Meta Taught Its Muse AI Agent to Make Calls, but Many Businesses Hang Up
Meta has opened its Muse agent's calling feature to paid subscribers in the US. The agent dials a cable provider, a restaurant, or the DMV (the Department of Motor Vehicles) on the user's behalf and waits on hold. But many companies hang up as soon as they recognize an AI, and in September Meta temporarily routed calls to live operators. For builders of voice agents, the bottleneck is no longer voice quality but whether businesses are willing to talk to a machine.
AI-processed from The Decoder; edited by Hamidun News
Imagine you need to reach your cable provider: half an hour of hold music, then a menu that says "press five to speak with an agent." Meta wants an AI agent called Muse to take on that ordeal. It can now make calls to companies in the US on a person's behalf. But there is a catch: many companies hang up as soon as they realize an AI is on the line. For people building voice agents, this matters more than a polished demo. The calling works; whether the other end accepts such calls is still unclear.
What Meta launched
Muse is Meta's AI agent that runs continuously and carries out the tasks users give it. The calling feature was tested in September and is now available to all paid Muse subscribers in the US.
Alexandr Wang, Meta's head of AI, said the agent has more than 10 million users. According to him, Muse takes on tedious calls: to cable companies, restaurants, and the DMV, the American agency that issues driver's licenses and registers vehicles. The agent will soon also get a free email address.
Why people like this is simple. The AI can sit on hold indefinitely and work through complicated voice menus without getting annoyed. A person does not have to spend an evening on it; the agent does not mind.
Who else calls, and who doesn't
Meta is not alone here. According to The Decoder, Google Gemini, the startup Instinct, and xAI's Grok Bot can already call businesses. OpenAI's ChatGPT Dots and Anthropic's Claude Duet cannot yet.
So calling is becoming a standard item on the list of personal agent skills, alongside web search and calendar management. In 2018, Google showed Duplex, and the room gasped. Now a similar capability is becoming a routine feature of AI assistants that are always on.
The problem: the other end hangs up
The most interesting part of the story is not the feature itself but its weak point. Many companies end the call as soon as they recognize an AI. Because of this, in September Meta had to temporarily test Muse with live operators.
The history repeats itself. At Duplex's launch, a human handled every fourth call rather than an algorithm. Eight years later, companies are running into the same barrier.
Synthetic voices are becoming more realistic, and it is getting harder to tell an AI apart by ear. The source does not say this, but the logic is hard to miss: the better the voice, the stronger the business's incentive to look for other filters, such as asking callers to verify their identity or transferring them to another channel.
What this means for practitioners
First. If you are building a voice agent for calls, plan a fallback. Meta used live operators. The approach works, though it is not cheap: a human picks up the conversation when the agent fails or when the other party refuses to talk to an AI.
Second. Think about transparency. Some companies may be more willing to accept an agent that introduces itself honestly and explains why it is calling. This is our assumption; the news report contains no data on it.
Third. Track metrics that matter more than a pleasant voice: what share of calls reach their goal, what share were dropped, and in how many a human was needed.
There are limits too. We do not know what percentage of Muse's calls are dropped today or how often operators are involved. The Duplex figure refers to the launch in 2018. The feature works only in the US and only on a paid subscription. The direction is nonetheless clear: the question is no longer whether an AI can make the call, but whether businesses will answer it.
- Nov 21, 2026OpenAI Launches ChatGPT Wallet: Agents Now Make Purchases With Face ID
- Nov 21, 2026OpenAI launches ChatGPT Wallet: the agent buys goods, but never sees the card number
- Nov 21, 2026Meta's Muse Learns to Make Calls to US Companies on Your Behalf — But Only with a $20+ Subscription
- Nov 21, 2026CoreWeave Trains DeepSeek-V3 671B in 17.65 Minutes on Vera Rubin, 2.7x Faster Than GB200
Key Questions Answered
What Meta launched?
Muse is Meta's AI agent that runs continuously and carries out the tasks users give it. The calling feature was tested in September and is now available to all paid Muse subscribers in the US. Alexandr Wang, Meta's head of AI, said the agent has more than 10 million users. According to him, Muse takes on tedious calls: to cable companies, restaurants, and the DMV, the American agency that issues driver's licenses and registers vehicles. The agent will soon also get a free email address.
Who else calls, and who doesn't?
Meta is not alone here. According to The Decoder, Google Gemini, the startup Instinct, and xAI's Grok Bot can already call businesses. OpenAI's ChatGPT Dots and Anthropic's Claude Duet cannot yet. So calling is becoming a standard item on the list of personal agent skills, alongside web search and calendar management. In 2018, Google showed Duplex, and the room gasped. Now a similar capability is becoming a routine feature of AI assistants that are always on.
What are the key details on The problem: the other end hangs up?
The most interesting part of the story is not the feature itself but its weak point. Many companies end the call as soon as they recognize an AI. Because of this, in September Meta had to temporarily test Muse with live operators. The history repeats itself. At Duplex's launch, a human handled every fourth call rather than an algorithm. Eight years later, companies are running into the same barrier. Synthetic voices are becoming more realistic, and it is getting harder to tell an AI apart by ear.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.