- Published on
Everything new coming to Meta’s AI agent Muse
- Authors
- Name
- DeepHorus
Meta is rapidly expanding Muse, the personal AI agent introduced only weeks ago, as it targets everyday tasks across apps, devices and user accounts. At the company’s annual Connect event in Menlo Park on Wednesday, CEO Mark Zuckerberg called Muse “the centerpiece” of Meta’s consumer AI strategy, predicting it could become a personal superintelligence used by billions. The social networking giant is betting that software capable of taking action, rather than merely answering questions, can become its biggest consumer AI platform yet. The move positions Meta against rival agent efforts from OpenAI and Anthropic and extends its ambitions from apps and wearables into desktop computing.
Muse launched in early September as an assistant that connects to services including email and calendars and works across different devices. It is powered by Muse Spark, a multimodal Meta model built for agentic work, with automation of routine tasks at the center of Meta’s pitch. Zuckerberg also referred to Jolly as the physical avatar associated with Muse, highlighting the company’s interest in giving its AI a recognizable form. Meta presents Muse as a broadly accessible alternative to more specialized competitors, although its real value will depend on how accurately it handles sensitive, everyday work.
One of Connect’s most eye-catching announcements is Muse Realtime Avatar, a new model that will let users create a digital character for their agent and hold real-time video conversations with it. The avatar will have a face, body and voice, allowing users to ask questions and assign tasks during the call. Meta demonstrated the capability onstage, giving a practical AI feature a distinctly Metaverse flavor while also aiming to make the assistant more personable. The feature is described as coming soon, without a precise launch date.
Meta is also bringing Muse to its smart glasses, where users will be able to invoke the agent by voice using a wake word. Proposed uses include personalized workout guidance, meal logging, appointment booking and help buying products the wearer sees, with the agent working in the background and checking back when a task is finished. Meanwhile, a forthcoming Mac version will let users delegate work involving desktop applications, from business operations to routine computer jobs. Alexandr Wang, Meta’s chief AI officer, said users can queue up tasks and walk away while Muse continues working; Meta also previewed a dedicated email address for the agent.
Zuckerberg framed Muse as a tool that could save users time and eventually help them make money. Meta plans to offer a large allowance of free usage tokens, but expects to profit over time by charging a small fee on transactions completed by the agent. That approach could reinforce Meta’s ambitions in commerce by placing an AI intermediary between users and products or services. It also means transaction access may become a central part of the agent’s business model, making fee transparency and meaningful user control important issues for its wider rollout.
The breadth of the keynote drew attention to the ambitious scope of the rollout: Meta is treating Muse as a cross-device platform, not a collection of isolated AI features. Its background operation and access to personal accounts could be genuinely useful, but they also raise questions about permissions, privacy, safeguards and what happens when an automated action goes wrong. Smart-glasses support is expected “in the coming months,” providing the first clear deployment window, while other capabilities remain less precisely scheduled. The next phase will test whether Meta can make agentic computing reliable enough for daily use; success could help the company normalize AI that acts across a user’s digital life, while failure would intensify concerns about security and loss of control.
