Meta Unveils Muse Code Terminal Agent to Replace Junior Devs
<p>Just when developers thought they could sit back and write code in peace, Meta drops a terminal agent that runs sub-agents in secret git branches. Is this the end of human debugging or just another expensive telemetry vacuum?</p>
The engineering team at Meta Superintelligence Labs, led by former Scale AI chief executive Alexandr Wang, has launched Muse Code straight into public beta. Operating directly inside the command terminal, the system handles background planning, automated execution, and self-testing without needing developer supervision.
Powering this setup is the Muse Spark 1.2 model, tailored explicitly for repository navigation, heavy debugging, and prolonged task execution. To keep developers from pulling their hair out over broken main branches, the agent spins up isolated background sub-agents across separate repository branches, letting multiple solutions fight for survival simultaneously.
The architecture logs every tool invocation and file modification prior to execution, maintaining persistent state even after unexpected system crashes or terminal reboots. Benchmark tests reveal that Muse Code paired with Muse Spark 1.2 scored roughly 59% on DeepSWE 1.1 and delivered strong numbers on Terminal-Bench 2.1, though Anthropic's Claude Opus 5 still holds the performance crown.
Access comes via API at standard rates of $1.25 per million input tokens and $4.25 per million output tokens. Developers willing to feed their telemetry data back to Meta get a massive discount down to $0.10 input and $0.20 output tokens per million.
Trading private codebase telemetry for cheap tokens feels like signing away corporate secrets just to save a few dollars on bug fixes. Corporate boardrooms will inevitably jump on the discount while engineers slowly realize their terminal history is training their digital replacements.
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.