Meta’s Muse points toward a useful change in computing: being able to ask for help with what is in front of you without first opening an app and explaining the situation.
At Meta Connect, the company described a personal AI agent that works across connected services, handles tasks in the background, and returns with the result. Bringing that agent to glasses would let it respond to what a person sees and hears. A product on a shelf or a sign on a wall could become part of the request without the user typing a description.
Muse is available through its app, the web, WhatsApp, and a Mac desktop version. Meta says it is adding connections to services including Shopify, Shop Pay, PayPal, GitHub, Notion, Granola, and Box, along with retailers and travel services. Its plans for glasses include helping users log meals, book appointments, find travel options, and act on something in view.
Those are Meta’s descriptions of announced and forthcoming capabilities. Availability varies, and a demonstration does not establish how reliably the system will perform an ordinary task. Booking the wrong appointment is still a problem even if the process takes fewer steps.
What interests me is the possibility of carrying context across those steps. Today, a person might find information in one app, copy it into another, and open a third to complete a transaction. An agent with the right connections could handle that sequence from a single request. Glasses would extend the same approach to situations where taking out a phone interrupts what the person is doing.
That could be useful while cooking, making a repair, or moving through an unfamiliar place. The value would come from completing the task while allowing the person to keep paying attention to the activity itself. A device that constantly interrupts with suggestions would miss the point.
I am optimistic about this direction. Computing has become powerful enough to assist with much more of daily life, but using it still requires us to spend considerable time managing screens. An agent that can understand a request in context and follow through could remove some of that work.
The permissions need to be equally understandable. People should know what the agent can see and remember, which accounts it can use, and when it needs confirmation before acting. The consequences of a mistake change when a system moves from answering a question to spending money or making a commitment on someone’s behalf.
Meta still has to earn that trust through reliable products. Phones and laptops will remain useful, and it is too early to know which company will make this approach work best. But the test is practical: can the agent complete something useful with less effort from the person using it? That is what I will be watching as these capabilities reach everyday use.