Meta launches Muse Code, its first AI coding agent to rival OpenAI and AnthropicMeta has launched its first AI coding agent, Muse Code, as the tech giant steps up its push to compete with OpenAI and Anthropic. Currently released in beta, Muse Code is a terminal-based coding agent powered by Meta’s new Muse Spark 1.2 model. Announcing the new models, the company says that it can handle complex software engineering tasks across large code repositories, including planning changes, writing code and checking whether the results work correctly. What is Meta Muse Code and how does it workUnlike a basic AI coding assistant that mainly suggests or generates pieces of code, Muse Code is claimed to be designed to work as an agent capable of carrying out longer and more complicated software engineering tasks. According to Meta, Muse Code uses a main AI agent along with multiple specialised background agents. These background agents remain active throughout a coding session and can gather information, work on next steps and decide when they need to report their findings to the main agent.Meta says this approach can help the coding agent solve difficult problems faster while requiring less intervention from developers. The company says it has also designed Muse Code to handle long-running jobs. Every model call, tool run, approval and code edit is stored in a local event log. If the agent crashes or stops unexpectedly, it can restart from exactly where it left off instead of beginning the task again.Muse Code also comes with built-in skills. The /plan command can create a plan that requires user approval, while /grill can test that plan for possible problems. Another command, /goal, directs the agent to keep working towards completing a specified objective.Muse Spark 1.2 powers Meta's new coding agentAt the centre of Muse Code is Muse Spark 1.2, Meta's latest AI model and a coding-focused update to Muse Spark 1.1. Meta says it increased the amount of training compute devoted to coding tasks while also expanding the variety of training environments used to develop the new model.The company claims Muse Spark 1.2 improves code generation, complex debugging, understanding of large codebases and end-to-end developer workflows. It has also been trained for long-running coding jobs, including generating entire repositories and working on large projects.