been trying Meta's new Muse Code terminal agent this weekend
its a coding agent that runs in your terminal but the interesting part is it keeps background agents alive during your session. like they dont start fresh for every task. they remember context and keep working in the background while u do other things
ran it on one of my repos for about 3 hours. it made like 400 tool calls, refactored a whole module, wrote tests, caught its own errors. the crash recovery is wild too. it logs every action so if it crashes it picks up where it left off
the model is called Muse Spark 1.2 and Meta trained it specifically for long repo scale tasks. in their own test it ran for 24 hours straight optimizing GPU kernels. 24 hours
my main concern is the same as with every agent tho. how much do u trust it to touch production code. i kept it on a feature branch and reviewed everything before merge
honestly feels like we are heading toward a world where u dont write code anymore. u review it. and that shift is happening faster than i thought
anyone else tried Muse Code yet or still sticking with Claude Code and Cursor?
Top comments (0)