Meta plans to bring Muse, its task-performing AI agent, to its AI glasses in the coming months. The concept is straightforward: rather than reaching for a phone or opening a desktop app, a wearer would say Muse’s name, state a request and let the assistant work in the background until it has something to report.
That could mean asking it to check flight prices, arrange an appointment or locate a particular item. Meta also presented the glasses version as a hands-free aid for ongoing personal routines, including customized workouts, meal logging and nutritional information. The central promise is not simply answering a question aloud; it is delegating a task while the person keeps moving through their day.
Muse is already available through standalone iOS, Android and Mac apps, as well as inside WhatsApp chats. Its expansion to glasses therefore appears to be a new interface for an existing service, rather than a separate agent built exclusively for eyewear. For anyone following the company’s wearables strategy, it is another step in turning smart glasses into an always-near access point for communication, assistance and now task execution. Meta has also been broadening what its glasses are meant to do in other ways, including a camera-free audio-oriented smart-glasses option.
Muse is an agent, not just a chatbot
The important technical term here is agentic AI. A conventional chat assistant generally responds with information, drafts or suggestions. An agent is intended to take a multi-step task and pursue it across connected services. Meta says Muse can use the web, book travel, make phone calls, send emails and make purchases on a user’s behalf.
That distinction matters because the risk changes with the action. Getting a response to “what are the cheapest flights?” is one thing. Having software move from searching to selecting an itinerary, providing details and completing a booking is considerably more consequential. The same applies to calls and email: the value is in removing steps, but each removed step can contain personal information, an irreversible choice or a real-world cost.
Meta says Muse requests permission before sensitive actions, including purchases. Users can decide which external services Muse is allowed to access and can revoke those connections at any time. Those are meaningful controls in principle. Permission is the point where a user can intervene before an assistant crosses from research or preparation into an action with financial, personal or reputational consequences.
The company has not supplied a firm release date for Muse on glasses, nor has it detailed exactly which glasses will support it, which services will be available through the wearable interface or how each type of request will be confirmed. The stated timing is simply “the coming months.” That uncertainty is worth keeping in mind when considering the examples: they establish the direction of the feature, not a complete feature list or a guarantee that every action will work identically in every setting.
What the glasses interface changes
On a phone, asking an agent to work usually means opening an app, typing or speaking the task and returning later for the result. Glasses may compress that opening step to a wake phrase and a spoken request. Meta says Muse will work in the background after being activated and will check in when it is finished.
That makes the most sense when hands and attention are already occupied. A runner could request an action without interrupting their stride to unlock a handset. Someone cooking, carrying bags or walking between appointments could begin a task without pausing to navigate an app. The proposed workout and meal examples similarly rely on the fact that glasses can stay on during activities where holding a screen is inconvenient.
Still, the strongest benefit described so far is convenience, not a wholly new capability. Muse can already be accessed on phones, computers and WhatsApp. A person who is comfortable issuing a request through one of those routes may not find pulling out a phone especially burdensome, particularly for a task involving money, travel details or a message that deserves a careful review.
That does not make voice access pointless. Reducing the friction of starting a task can be valuable, and the background-processing model means the wearer does not need to wait for every step. But it does frame the near-term question for prospective users: is hands-free initiation enough of an upgrade to justify linking more of one’s services and daily routines to a device worn on the face?
Trust is the real product requirement
Muse’s listed capabilities demand an unusually high level of trust. Booking travel can involve schedules, passenger details and payment. Sending email may involve private correspondence. Making calls can expose a user’s intent or identity to another person. Tracking items may reveal shopping interests. Meal and workout use can create a record of personal habits.
Smart glasses intensify that calculation because the interaction is designed to be ambient. An assistant that can be summoned with a spoken phrase and assigned an errand without a screen is meant to feel effortless. Yet effortlessness is not automatically the same as clarity. For sensitive work, users may reasonably want to know what service will be accessed, what information will be shared, whether the action is a draft or a final submission, and when they will be asked to approve it.
Meta’s permission-before-purchase policy is therefore one of the most important stated details. It indicates that at least some high-impact actions are not meant to happen silently. The ability to choose external connections and later revoke them gives users a way to limit the agent’s reach. In practical terms, a cautious approach would be to connect only services that are necessary for a desired use case, and to treat revocation as an available safety valve rather than an obscure account setting to ignore.
There is also a difference between delegating discovery and delegating commitment. Asking Muse to compare flight prices or find an item is a lower-stakes starting point than allowing it to complete a purchase. Asking it to prepare the ingredients for an email is different from authorizing it to send the message. The available information establishes that Muse can perform substantial actions and uses permission for sensitive ones, but it does not spell out the confirmation flow for every kind of task. Users should not assume identical safeguards across every connected service until Meta provides those details.
Practical use cases, and where caution belongs
The travel example is easy to understand. Voice initiation could be useful when a person remembers a task mid-activity: “check flight prices” can become a background job instead of a note to deal with later. The potential limitation is equally obvious. Flight shopping can involve flexible dates, airports, baggage rules and many preferences that may be difficult to convey in one short spoken request. The assistant may be most helpful in narrowing options, while the traveler retains the final comparison and booking decision.
Appointments offer another plausible case. A user could ask Muse to find or arrange a time while busy elsewhere. But appointments may involve calendars, locations and sometimes sensitive context. The convenience of hands-free setup has to be balanced against how much information needs to pass through linked accounts and whether the requested time actually matches the user’s priorities.
For item tracking, the value may be speed: a quick spoken request can send the agent looking while the user continues with another activity. As with travel, the result may be more useful as a shortlist than as an instruction to buy immediately. Meta specifically says users can determine which services are connected, so the scope of what Muse can do should depend on permissions the user has chosen rather than on an unrestricted mandate.
The fitness and nutrition ideas are somewhat different. They suggest Muse could act as a continuing companion, helping guide a workout, record meals and surface nutritional insights. These are less about finishing one discrete errand and more about building a habit around repeated voice interactions. That could be appealing to people who dislike handling a phone while exercising. It also means the utility depends on consistent use and on a wearer’s comfort with the personal information involved.
An early test of whether AI glasses earn their place
Muse on glasses puts a useful question to Meta’s broader AI hardware vision: what should a wearable assistant do that a phone-based assistant cannot do well enough? The answer offered here is immediacy. A request can begin in the moment, hands-free, and the software can return when it is done.
For some people, that small change in access could be the difference between acting on an intention and forgetting it. For others, the phone will remain the preferred place for tasks that require context, review and confidence. The glasses’ appeal will likely hinge not just on whether Muse can perform errands, but on whether it can make its progress, permissions and final actions understandable without making the supposedly frictionless experience feel risky.
Meta has laid out a broad ambition for Muse: move from answering requests to taking meaningful work off a user’s plate. Bringing that ambition to AI glasses makes the pitch more personal and more immediate. It also raises the bar. A hands-free agent must be useful enough to save time, transparent enough to supervise and restrained enough that wearers can decide exactly how much of their digital life it is allowed to touch.
