Muse Code goes after the large repo with persistent background agents
Meta's terminal coding agent entered beta claiming complete software engineering tasks across large repositories — planning changes, writing code, validating results — by launching its own sub-agents to work in parallel. It was developed and trained alongside Muse Spark 1.2.
The large-repo framing is the competitive claim. Single-file and single-module assistance is commoditised; the remaining hard problem is a change that touches twenty files across a codebase nobody currently holds in their head. Muse Code's answer is fan-out — the harness spawns parallel agents rather than working one edit at a time.
Persistent async background agents is the other half. An agent that survives the session and keeps working changes the interaction model from request-response to delegation, and brings with it the supervision problem that every long-running autonomous process has ever had.
Trained alongside the model is the part competitors will have to answer. Meta argues the coupling itself improves coding performance, which if true makes standalone harnesses structurally disadvantaged against vertically integrated ones.
TechCrunch — Meta launches Muse Code, an AI agent for large code bases → · VentureBeat — Meta enters the AI coding wars with Muse Spark 1.2 and Muse Code →