Skip to main content

Meta’s Muse Phone Beta Uses a Human-Agent Layer

22 SEPTEMBER 2026·2 MIN READ·2 SOURCES·Independently corroborated

Meta is testing human call-center support for Muse, its AI agent that contacts businesses on users’ behalf, leaving unanswered when calls are handled by people instead of AI.

Meta’s Muse Phone Beta Uses a Human-Agent Layer

Key takeaways · 3

  • 01

    Treat Muse’s outbound calling feature as a hybrid human-and-AI service rather than assuming complete automation.

  • 02

    Organizations evaluating similar agents should establish clear disclosure policies for human access to user requests.

  • 03

    The sources do not explain Muse’s routing criteria or how frequently trained people complete calls.

How Muse Handles Calls

Meta executives announced that Muse’s beta for outbound calls to United States businesses had expanded, describing a feature that can call businesses to make restaurant reservations or haircut appointments. [1] An internal Meta post said Muse can call on a user’s behalf, handle a conversation, complete a request, return a transcript and summary, and hand the request to a trained human agent who places the call, while another internal announcement said a “human agent layer” had been added for calls to be completed. [1]

Questions About Human Involvement

404 Media reported that it remains unclear how often or under what conditions calls are routed to human agents rather than completed exclusively by AI; the outlet also said employees questioned the privacy implications of sharing requests with people and the perception that the assistant was not advanced enough to make calls on its own. [1] Bloomberg separately described Muse as a new AI app that makes shopping easier, attributing that characterization to Singh. [2]

What it means

The central issue is not whether Muse can initiate a business workflow, but how much of that workflow is automated and when a person takes over. The public beta messaging emphasizes completed tasks, transcripts, and summaries, while the internal description identifies a trained human agent as part of completion. That gap makes disclosure and request handling central to evaluating the product, especially because employees themselves raised privacy and perception concerns. What the sources don't address: what routing rules trigger a human handoff or how users are informed before their requests are shared.

Muse illustrates why practitioners must distinguish an AI product’s user-facing behavior from the human operations supporting it. Agent evaluations should examine handoff rules, disclosure practices, and who can access requests—not merely whether the requested task gets completed.

Why it matters
Daily session

Put this to work — one session a day, built for your industry.

Create a free account for a daily session — eight questions and one real-work challenge, on the news that affects your role.

Start free

How this developed

  1. 22 September 2026

    Meta’s Muse Phone Beta Uses a Human-Agent Layer

  2. 22 September 2026

    Event created from source cluster.

Sources

AI fluency, one session a day, built for your work.