Meta rolls out Muse voice and animated avatars in the US Meta rolls out Muse voice and animated avatars in the US

An agent with a speaking face

Users start a conversation through the phone icon. Meta offers voice-only, alongside-chat and full-screen views, with animated expressions and gestures. The company says people can interrupt responses. Support for AI Glasses and Muse Charm is still forthcoming.

The September preview becomes an app feature

Meta presented the broader plan at Connect in September. Its September 24 recap described conversations continuing while Muse worked on tasks in the background. It also named Notion, Granola, GitHub and Box among work connections, alongside shopping and payment partners. That context matters because spoken requests need somewhere useful to go. A conversational interface can collect instructions, while a connected service supplies the document, repository or purchase workflow. The Connect presentation tied the same agent to glasses and a pocket-sized Charm device, putting conversation at the center of its proposed hardware family.

Google puts the camera into the conversation

Google took a different route with Guided Vision in Gemini Live, announced October 1 for compatible Android devices. Users share a camera view and receive spoken descriptions, follow-up answers and instructions for reframing the image. Google developed it with blind and low-vision users and reports participation by more than 1,000 Aira testers. Its examples include reading labels and finding misplaced objects. That makes the camera an input to a practical task. Google explicitly excludes navigation, obstacle detection and replacement of established mobility aids from the feature’s intended use.

OpenAI connects speech to existing work

OpenAI’s September 23 release notes described plugin support during Voice conversations on web, iOS and Android. They also described Voice in Work, including document, presentation and spreadsheet creation, connected apps and browser tasks. Unfinished work can continue in text after a call ends. Access still follows the account’s available plugins, permissions and usage limits; Voice in Work requires both capabilities. The distinction gives this comparison a useful baseline: speaking to an assistant, granting it access to a service and reviewing its output are separate parts of the experience.

The result needs to survive the conversation

These releases make different choices about what speech controls: an agent’s tasks, a live camera interpretation or an existing workspace. For software teams, that shifts attention toward the transition between a conversation and a lasting result. A pleasant exchange supplies little evidence about whether a file was created correctly. The artifact, transaction or recorded change remains the thing a user can inspect.