Meta's Muse Code AI agent tackles large codebases

Meta is stepping up its AI game with the launch of Muse Code, a terminal-based AI coding agent currently in beta that aims to assist developers with complex tasks across large software repositories. The agent, powered by Meta's existing Muse Spark coding model, can handle complete software engineering workflows—planning changes, writing code, and validating results—by deploying its own sub-agents that work in parallel within isolated worktrees. This approach ensures no collisions occur in the working copy, even when handling multiple concurrent tasks. Meta CEO Mark Zuckerberg highlighted in a social media post that the agent successfully built six features for a game simultaneously during testing, demonstrating its ability to manage large-scale projects efficiently.

Meta is positioning Muse Code as a more affordable alternative to competitors like OpenAI's Codex and Anthropic's Claude Code, particularly emphasizing its cost advantages. The move reflects Meta's broader push to expand its AI presence beyond its traditional focus on advertising, with recent investments in enterprise AI tools such as customer service agents. Meta's AI chief, Alexandr Wang, told the Wall Street Journal that the agent could be an incredibly good option for many workflows and use cases, especially from a cost perspective. The release underscores Meta's efforts to catch up in the AI tools race while leveraging its existing models to deliver competitive solutions.