Meta Releases Muse Code, A New Terminal Coding Agent

Meta has released Muse Code in beta, a terminal-based coding agent designed to handle complete software engineering tasks rather than simply suggesting snippets. The tool is built to operate across large, complex codebases, taking on the planning, writing and validation stages of development with minimal human intervention. Muse Code is powered by Muse Spark 1.2, a coding-focused update to Meta's model lineup. This positions Meta more directly against rival coding agents from OpenAI, Anthropic and Google, all of which have pushed into the agentic coding space over the past year. Rather than acting as an autocomplete tool inside an editor, Muse Code operates from the terminal, giving it broader access to a project's files, build systems and test suites. The emphasis on handling full engineering tasks marks a shift in how Meta is positioning its coding tools. Instead of assisting a developer line by line, the agent is designed to take a task description, work out an approach, implement the necessary changes across multiple files, and then check its own work before handing it back. This kind of end-to-end workflow is becoming the standard test for coding agents, with the ability to validate results seen as a key differentiator from earlier generations of code-completion tools. Meta has not detailed pricing, availability outside the beta, or how Muse Code compares against competitors on standard coding benchmarks. As with other agentic coding tools launched in beta, expect further updates on capability, reliability and access as Meta gathers feedback from early users.