Meta now lets you talk to your Muse AI agent by voice, in real time. Mark Zuckerberg announced it on 5 November 2026. The Verge’s reporter had the feature on the iOS app the same day and tried it while writing the story, with the avatar at the top of the screen responding as the conversation went on.

A voice and a face feel like progress

A bot that talks back and looks at you feels closer to a person. That is the excitement around a launch like this, and I understand it. The demo is impressive, and I have no argument with the engineering. My question is the one I ask of every launch. What need does it meet, and what changed for a business on a normal day?

Voice helps when you are giving instructions

Voice works well on input. I use Wispr Flow to talk to Claude, and I dictate most of my instructions that way. Speaking a brief is faster than typing it. The brief also comes out more complete, because I say what I mean instead of trimming it to save my fingers. In the instructional space, voice clearly helps, and people will use Muse for that.

Voice helps much less on output and two-way conversation

Output is a different job. When I need an answer, I want to read it, scan it and copy it. A spoken answer runs at the speed of speech and leaves nothing I can check later. A two-way voice conversation with a bot is slower still, because you wait for it to finish before you can correct it.

In commerce the gap is wider. Look at what actually happens in a customer conversation that leads to money. An order gets looked up. A slot gets checked. A payment link gets sent. A delivery gets booked. Every one of those steps is data moving between systems. The voice sits on top of a text and data exchange, and it does not make that exchange any better.

The conversations that turn into orders are becoming agent to agent

More and more of the conversation will be agents talking to agents. A customer’s assistant asks a shop’s assistant whether a product is in stock, what it costs delivered, and whether Thursday works. Agents don’t talk to each other in voice. They exchange data, and data is text. So voice has limited room to expand into exactly the conversations a business cares about most.

That is why Meta’s news solves nothing on the output side for an operator. The face at the top of the app is talking to a human who is listening right now. The orders and bookings heading your way are increasingly placed by software that reads, compares and decides. That software is not listening to your bot’s voice.

What this means for a clinic or a shop

Take a clinic. A patient’s assistant asks whether there is an appointment this week, what the consultation costs, and whether the deposit is refundable. What the clinic needs is for those answers to exist as data: opening hours, prices, availability and deposit terms, written precisely enough that software can read them without guessing. A talking bot on the clinic’s website adds nothing to that. A plain, specific, machine-readable answer adds everything.

A shop is the same. Stock, price, delivery cost by area, return window, what is different about this product compared with the one next to it. The agent on the other side picks the business it understands most clearly and can quote most precisely, and that is something you can work on. The attributes have to be materially different from your competitors’, and they have to cover the many slightly different ways people ask the same question.

I see this from the inside at AeroChat. Customer conversations that turn into actions in a merchant’s systems run on text and data, which is the layer we build on. Even when a customer sends a voice note on WhatsApp, it is transcribed to text before anything happens with it. Voice automation is parked on our side for the same reason.

What a business should actually do

Spend on making your answers data, not on making your bot talk.

  • Write down every fact a customer’s agent would ask for: prices, stock, hours, delivery areas and costs, deposit and return terms. Keep them specific and dated.
  • Put those facts on your own pages in plain text and in a form a machine can read, not buried in an image or a PDF.
  • Connect the systems where the real answers live, so the order status, the diary and the payment link come from the source and not from a script.
  • Use voice where it helps you: dictate your briefs, your workflow descriptions and your replies.
  • Skip the talking avatar for now. Nothing in your order flow is waiting for it.

Source: The Verge: Muse has a voice mode now