Meta explained the similarity between the new AI agent Muse and the OpenClaw project. Head of Product at Meta Superintelligence Labs, Nat Friedman, stated that the team indeed independently created Muse, but was "strongly inspired" by OpenClaw during its development. The discussion began after users noticed similarities in the structure of the two systems, including workspace file names and the content of SOUL.md – a document that sets the agent's behavior principles, communication style, limitations, and skill set. In light of this, co-founder of the AI project Vibecode, Ansh Nand, called Muse an adaptation of OpenClaw for a mass audience. Friedman did not deny the influence of the project and shared that he first became acquainted with OpenClaw in January. After that, the MSL team acquired hundreds of Mac minis, and employees began actively using OpenClaw and other tools for working with personal AI assistants.
According to Friedman, Meta's interest in OpenClaw is related not only to the concept of a personal agent but also to its infrastructure. A company representative specifically noted the contribution of one of OpenClaw's key developers, Peter Steinberger, and called the architectural shell he created for interacting with the agent one of the project's notable innovations. However, Meta emphasizes that Muse was developed as an independent system. The main task was to transfer the idea of a constantly available personal AI to an environment where additional security and user data processing requirements could be met. Therefore, the similarity of concepts and individual interface elements is seen by the team as the result of borrowing common ideas rather than a technical reproduction of the original project. This approach reflects a broader interest of AI developers in agents capable of independently performing a sequence of actions instead of responding only to individual user requests.
Meta also outlined several requirements that should distinguish Muse from existing personal agents. The system must be secure when working with user data, ensure confidentiality, and remain simple enough for everyday use. Another task developers mention is scaling the technology to a very large audience. That is why, according to Friedman, the company did not limit itself to transferring the already existing concept of OpenClaw but built its own implementation considering Meta's infrastructure and requirements. As a result, Muse is viewed as a separate product, although its architectural ideas and application scenarios were formed under significant influence from OpenClaw. The discussion around the two projects shows that personal AI agents are gradually transitioning from experimental tools to systems that should independently work with files, applications, and other digital user resources.
Related articles
