [22].

Anyone planning their own launch should weigh two numbers: surfaces you can reach, and consent decisions between install and first finished task. Meta has bought the first outright, but the second is untouched by distribution, and nothing in the launch material says how many people get through it [5].

What to watch

  • Whether Meta publishes how many Muse users get past the third connector and finish a task, rather than surface or open counts.
  • Whether Muse on the AI glasses ships, and how Sentinel's approve-this-action prompts work on a device with no screen to show them.
  • Whether the promised Shop Pay and 1Password integrations arrive, since checkout and stored credentials carry most of the paid tier's use cases.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories