Meta Is Bringing Muse to AI Glasses—What Changes When an Agent Can See

What changed

At Connect 2026, Meta said its Muse personal agent will come to AI glasses in the coming months. The company demonstrated an agent that can respond to what a wearer is looking at, continue voice conversation while it works, use connectors, and eventually transact through services such as Shop Pay and PayPal.

The practical change is authority. Glasses can make the transition from observing an object, sign, shelf, screen, or person to proposing an action feel nearly frictionless.

Availability boundary

“Coming in the next few months” is not general availability. Meta lists work connectors including Notion, Granola, GitHub, and Box, plus shopping, payment, and future travel services. Buyers should record the exact device, account, region, connector, permission, and rollout they can access.

Why small teams should care

Hands-free context could help field teams retrieve procedures, create follow-ups, inspect stock, translate signs, or document maintenance. It could also expose customer faces, private screens, health information, payment details, or confidential conversations before a bystander understands that AI is involved.

An agent's own email address adds another identity to govern: who owns it, what it may send, how recipients identify agent mail, which credentials it can use, and how access ends during offboarding.

A cautious test

1. Begin read-only with public or synthetic data.
2. Test clear visual and spoken disclosure.
3. Disable purchases, outbound email, and sensitive connectors until logs and approvals work.
4. Include bystanders, accessibility needs, weak connectivity, mistaken objects, environmental prompt injection, and offboarding.
5. Compare with a phone and non-camera voice option.

Measure completion, wrong-object selection, unauthorized disclosure, approval comprehension, objections, corrections, failures, and time saved.

Bottom line

Muse on glasses could make an agent useful where physical work happens. It also compresses seeing, remembering, deciding, and acting into one wearable surface. Start read-only, make capture unmistakable, and grant one connector and one action at a time.