After Connect in 2026, Meta announced a new set of interconnecting products: the personal intelligent agent Muse will be integrated into AI glasses in the coming months. Ray to Ban Meta marks the launch of the third generation of products, and the company also introduced lighter Audio glasses as well as the VR Glasses weighing about 100 grams. Unlike in the past where a voice assistant was added to the glasses, what is more noteworthy this time is that the interface, model, and execution capabilities are all incorporated into the same system. Muse can understand tasks based on the items, posters, or lists in front of the wearer, call upon cooperative services to continue processing them, and work in the background while the user is still talking to it. It is not yet fully available to everyone; Meta stated that it will be integrated into the glasses in the coming months, and some of the new hardware and features will also be implemented later this year.
What this release truly changes is not the parameters on the frame of the glasses, but rather the fact that intelligent agents are beginning to emerge from mobile applications. On mobile phones, AI typically waits for the user to open the interface, enter a question, and then display the answer on the screen; however, glasses naturally possess continuous visual and vocal context. When users see products on a shelf, they don't need to describe the brand and specifications again; when they see promotional posters, they don't have to take photos first, then upload them, and then copy the dates. With reduced entry barriers, the value of AI shifts from "responding in a human-like manner" to "being able to take on tasks at the right moment." This is also why Meta releases Muse, hardware, connectors, and payment systems together: without accessible services alongside model capabilities, intelligent agents remain at the chat level; with just glasses and no reliable agent, they are merely another wearable screen.
From voice answering to continuous proxying, glasses have changed the starting point of tasks.
According to Meta, Muse will be able to take action when the wearer sees real-world objects, and it will reach retail, travel, productivity, and payment services through more connectors. New partners include Walmart, Best Buy, Sephora, Wayfair, Notion, GitHub, and Box, among others. Shop Pay and PayPal have also been incorporated into the payment process. The key point here is not how long the list of partners is, but that intelligent agents are now possessing a complete chain from recognition to execution. They first understand the situation on-site, then select services, and subsequently complete searches, organization, filling out forms, or purchases; users only confirm at sensitive stages. For platforms, this represents a new competition for distribution rights: in the past, app stores decided what users downloaded, and search engines decided what users clicked on, but in the future, personal intelligent agents may decide who performs tasks.
This model also raises the requirements for real-time performance. In the context of glasses, users cannot afford to wait for long periods, nor can they repeatedly correct their inputs. The model needs to understand what "this," "that one on the left," and "the previous document" refer to, and it must also continuously process speech in noisy environments. Meta describes Muse as a personal agent, rather than a regular voice chat service, which means the system needs to maintain a longer task state, continue running while the user engages in other activities, and then re-engage when authorization is required. Although backend execution may seem seamless, it actually demands that task planning, tool invocation, identity authentication, exception recovery, and audit logging all be reliable simultaneously. If any one of these components fails, the consequences for the user could be an incorrect response, a misplaced order, a wrongly sent email, or a flawed navigation experience.
On the hardware side, there is also a re-layering around these all-weather entry points. The third generation emphasizes being thinner, lighter, having longer battery life, and offering faster operations; products with display capabilities provide visual feedback such as navigation and video calls. Those designed for immersive work and entertainment weigh around 100 grams. There are still efforts to integrate headset capabilities into everyday glasses. Different devices do not replace each other but cover different use cases: "just listening," "seeing prompts," and "immersive spaces." If a unified platform becomes the standard, users may no longer care which pair of glasses a particular function is associated with, but rather whether their tasks and memories can be carried across devices seamlessly.
The real barriers are permissions, privacy, and the cost of making mistakes.
Once personal intelligent agents are integrated into glasses, privacy concerns will become even more acute than during the smartphone era. Cameras and microphones may come into contact with passersby, family members, work documents, and consumer habits. Visual context contains far more incidental information than text-based inquiries. Meta It is essential for users to be clearly informed about when the devices collect data, which information is sent, what permissions the connectors have, and what actions are taken in the background for each task. A one-time consent clause is not sufficient; permissions should be specified down to the service, data type, and level of operation, and users should be allowed to revoke them at any time. Clear prompts regarding data collection and restrictions on usage scenarios are equally important for bystanders; otherwise, just because the technology is available does not mean it is socially acceptable.
Payment and shopping further amplify the cost of errors. Muse will be able to integrate with multiple retailers and payment services in the future, but being able to make payments does not mean one can completely let go of responsibility. Agents need to provide verifiable summaries regarding prices, quantities, addresses, subscription terms, and return conditions, and high-risk operations should be confirmed by the users. The platform also needs to distinguish between recommendations and advertisements to prevent agents from presenting a particular supplier as the best option due to commercial commissions. If agents communicate with merchants on behalf of users, the boundaries of responsibility must also be clear: whether the error stems from the model, the connector, merchant data, or user authorization, it cannot become an unaccountable gray area afterwards.
From an industrial perspective, the advantage of Meta is that it can form a closed loop with hardware shipments, social relationships, and consumption scenarios, but this does not automatically translate into trust. The more proactive the individual intelligent agents are, the more users need visible control panels, complete operation records, and easy-to-understand exit mechanisms. The more connectors there are, the greater the potential for attacks; therefore, the permissions of third-party services and their data retention policies must be incorporated into the overall security design. For developers, the new opportunity lies in providing structured, verifiable, and revocable service interfaces for proxies, rather than just creating pages that people click on.
Therefore, Connect 2026 is more of a directional statement: AI glasses are shifting from being shooting devices and voice accessories to becoming real-world terminals for personal intelligents. Muse The development of these glasses is still in phases, with many functions yet to be released and verified, but the product logic is already clear. The next round of competition will not be about who has the highest camera resolution, the most stylish frames, or the best model rankings; rather, it will be about who can enable agents to understand and execute tasks more quickly in real environments, while keeping control over permissions and errors firmly in the hands of users. What truly determines whether such products can become a daily necessity is not how smoothly they perform on stage, but whether users will dare to entrust them with those ordinary yet important tasks months down the line.
