Meta已发布AI编程代理Muse Code，目前处于测试阶段[1]。该工具由Meta的编码模型Muse Spark提供支持[1]，旨在在大型代码库中完成完整的软件工程任务[1]。根据Meta首席执行官Mark Zuckerberg的介绍，Muse Code可以处理"规划变化、编写代码、验证结果"等工作[1]。
该产品的核心创新在于其并行处理能力[1]。当任务规模足够大时，Muse Code会启动多个子代理在隔离的工作树中同时进行开发[1]。Meta在测试中已验证了这一能力，团队曾同时为游戏构建六个功能，期间"没有冲突"[1]。Meta AI首席Alexandr Wang表示，从成本角度考量，这个解决方案"可以是一个非常好的选择"[1]。这一发布扩展了Meta的企业AI产品线——公司曾在六月推出了针对客户服务的代理[1]。
Meta has unveiled Muse Code, an artificial intelligence agent currently in beta that is designed to execute complete software engineering tasks within large code repositories [1]. According to Mark Zuckerberg, the tool can handle "planning changes, writing code, and verifying results" across extensive codebases [1]. Powered by Meta's coding model Muse Spark, Muse Code distinguishes itself through its ability to manage complex projects by spawning multiple sub-agents that work in parallel [1].
When projects reach sufficient scale, the system distributes work by creating separate sub-agents that operate simultaneously in isolated work trees, preventing conflicts between concurrent tasks [1]. During testing, Meta used Muse Code to simultaneously build six game features "without conflicts," demonstrating the agent's capacity to coordinate parallel development efforts [1]. Alexandr Wang, Meta's chief AI officer, highlighted the product's economic benefits, stating that "from a cost perspective, this can be a very good choice" [1].
This launch marks Meta's continued expansion into enterprise AI solutions, following its entry into the corporate AI market in June with the release of agents tailored for customer service applications [1].