Login
You're viewing the functional.cafe public feed.

Replies

  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 1
  • Nov 17, 2026, 3:54 PM

    @cmconseils there is a weird sort of imposter syndrome that comes from this too. Like wait, if my boss is satisfied with crap, then maybe I sucked all along? It's like when someone leaves you for someone else who's ugly and it's like oh shit, am I also ugly?

    💬 2🔄 0⭐ 5
  • Nov 17, 2026, 4:27 PM

    @thomasjwebb @cmconseils there's so much imposter syndrome with LLMs, yes! Because also: it creates code you can't understand (so can I even read code?) yet somehow works (more or less), it adds all kinds of tests you would never think to add but might make sense (so am I bad at creating TTD?) and it is actually quite good at figuring out why something you wrote doesn't work but not at fixing its own work, so surely your own errors are trivial and the LLM's errors are profound??

    💬 2🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Nov 17, 2026, 5:11 PM

    @meuwese

    My meat-proxy coworker practically worships Claude and constantly emails me its output.

    99% is incoherent dogs**t.

    The most likely reason you can't understand its code is that its code is also incoherent dogs**t that it brute-forced into a state where it just barely works.

    Plenty of humans have written incoherent dogs**t code before. You'd struggle to understand it, too, but that doesn't mean it's good. Dogs**t that's incomprehensible is still dogs**t.

    @thomasjwebb @cmconseils

    💬 0🔄 0⭐ 4
  • Nov 17, 2026, 5:23 PM

    @meuwese

    This applies to tests, too. Just because it generates all kinds of tests doesn't mean those tests are actually useful and correct.

    I haven't looked at any LLM-written test suites thus far, but I have looked at my coworker's emails, and Claude loves to generate loads and loads of unnecessary noise, so I'd assume it will do the same when writing tests.

    Basically, everything it writes, including impressive-looking test suites, is crap until proven otherwise.

    @thomasjwebb @cmconseils

    💬 1🔄 0⭐ 2
  • Nov 17, 2026, 5:28 PM

    @argv_minus_one @meuwese @cmconseils I would say from experience that these tools are better at code than prose (at least when working within a repo). But the problem is the code has prose in it, comments. And I just saw that a dev much smarter than I independently came to the same conclusion as me that AI-generated comments shouldn't be allowed (sunshowers.io/posts/nextest-in). My experience using AI at my day job makes me pretty confident in my decision to stop allowing AI in my oss projects.

    💬 1🔄 0⭐ 0
  • Nov 17, 2026, 5:31 PM

    @argv_minus_one @meuwese @cmconseils I think if the OSS community tries not to jump on this bandwagon just because enterprise is hyping it, we can come up with better ways to "vibe code." I was thinking a little about this a couple of days ago that I'm on the verge of making something that automatically makes UIs for openapi endpoints. And I was thinking holy shit, I'm making a software factory! And I don't have to pay Anthropic for it to work.

    💬 0🔄 0⭐ 0
  • Nov 17, 2026, 5:42 PM

    @thomasjwebb

    “I’m not going to give up my values in service of getting something like this out quicker,” said the slop bro. 🙄

    I took a look at one of the cited commits github.com/nextest-rs/mukti/bl and holy smokes that's a lot of unnecessary code. Just generate a non-expiring snake-oil certificate with openssl, store it in the source tree, and be done.

    And why the hell is it generating static PowerShell code programmatically??? 🤦‍♂️ github.com/nextest-rs/mukti/bl

    @meuwese @cmconseils

    💬 0🔄 0⭐ 1
  • Nov 17, 2026, 5:44 PM

    @thomasjwebb

    So yeah, this is actually an excellent demonstration of why reaching for a bot is a horrible idea: it will generate a landfill's worth of garbage code, and bot users tend to do this blindly, without reviewing it and understanding it and realizing that it's garbage.

    That or they're using the bot to compensate for a skill issue instead of actually solving the skill issue. Most folks don't know, for example, how to make a certificate that never expires.

    @meuwese @cmconseils

    💬 0🔄 0⭐ 0
  • Nov 17, 2026, 5:30 PM

    @meuwese

    As for it being unable to fix its own work, that's because it's not capable of logic or learning.

    When you have a bug and you fix it yourself, you're reasoning about how your code does, why it does that, and what you need to change to make it work. And once you find the mistake, you learn not to make that mistake again.

    Bots can't do that. They can only plagiarize and make random changes until the code accidentally works. Failing that, they're helpless.

    @thomasjwebb @cmconseils

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Nov 17, 2026, 5:29 PM

    @cmconseils I don't believe that. If it's true then it's probably at least partially my fault. I don't think it is though. I think they can tell. I just don't think they care. My efforts fighting them on shit instead of just building it the broken way did NOT go unnoticed. In the end, it was in fact the broken way that I did it because I was too chicken shit to quit, but still... They're happy not dealing with that anymore and can now get to the important bit: making bank.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Nov 17, 2026, 6:32 PM

    @cmconseils it really pisses me off how management will nitpick literally every single thing you do and hold you to an impossibly high standard, but is happy to overlook repeated and egregious errors so long as they were output by an LLM.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Nov 18, 2026, 10:18 AM

    @cmconseils What the alt text doesn't say is that the first head-with-hands is sharp and the second one is fuzzy. (If the boss can't tell the difference, they might need new glasses!)

    💬 0🔄 0⭐ 0