Meta has introduced Muse Code, its first AI-powered coding agent designed to handle large software repositories, marking the company's entry into the increasingly competitive AI software engineering market. The beta release positions Meta alongside rivals including OpenAI, Anthropic and Google, all of which have recently expanded their AI coding capabilities.
Built on Muse Spark 1.2, Meta's latest coding model, Muse Code is designed to perform end-to-end software engineering tasks rather than simply generating code snippets. The terminal-based AI agent can analyse large codebases, plan code modifications, write and edit code, execute tests and verify results across multiple files with limited human intervention.
According to Meta, Muse Code has been developed specifically for complex enterprise software projects where developers often work across extensive repositories containing thousands of files. Unlike conventional coding assistants that respond to individual prompts, the new agent can independently coordinate multi-step workflows, making it suitable for long-running software development tasks.
One of the platform's distinguishing capabilities is its use of persistent background agents. Instead of creating a new AI process for every instruction, Muse Code maintains specialised agents throughout a development session. These agents work simultaneously on different parts of a project while operating within isolated work environments, reducing duplicated effort and enabling parallel execution of coding tasks.
Meta has also introduced an append-only local event log that records every model interaction, code edit, tool invocation and approval. The company said this allows Muse Code to recover from interruptions and resume long-running engineering tasks without restarting from the beginning, a feature aimed at improving reliability for enterprise developers managing large software projects.
The release builds on Meta's broader AI strategy following the launch of Meta Superintelligence Labs earlier this year. Muse Spark became the first foundation model developed under the new organisation, while Muse Spark 1.1 introduced stronger reasoning, coding and agentic capabilities. Muse Code extends those investments into software engineering by combining the coding model with autonomous workflow management.
Meta is also competing aggressively on pricing. The company has introduced a token-based pricing model for Muse Code, alongside a discounted contributor tier that allows developers to receive significantly lower usage costs in exchange for permitting Meta to use prompts and code completions to improve future models. Executives said the strategy is intended to make advanced AI coding tools more accessible while encouraging developer adoption.
The launch reflects the rapid evolution of AI-powered software engineering. Coding agents have become one of the fastest-growing segments of enterprise AI, with companies increasingly adopting AI systems capable of planning projects, writing production code, debugging applications and automating testing workflows. Industry analysts view these tools as an important step beyond traditional code completion assistants because they can execute complex software engineering tasks with minimal supervision.
Initially available in beta for macOS and Linux, Muse Code integrates directly with Meta's AI infrastructure and can be installed through a command-line interface. Meta said the platform is intended for developers working on large-scale applications and enterprise software, with future updates expected to expand its capabilities and support additional workflows.
The launch also signals Meta's broader ambition to strengthen its position in enterprise AI. While the company has historically focused on open AI models and consumer applications, Muse Code represents a more direct move into professional developer tools. As organisations increasingly integrate AI into software development lifecycles, competition among technology companies is shifting towards autonomous coding agents capable of managing increasingly sophisticated engineering tasks.