Another reason I try my darndest to avoid LLM-built "FOSS" tools: the humans who publish them often enough aren't prepared to support them.
When that neat utility that Claude burped out is deprecated by some breaking change, are you going to have someone to reach out to? Are you going to understand it? Or will the "author" have moved on toward some other thing that nabbed their attention?
No matter how "impressive" these LLM services get, no matter how powerful their capabilities, there's no accounting for the human drive toward "getting bored and doing something else".
A decade ago, if I was slinging a few thousand lines of code, it was obvious that I, a living human being with an attention span of some description, had to have worked at it. I could potentially be trusted to continue working at it. Obviously, I cared enough to write this much code.
That's not the case anymore. If I throw code at a problem, you don't know that I care enough to support it in a week, let alone a year or more.
That alone feels like enough of a reason to distrust LLM-generated output in the FOSS world. To say nothing of all of the other reasons.
@veronica
Even worse is #vibeCoding. It works in one direction only. Don't expect it to 'magically' maintain itself. A pro tip to programmers who are called in to debug / fix broken generated code: 'charge by the hour', do not offer 'fixed price'.
Aug 12, 2026, 9:56 PM@veronica I mean, that’s already a problem with non-LLM-developed FOSS tools a lot of the time. Claude definitely exacerbates this : (
@veronica I agree with you, although I guess the retort might be "Claude will support/fix it for you"
@aburka For thousands of dollars a year, potentially! I don't want to lease my codebases! :P
Aug 12, 2026, 10:30 PM@veronica
For me it's just the … insulting, hostile vibe of most of those LLM-based projects.
A lot of them don't appear to have been developed in good faith to fulfil one of the person's needs or interests, or because they enjoy software development.
Instead it often looks like they were created solely for ulterior motives like grifting money, fame or deceiving others. Often it even seems like the sole intention is actually to upset people on purpose, like "haha, look how mad this makes the anti-AI folks! look how much smarter and greater I am than you!!"
The program itself is then completely meaningless and irrelevant, just a means to an ulterior end.
It all has a similar vibe to those neo-nazi/alt-right "forks" of popular FOSS projects that keep popping up occasionally, like Redot (Godot), PolyMC (MultiMC), XLibre (X11), Sneed-whatever it was (Audacity).
They often don't make any technical progress at all; they just exist specifically to pretend that there is a serious alt-right software project that has some kind of authority, and to thereby upset the 'enemy'. They often go under in just a few months because they end up realising they aren't actually interested in programming, they just wanted to participate in a culture war.
@veronica even worse than deprecated: a recent vibe-coded feature by someone else at work included the AI using an older version of a dependency that has severe security vulnerabilities, but the major version bump is so breaking and complicated, that it's now very difficult to upgrade the dependency even just a couple of weeks later. No one else but me is even keeping up with dependencies and security vulnerabilities, there's too much other stuff to vibe code!
@brettcodes @veronica Well, why wouldn't it happen that way, you haven't contributed the fix to the next version to the slop machines yet.
Get on that, please.
No
Not ever.
Never.
@veronica my favorite word describe the act of generating anything with an LLM is "extrude"
@veronica this is exactly one of my observation that I wrote on my blog recently: https://enblog.eischmann.cz/2026/07/23/how-ai-is-changing-open-source/