Meta announced the launch of Muse Code on Wednesday, an agentic coding tool for developers of artificial intelligence products, as the company competes with other major AI labs for customers and revenue. Meta described Muse Code as capable of performing "complex software engineering tasks" involving large amounts of data, including planning, writing and validating code. Agentic tools are software applications programmed to handle tasks without human supervision, though a person typically still makes final decisions or supervises task execution. Muse Code launched in beta, or preview mode, and is a terminal-based agent available for macOS and Linux, allowing developers to coordinate AI assistants from a single command line. It can also fan out to sub-agents working in parallel on separate parts of a project. Meta also released an updated version of its flagship model, known as Spark. The update, called Spark 1.2, is coding-focused and available through Muse Code and Meta's separate developer application, the Meta Model API. Developers can access Muse Code through a standard pay-as-you-go pricing structure or a discounted "contributor tier" priced at $0.30 per million tokens, in exchange for allowing Meta to use submitted code to train future models. Meta has invested heavily in its AI division, now known as Meta Superintelligence Labs, as it works to keep pace with OpenAI and Anthropic. Amazon, Microsoft, Alphabet and Meta are collectively on track to invest around $700 billion in AI data centers, chips and computing infrastructure this year. In February, OpenAI