Muse Daily: Oct 1 — The Phone Calls Weren't Always AI: What the "Human Concierge" Story Means for You

Here's a scenario that should make you pause. You ask Muse to call your internet provider and haggle over your bill. The call goes through, the negotiation happens, and you get a nice summary afterward. Everything looks like AI magic, except the voice on the other end of that call belonged to a human contractor sitting in a call center somewhere, and nobody told you.

That's what Reuters found happening inside Meta last month, and the details are worth your attention if you plan to let any AI agent make calls for you.

Photo: Petiatil (CC BY-SA 3.0)

What the phone feature actually does

Muse's phone-calling feature works the way you'd expect an agent to work. You tell it to dial a US business and handle an errand: book a haircut, check whether a store has an item in stock, get quotes from contractors. The agent places the call and talks to whoever picks up. Meta started testing it with employees in August and rolled it out to users in the days after the app's September 8 debut.

The problem Meta ran into is a very human one. People on the receiving end kept hanging up. One employee told Reuters he tried calling his insurance company and "they keep hanging up on Muse when they hear it is AI." Businesses don't like talking to robots, and an agent that can't complete calls isn't much of an agent.

So Meta tried something else. For half of its employees, it switched on a "human concierge," also called "human agent calls" in internal posts. The pitch, quoted by Reuters: "Muse is now able to hand requests to a trained agent, who places the call and works it through." When the AI couldn't get the job done, a person quietly took over. Employees who didn't want in could join an opt-out group.

Why employees pushed back

Some employees were not happy. One wrote in an internal post: "It's baffling to me why we think this feature is worth the risk. We are one bug away from unnecessary information being leaked to human callers."

That risk stopped being theoretical fast. Another employee had asked Muse to call his internet and cable provider to negotiate his bill. The transcript showed the human contractor had made a racist reference during the call. The contractor was removed from Meta projects, but the incident made the core issue concrete: when a stranger handles your call, your account details, your billing history, and your identity are in someone else's hands.

Meta's response was quick. A vice president in the Superintelligence Labs unit told employees that "it was a miss" to start testing contractor-placed calls without proper disclosures, and said the company had "rolled back this feature" for now. She added that Meta was still working on AI calling, noting that human-handled calls were hitting 95 to 98 percent success rates, compared with the lower success rate of AI-only calls. She also apologized to the employee affected by the racist remark.

We've seen this movie before

This isn't Meta's first rodeo with humans behind the AI curtain. About a decade ago, when the company was still called Facebook, it trialed a digital assistant called "M" inside Messenger. Media reports later estimated that around 70 percent of the assistant's tasks were actually carried out by humans. The technology demos looked autonomous. The labor wasn't.

A Meta spokesperson, Daniel Roberts, told Reuters the employee response to the concierge test had been "overwhelmingly positive" and that its purpose was to "get feedback so we can implement safety and privacy protections and improve features before we release them publicly." He added: "We're working with merchants to continue improving this potential calling feature, and will only roll it out when it's ready and with the proper disclosures."

Photo: Shixart1985 (CC BY 2.0)

What this means for you

Here's my read. The human-concierge episode tells you two things at once. First, AI phone calls are genuinely hard: real businesses hang up on robots, and success rates without human help are low enough that Meta felt the need to cheat a little. Second, any feature where an agent "makes calls for you" deserves more scrutiny than the average chatbot trick, because a call involves your identity, your accounts, and a stranger on the other end.

If you use Muse's calling feature, or any agent's, here's the order I'd follow:

1. Assume a human might hear it. The concierge test was rolled back, but the concept isn't dead. Meta said it's still improving the feature. Before you hand over a call, decide whether you'd be comfortable with a contractor hearing everything on it. If the answer is no, don't delegate that call.
2. Never put account numbers or passwords in a phone task. Account numbers, PINs, passwords, Social Security numbers: keep them out of anything an agent might read aloud or a human might transcribe. If a call genuinely needs verification details, handle that part yourself.
3. Check what the disclosure says before you use it. Meta has promised "proper disclosures" before any public rollout. Read them. Specifically, look for whether calls are AI-only, whether humans may be involved, and whether the business on the other end is told it's talking to an agent.
4. Use it for the boring part first. The safest version of this feature is hold music. One practical approach people are already using: ask Muse to call customer service, wait on hold, and ring your own phone back when a human agent picks up. You do the talking; the AI does the waiting. If the calling feature isn't available on your account yet (it's still in beta and rolling out gradually — ask Muse whether it can place calls before assuming it can).
5. Remember what you're trading. An agent that negotiates your cable bill needs your account. An agent that books your haircut needs your name and number. Every delegated call widens the circle of who knows your business. That's fine when the payoff is real, but make it a conscious trade.

The bigger picture is simple: agents are moving from typing to talking, and talking is messier. Typing leaves a transcript you can review. A phone call happens in real time, with real people, and as Meta just learned, sometimes the "AI" on the call isn't AI at all. Treat delegated calls like you'd treat handing your phone to a stranger: useful sometimes, but never casual.

---

Sources:
- Reuters via CNA, "Meta testing a 'human concierge' for its new personal AI agent, Muse" (Sept 22, 2026): https://www.channelnewsasia.com/business/exclusive-meta-testing-human-concierge-its-new-personal-ai-agent-muse-6402946
- WDC News 6, "10 Ways People Are Using Meta's Muse AI App": https://wdcnews6.com/10-ways-people-are-using-metas-muse-ai-app/

Want to try Muse yourself? muse.ai — use my invite code P74B9O at signup. This is my referral code, so I'll receive tokens too.