Meta enters the AI coding wars with Muse Spark 1.2 and Muse Code with persistent async background agents

**Meta** released **Muse Code**, a terminal-based AI coding agent now in beta, alongside **Muse Spark 1.2**. The launch puts Meta into the AI coding-agent race with a product framed around **persistent async background agents** rather than only one-shot chat completions inside an editor.

**Muse Code** runs from the terminal and is designed to keep work moving after the first prompt. **Persistent async background agents** point to long-running sessions that can keep operating while the developer does other work, instead of blocking on each turn. **Muse Spark 1.2** ships with that release as the model side of the stack; the public details frame the pair as product plus model, not a single chat endpoint.

For engineers and builders, a terminal-first agent changes the integration surface. It can sit next to existing CLI workflows, scripts, and local repos without requiring a new IDE. Beta status means the useful test is how well agents stay coherent across multi-step tasks and how much control you keep when work continues in the background.

The move is market positioning as much as product. Coding agents are already a crowded category of terminal and IDE tools; Meta entering with a named agent product and a numbered model line signals it wants a direct seat in that fight, not only foundation-model distribution. Expect rivals to answer on agent runtime design—especially persistence, async control, and terminal UX—not only raw completion quality.

What to watch next is how the beta behaves under real multi-file work: whether background agents stay scoped and reviewable, how **Muse Spark 1.2** performs as the driver model for those sessions, and whether Meta tightens the product around agent lifecycle controls before a wider release. For teams evaluating agents, the practical bar is simple—does async background work reduce cycle time without creating opaque changes you cannot audit.