Assistants tend to become habits when you can simply talk to them. Meta has now added real-time voice conversations to Muse, its personal AI agent, along with an avatar at the top of the chat that animates in time with what it says. Mark Zuckerberg announced the feature, and it is rolling out now in the Muse app in the US.
In The Verge's early hands-on, the agent answered quickly by voice, pausing at times while it looked something up or checked a task, much as it does in text. You can change the tone of the voice by asking the agent to pick one or by browsing options in the avatar settings. The avatar can be customised heavily: the reviewer set his to look like Mark Zuckerberg as a reminder of whose service he was using, and found the talking likeness uncanny, though he could not make the voice sound exactly like Zuckerberg. Meta says voice mode and the expressive avatars are coming soon to its AI glasses and will be available on the Muse Charm device, which is due to ship this year.
For readers of our Meta Muse review, this lands on a product that already stands out for price. Muse is free to start, with paid Power and Maximum plans at $20 and $100 a month that buy more weekly usage rather than extra features, and it works in its own app and inside WhatsApp. Our review rates its security design as the most ambitious of the four always-on agents we compared, while warning that its terms leave users holding the risk. Voice makes those trade-offs more pressing, because spoken requests are faster to make and easier to make carelessly.
The business logic is visible elsewhere. On Meta's latest earnings call, Zuckerberg said Muse will only become profitable if Meta can bring compute costs down, and real-time voice and animation are not cheap to run. Wider reach through glasses and a dedicated device is the bet that makes those costs worth it.
The open questions are about likeness and control. Avatars that can be made to resemble real people raise obvious impersonation concerns, and Meta's launch materials, as reported, do not say what limits apply. Nor is it clear how voice conversations are stored or used for training, or whether the free tier's unpublished usage limit covers voice the same way as text. Those details matter more than the animation.