Meta Muse calls US businesses, but many hang up on AI

Meta opened Muse’s business-calling feature to every paying Muse subscriber in the US after a September test, but many companies still end the line when they detect an AI, according to The Decoder.

The launch gives consumers direct access to an AI agent (software that acts on a user’s behalf), a capability The Decoder places in the lineage of Google Duplex’s 2018 demonstration.

Access does not guarantee completion. A company can terminate the call before the agent reaches a person, making acceptance by the receiving business the immediate deployment constraint.

Meta AI chief Alexandr Wang said Muse has more than 10 million users. He described its work as “tedious calls” in three categories:

• Cable companies
• The DMV
• Restaurants

Wang also said Muse will soon get a free email address. The report gives no rollout date for the address and no Muse subscription price.

What changed since Google Duplex?

Google Duplex demonstrated AI-mediated calls in 2018, prompting widespread amazement, The Decoder reports. Meta has now put similar functionality into a US product available to every paying subscriber. The Decoder frames that shift as a standard capability of always-on AI agents.

Meta’s change is product status, not invention. Duplex showed that a machine could navigate a business call. Muse distributes that capability through a service Meta says has more than 10 million users.

The market stakes are practical. Meta can grant access immediately, but a receiving company controls whether the interaction continues. Every rejected call turns that access into an unused feature.

Which rivals can already place calls?

The Decoder’s competitive snapshot divides five named services into two groups.

• Google’s Gemini — can already place calls.
• Instinct, a startup — can already place calls.
• xAI’s Grok Bot — can already place calls.

Still cannot place calls:

• OpenAI’s ChatGPT Dots
• Anthropic’s Claude Duet

Meta’s business-calling feature is therefore not an exclusive capability. It combines subscriber access with a user base Meta puts above 10 million, but The Decoder provides no launch dates, prices, geographic reach or call-volume data for the competing services. The comparison establishes feature availability, not market share.

The 10 million figure measures Muse users, not callers. The Decoder gives no paying-subscriber count, feature adoption rate or completed-call volume. Meta can therefore claim a large audience without evidence in the report that a large share has used business calling.

Why do businesses hang up?

The failure point sits on the receiving end. Many businesses hang up the moment they detect an AI, The Decoder reports. That reaction forced Meta to test human operators temporarily in September.

Human assistance provided a fallback when automated calls failed to pass the receiving end. It also complicates any claim that each Muse interaction is fully autonomous. The source does not disclose how many September calls involved an operator.

Google encountered the same constraint with Duplex. At launch, one in four calls, or 25%, was handled by a human, according to The Decoder. That 2018 figure shows human-assisted calling operated at a substantial share from the beginning.

Synthetic voices (machine-generated speech) keep getting more realistic, and The Decoder says that makes AI detection increasingly difficult. Businesses can still recognize an agent and terminate the call, but identification itself remains unreliable. Improved realism helps the caller and hurts the detection system at the same time.

What do users gain from waiting?

Users are drawn to Muse largely because it will “sit on hold indefinitely and work through complex phone menus without complaint,” according to The Decoder. The value proposition centers on endurance and menu navigation.

Wang named cable companies, the DMV and restaurants. The Decoder provides no task-completion figures for those categories, leaving the practical benefit of each use case unquantified.

For consumers, delegation changes who absorbs the waiting. The source emphasizes indefinite waiting and menu navigation. A service can sound natural and still fail if the company ends the call.

The forthcoming free email address adds a user-facing identifier for the agent. The report does not say how or when businesses will treat that address, nor whether it changes the detection problem that prompts hang-ups.

Can businesses accept machine callers?

Meta can now distribute machine callers through a paid US service. That does not mean businesses will accept them. Each call still crosses a company-controlled phone system, and the receiving operator can end it before any useful work occurs.

The report offers no figures for rejected calls, completed requests, human handoffs or accepted business categories. It also does not describe whether businesses are required to disclose that a caller is an AI. Those omissions prevent an assessment of how often Muse produces a completed service interaction.

For Duplex, the disclosed benchmark was one human-handled call in four at launch. Meta’s September test used people temporarily, but The Decoder reports no comparable rate. The older Google experience provides context; it does not supply a direct measure for Muse.

The Decoder does not quote a business explaining the refusal, so the motive beyond AI detection remains unestablished. The report also gives no detail on whether businesses detect an agent before pickup, during the greeting or after hearing a request. The source supports no comparison among those stages.

What will decide whether Muse succeeds?

Muse has cleared three visible hurdles: access for all paying US subscribers, a reported user base of more than 10 million, and competition from Google’s Gemini, startup Instinct and xAI’s Grok Bot. It has not cleared the endpoint problem. Many companies still hang up, and Google’s earlier launch handled 25% with a human.

The next decisive number is not another user-count milestone. It is the share of Muse calls that remain connected and complete their assigned task without a hang-up or human handoff. Meta has not disclosed that figure. Until it does, the US rollout demonstrates distribution of AI calling, not proven acceptance by the businesses Muse intends to call.