Meta has handed its Muse assistant the phone. On November 12, the company’s chief AI officer, Alexandr Wang, announced that Muse can now place outbound calls to businesses in the United States on behalf of people who pay for the service — the sort of calls most of us postpone for days: the insurer, the cable company, the DMV, the restaurant that still takes bookings by phone.

The idea is simple: describe the errand in a chat, and Muse rings the business, talks it through, and reports back. — the latest step in Meta’s push to make Muse, launched in September, an everyday utility.

What changed#

Calling is not brand new. Meta tested it from September with a small group of early adopters who had to ask for access. What changed on November 12 is the audience: the feature is now open to any paying Muse subscriber in the US. Free users are still locked out, and so is everyone outside the United States — at least for now.

The jobs Muse is pitched for are deliberately mundane: reserving a table, querying an insurer, sorting out a cable bill, booking a DMV appointment. These errands run on waiting, and Meta’s bet is that an assistant with infinite patience is the right tool for them.

A woman relaxing on a sofa with coffee while her phone on the table shows a glowing voice waveform, holding a call on her behalf
AI-generated illustration for AI Frontier Post

What it costs#

Outbound calling sits behind Muse’s paywall. Plans start at $20 a month for 500 million Muse tokens a week; a heavier tier at $100 a month raises that to 3 billion tokens. No per-call price is published, and Meta has not said whether heavy callers hit a separate limit.

The timing matters for Meta. The company said last month that Muse has passed 10 million users.

The person on the other end#

Meta is not the first to hand an assistant the phone. Google’s Gemini offers a similar “Call for Me” feature, though only for eligible Pixel owners in the US; the startup Instinct has been selling the same idea; Grok Bot handles outgoing calls too, and Claude can do it through third-party integrations such as AgentPhone. The question now is whether anyone on the other end will stay on the line.

That is the awkward part. Reporting cited by TechCrunch notes that many businesses and receptionists hang up as soon as they realise the caller is not a person. And the voices are getting better, which cuts both ways: a more natural voice books more tables, and it also makes it harder for the person answering to know they are talking to software. User reports say the calls arrive with a META caller ID and that Muse introduces itself as an AI, though Meta has published no detailed documentation of the feature.

There is also a trust question. During the September test, human contractors stepped in on calls Muse struggled to complete; Meta pulled that back after employees raised privacy concerns about users unknowingly being overheard. The feature now on general release is, as far as public reporting goes, fully automated.

A smartphone on a restaurant table with a glowing blue voice waveform rising from it, suggesting an AI agent booking a table by phone
AI-generated illustration for AI Frontier Post

Why it matters#

Phone calls are a good test of how far agentic AI has come. Booking a table is easy for a person but hard for software: it means navigating a phone menu, understanding a stranger’s questions in real time, and confirming the result — all without a screen to fall back on. An assistant that does that reliably is doing real work, not just producing text.

It is also a test of the business model. Meta is charging for the privilege, so calling has to be good enough to justify $20 a month. If Muse books the table and reports back cleanly, the subscription starts to look like a personal assistant. If it gets hung up on, or hands the call back half-finished, it is an expensive party trick.

What to watch#

  • Hang-up rates. Will businesses adapt to AI callers, or screen them out?
  • Disclosure. How clearly does Muse identify itself, and do regulators push for stricter rules as AI voices get harder to tell from human ones?
  • The free tier and the rest of the world. Meta has not said when calling reaches free users or subscribers outside the US.
  • Reliability. The real metric is completion: how often Muse finishes the errand without handing it back, and how it copes with a changed time or a question it cannot answer.
  • The rivals. Google, xAI and a crop of startups are shipping the same capability. Whoever makes the fewest awkward calls wins the habit.

Sources#