Meta Muse Code: AI Agent Tackles Complex Software Repositories

Meta announced earlier this week the release of Muse Code, an autonomous AI agent designed to navigate and modify complex software repositories with millions of lines of code. The company said the tool moves beyond reactive autocomplete functions to perform multi-step reasoning tasks such as migrating authentication services or debugging race conditions across tangled dependencies. Muse Code uses a proprietary Repository Graph Architecture that maps data flow and function hierarchies, allowing the agent to retrieve only relevant code sections rather than scanning linearly. In internal testing, the agent resolved a flaky test suite in the Instagram backend by identifying and fixing a concurrency issue in a caching layer. Meta said Muse Code has been mandatory for all new backend development since August 1st, increasing developer velocity on infrastructure tasks by an estimated 40%. The system is built on Meta's next-generation Llama 4 architecture and includes Muse Guard, a safety layer that prevents destructive commands without human approval and logs actions for audit compliance. Meta is opening a private beta for select enterprise partners in a managed cloud environment. Pricing has not been released, but analysts expect a subscription model based on agent compute hours.