On Tue, May 5, 2026 at 9:44 PM Nguyễn Gia Phong <[email protected]> wrote: [...] > On 2026-05-05 at 15:04-04:00, Greg Hogan wrote: > > I would be very interested to see an LLM attempt to > > update Guix packages. Not simple leaf packages but core tools or > > libraries with hundreds or thousands of dependent packages, many > > requiring version updates or pinning or patches from upstream sources. > > The Guix package set is more assembled than built. > > Since each partial world rebuild costs days to a week of CI time, > I don't think that's an option for us. Socially we also need > someone to take responsibility in using the shared resources fairly. > Due to the costly QA, patches updating core-ish packages > are left hanging for months or even years.
The benefit would be that the agent could update many core packages at the same time given that computers have more memory and never need to take a break. And you would build a system that remembers past failures rather than our current system of discarding partially successful human effort. There is no inherent reason for our current system of random contributions rather than, as Hugo noted, having the humans focused on the challenging 1% (though many if not most of our patches come from upstream sources which would be discoverable by an agent, so this might be < 1% or << 1%).
