Packages are code, in large part, and they probably represent the main value and work product of a distribution (though a lot of automation already exists for this). I would not be surprised if an LLM could manage most of the day-to-day effort of creating and maintaining a standard debian package. The other main part that is valuable, though, is the process of getting all those packages to play nicely with each other, which is a bit more political and probably will not benefit so much from LLMs, other than perhaps making some approaches or experiments lower effort.