Meta introduced Muse, a personal artificial intelligence agent that can shop online, prepare and send e-mails and plan travel on behalf of the user. The system, which the company has been working on for a long time, can carry out multiple operations independently in line with the given target, instead of just answering questions, unlike Meta AI. Meta states that no technical knowledge is required to install and use Muse. The artificial intelligence agent, which has its own applications and website, can prepare an appropriate plan after learning the user’s goal, coordinate time and resources, and advance the determined tasks. However, how comprehensively the system can work will be directly linked to which accounts and personal data the user allows access to.
Meta CEO Mark Zuckerberg previously announced that the company wants to develop artificial intelligence agents that can proactively carry out people’s daily tasks. Muse aims to bring this approach to a wide range of users, unlike more technical solutions such as OpenClaw. According to the information provided by Meta, the user will not only be able to give Muse a task, but will also be able to define a more general goal they want to achieve. The system will then be able to create a personalized work plan for this goal and continue the necessary steps on its own. Therefore, Muse is positioned as a software that can operate between different applications and works independently to a certain extent, rather than a chat-based artificial intelligence assistant.
Meta Muse works with Google Workspace, Spotify and Apple Health
Muse is not limited to services already supported by Meta. If a service offers a public API, the artificial intelligence agent can create the necessary connection to work with this service on its own. In his statement to Engadget, Vishal Shah, Vice President of Artificial Intelligence Products at Meta, states that Muse is also capable of creating its own software, and therefore it is not easy to draw the exact limits of what it can do. However, this flexibility brings with it the issue of access permissions. For Muse to perform more complex tasks, the user must grant access to more services and share more information about their personal preferences with the system.
Meta says that it has developed various security mechanisms to prevent an artificial intelligence agent with this level of access from performing uncontrolled transactions. Muse will ask the user for confirmation before performing any consequential actions, such as sending an email or completing a purchase. On the other hand, some tasks that the company classifies as low risk can be performed without additional approval. In addition, each user’s Muse instance will run on a private virtual machine. According to Meta, this method aims to keep a user’s agent and the data it can access separate from other users’ systems.
Despite all these precautions, Meta does not claim that Muse will run error-free. The company openly acknowledges that AI agents can inherently exhibit unpredictable behavior, and that Muse will make mistakes from time to time. Meta expects these errors to occur less frequently and to limit the damage that may occur, thanks to the security systems it has created. The scope of transactions requiring user approval will be decisive at this point. How the system will establish the balance between convenience and user control, especially in areas such as shopping, messaging and accessing personal accounts, will be more clearly visible during real use.
Muse’s usage area will not be limited to desktop or standalone applications. Meta also plans to integrate its AI agent with WhatsApp and the company’s smart glasses. Thus, users will be able to assign tasks to Muse and receive information about ongoing transactions while on the move. Zuckerberg’s previous statements regarding artificial intelligence agents indicate that Meta wants to turn these systems into tools that can run continuously in the background, rather than helpers that are turned on only when necessary. In addition to small daily tasks, Muse is planned to be able to work on longer-term and multi-stage goals.
However, one of the most obvious question marks before the widespread use of Muse will be privacy. In order for the system to work in a personalized way, it is possible to access different types of information, from e-mails to health data, from social media activities to shopping preferences. Meta will allow users to opt out of their interactions with Muse from being used in subsequent AI training. The company also states that the details of Muse conversations will not be shared with advertising systems. However, he acknowledges that some interactions, such as shopping, may indirectly affect the ads shown to users.
Muse’s pricing model has not yet been fully disclosed. Meta states that most of the features that users will need will be free. It is stated that those who want to perform more comprehensive tasks may need a paid Meta AI subscription in the future. It is currently unclear which features will remain within the free usage limits and at what point subscription will become necessary. While Muse offers a system that can perform much more work on its own compared to Meta AI, issues such as access permissions, error probability, privacy and pricing stand out as the basic factors that will determine how useful the product will be for users.
Join Channel