Meta has introduced the beta version of its Muse Code AI agent, capable of autonomously planning changes, writing code, and testing results in large software repositories, according to infohub.kz.
The new tool, developed by Meta's Superintelligence Labs under the leadership of Alexander Wang, is powered by the Muse Spark 1.2 language model. Unlike traditional code completion assistants, Muse Code can autonomously perform the full engineering cycle: planning architectural changes, writing programs, and validating results.
When working with large-scale projects, the tool automatically distributes tasks among sub-agents in isolated working environments, eliminating the risk of damaging the main code. With the release of Muse Code, Meta aims to directly compete with autonomous solutions like Claude Code from Anthropic and Codex from OpenAI.
Developers are offered attractive pricing, including special reduced-cost access in exchange for providing data for algorithm training.