Meta launches Muse Code: AI agent for large codebases

Meta, which had once lagged behind in the artificial intelligence application race, is now accelerating to catch up with its peers. This week, the company launched a new terminal coding agent to help programmers handle complex development tasks within large software codebases.

Meta CEO Mark Zuckerberg stated on social media on Wednesday that the currently beta version of Muse Code can "complete full software engineering tasks in large code repositories," covering the entire process of "project planning, code writing, and result validation."

Users only need a single command to install Muse Code, which is based on Meta's previously released large code model, Muse Spark. When faced with large projects, it automatically generates multiple agents to work in parallel.

Zuckerberg explained, "When the task volume is large, the system will split off independent sub-agents to work in parallel in isolated working directories. The entire process will not alter your local code copy. In testing, this tool simultaneously developed six features for a game without any code conflicts."

Meta's move aims to enhance its competitiveness, benchmarking against competitors like OpenAI's coding agent Codex and Anthropic's Claude Code with a higher cost-performance ratio.

Alexander Wang, head of Meta's Super Intelligence Lab and AI director, stated in an interview, "We believe this product is highly competitive for a large number of development processes and application scenarios, with particularly notable cost advantages."

Meta continues to invest heavily in research and development, constantly expanding its AI footprint. In June of this year, Meta stepped out of its traditional core direction of AI-enabled advertising business and officially entered the enterprise-level AI market, launching agent products aimed at customer service scenarios