Yes! I experimented with it on my EXWM setup. After some back and forth via Aider, it made me a module to help me control and monitor tasks on my timewarrior setup, giving me a pomodoro-like indicator on my taskbar (changing colors as it approaches one hour).

It was a cool experience! I had it evaluate the code via commands to the emacs daemon, without reloading EXWM (ballsy, but I was prepared for failure).

EXWM is extremely flexible, but there is a high barrier of entry to using and customizing it. Having an LLM embedded to a live-evaluate desktop environment makes the interface more approachable without reducing its flexibility as much.

It also allows you to create explicit controls that map to the user’s muscle memory and sub-symbolic sensing of the environment, while staying out of the way during normal usage — a different paradigm than embedding an agent as an interface in its own right to control the environment (via speech or text).

Since open source software is readily modifiable, maybe soon it will unironically be the year of the linux desktop.