Meta launches Muse Code AI agent for end-to-end software engineering
The company's first dedicated coding agent moves beyond autocomplete to manage complex tasks across large repositories.
Meta CEO Mark Zuckerberg has announced the beta release of Muse Code, the company's first dedicated AI coding agent. The tool marks a strategic shift for the social media giant as it moves from providing simple code suggestions to offering a system capable of managing entire software development lifecycles.
Available as a terminal-based agent for macOS and Linux, Muse Code is designed to handle complex, end-to-end software engineering tasks. According to Zuckerberg, the agent can manage work across large repositories by planning necessary changes, writing the actual code, and validating the results. The system is powered by Muse Spark 1.2, a new model update specifically optimized for coding tasks. To handle the volatility of complex workflows, Muse Code utilizes restart-safe execution and persistent asynchronous subagents, allowing it to maintain progress across long-running tasks.
The Shift to Agentic Engineering
This launch arrives as Meta intensifies its competition with rivals OpenAI and Anthropic in the developer tools market. While previous AI assistants focused on snippet generation or autocomplete, the industry is pivoting toward "agentic" software engineering. In this model, the AI does not merely suggest a line of code but acts as a collaborator that can navigate a codebase, identify dependencies, and execute a full cycle of planning and testing without constant human intervention.
Market Implications and Performance
By targeting high-end software engineering workflows, Meta is attempting to reduce the manual overhead associated with maintaining large-scale codebases. However, the tool faces stiff competition in terms of raw performance. In benchmarks published by Meta at launch, including Terminal-Bench 2.1, Anthropic's Claude Opus 5 outperformed Muse Code. Despite trailing Claude, Meta's data indicates that Muse Code outperformed other industry competitors, such as OpenAI's Codex and Grok Build.
What's Next
As Muse Code moves through its beta phase, the industry will be watching to see if Meta's integration of persistent subagents provides a stability advantage over other LLM-based tools. While the company has signaled an aggressive pricing strategy to attract developers, the primary measure of success will be whether the agent can reliably handle the unpredictability of production-grade repositories without introducing critical regressions.