Hi,
On Fri, 20 Feb 2026 at 15:20, Ludovic Courtès <[email protected]> wrote:
> Could our tooling (‘guix refresh’ in this case) be compelling enough
> that one doesn’t feel the need to resort to LLMs for this kind of tasks?
Well, this could be try but I’m not sure it could “compete”.
For sure, many things could be improved and various labor tasks could be
smoothed, even maybe there is some low hanging fruits. :-)
However, it’s hard to compete with a virtual pair programmer that helps
you in various tasks, where “labor tasks” are ones among many others.
In front of a task where X doesn’t know how to, the typical workflow is
always a trial and error, and what evolves over time is the mirror
(virtual pair programmer) that helps X to understand the error then
adapt and run another trial:
• in the 90s, the Info manuals
• in the 00s, the forums
• in the 10s, the search engines
• now, it’s about LLMs
Teaching ’guix refresh’ for a subset of ’trial and error’ cannot hurt,
for sure! As for example the hints of ’guix build’ about
’define-module’ or about missing this or that module.
That said, once X has some habits for looping over the ’trial and
error’, then whatever the quality of the tools in this other specific
context, X still uses the same habits. Doomed.
Arf, the now too common situation:
« We cannot do anything, but we cannot do nothing. »
Cheers,
simon
