Post #3615657

2026-07-04 13:51 UTC

@vampiress@eigenmagic.net From my experiments with it: it can* be helpful in 3 scenarios. 1. Really small changes (like tracking down a bug) in existing code 2. Software that you essentially don’t care much about (like a one-off data converter or an internal app just for you, that sort of thing) 3. Truly trivial applications *often it’s either not at all helpful or it just takes the same time as hand-written code

Replies (2)

@eswag@dju.social 2026-07-06 05:24

@thomasfuchs@hachyderm.io @vampiress@eigenmagic.net It's helped me a little with code archeology. Given a codebase that was insufficiently described, with fading knowledge from its authors, an LLM did the donkey work of git blaming, reading commit messages, following documentation links & looking at GitHub PRs, to come up with a guess at *why* the logic does what it does, complete with checkable links. The most useful part of it was instances where it claimed it couldn't find a reason for the behaviour. I'm ambivalent.

Open ##3640068
@quincy@chaos.social 2026-07-07 03:48

@thomasfuchs@hachyderm.io Point 2 is interesting: not even going into the potential for skill loss, I find myself fundamentally disagreeing with its premise. Why would one "not care" about a one-off data converter? About an app that's just for me? @vampiress@eigenmagic.net

Open ##3643879