Meta Enters the AI Coding Race With Muse Code: Parallel Worktree Subagents Built on Muse Spark 1.2

Meta's Muse Code, launched August 5 for macOS and Linux on Muse Spark 1.2, is Meta's first dedicated coding agent — parallel isolated-worktree subagents, one-command install, pay-as-you-go or contributor-tier pricing.

Meta rolled out Muse Code beta on August 5, entering a space already occupied by Claude Code, OpenAI's Codex, and Cursor. The agent installs with a single command on macOS and Linux, connects to any git repo, and runs on Muse Spark 1.2 — a coding-optimized model variant trained specifically for the agent workload.

For large tasks, Muse Code fans out to parallel subagents working in isolated worktrees, so the main checkout is never touched mid-run. The pattern is structurally similar to Claude Code's workflow orchestration and Codex's parallel agent mode — all three converged on isolated-worktree subagents independently.

Pricing: pay-as-you-go at API rates, or a discounted contributor tier where developers consent to training data use.

Why it matters: The coding-agent market now has four serious contenders competing on the same architecture — parallel subagents, worktree isolation, agentic workflows. Your CLAUDE.md conventions, context-engineering patterns, and custom slash commands are increasingly portable skills. A competitive field accelerates feature shipping and applies downward pressure on pricing.

Sources: Meta debuts Muse Code to rival Claude Code — CNBC · Meta enters AI coding race — The Next Web · Meta Muse Code beta — Dataconomy · Meta Muse Code for macOS/Linux — 9to5Mac