AI BriefingMeta Launches Muse Code
Key point
Meta has released a terminal coding agent with session-based sub-agents and a dedicated model.
Details
Meta has released Muse Code beta, a terminal-based coding agent, along with the coding-specialized model Muse Spark 1.2.
Muse Code targets large repositories, performing change planning, code writing, and result verification while orchestrating multiple sub-agents. Key features include:
- Background sub-agents that persist throughout the session rather than being newly created for each task
- A runtime that logs all events in an append-only manner, allowing resumption from the point of interruption after a crash
- Built-in skills that independently review and adversarially test plans after they are formulated
- An agent architecture that operates workers and reviewers in parallel and as persistent processes
Meta showcased use cases for Muse Code and Muse Spark 1.2, including a WebGL-based photon orbit simulation, an isometric dungeon game, and a rain defense game. This release is a follow-up in the Muse Spark series, aimed at improving long-running agent workflows and coding performance.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.