Login
You're viewing the functional.cafe public feed.
  • Sep 18, 2026, 2:45 PM

    So: you started vibecoding your FOSS project? And now you don't really understand your codebase anymore *without* asking agents what's happening? And you basically outsourced all your favorite part of your passion project to managing a robot that doesn't care? And you're more burnt out than ever?

    I have to say that it's very difficult for me to feel sympathy for this. I see more people reporting on this, but it feels fairly predictable. But it also feels like it was clearly where it was going.

    💬 42🔄 563⭐ 1060

Replies

  • 💬 0🔄 1⭐ 8
  • 💬 0🔄 2⭐ 10
  • 💬 2🔄 1⭐ 1
  • 💬 1🔄 0⭐ 5
  • 💬 1🔄 0⭐ 0
  • Sep 19, 2026, 7:12 PM

    @6 @gunstick @cwebber I'm not "pro-AI". But I'm also pondering how it is different from the effect of automation in factories in early 20th century 🤔

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 3⭐ 4
  • Sep 18, 2026, 3:02 PM

    @cwebber and it all feels so commercial. i feel so different and unusual for disliking things that feel commercial and the business lingo, the "go-getter" attitude, they all make me feel gross working on personal projects. i'm surprised the bots don't bring in sprints "because they're a great way to pace and organize the project!"

    💬 2🔄 1⭐ 0
  • Sep 18, 2026, 3:17 PM

    @tyzbit @cwebber In a way, every prompt is a sprint step. You're forced to detangle what exactly was generated and finish all in one go lest you forget what that mess was about.

    💬 0🔄 1⭐ 2
  • 💬 0🔄 0⭐ 0
  • Sep 18, 2026, 9:44 PM

    @tyzbit @cwebber I already felt some social/cognitive dissonance years ago when workmates and managers kept pushing Scrum despite that:
    1. Those Fibonacci story point estimations are utterly meaningless (apart from as an exercise in just... discussing how we should implement something - but why don't we just do that then?)
    2. The idea of finishing a sprint and immediately starting the next one makes a mockery of the word "sprint"
    3. We literally never "completed" a sprint, so every single one felt like a failure and we had to explain what wrong at the retro. Nothing went wrong - we just thought things were simpler until we tried them, and we'll keep doing that forever
    4. Scrum related rituals used up like 30% of each workday and 83% of our energy and good mood.

    The embrace of slop by corporate execs and middle managers (who are perhaps desperate to "champion" every top-down initiative) shouldn't have surprised me, but it did.

    💬 1🔄 4⭐ 19
  • Sep 18, 2026, 10:52 PM

    @ganxel @cwebber YES! well articulated. the whole thing should have been a quick lesson in predicting complexity and then everyone moves onto something else but since business is a religious cult, it just won't die.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 8⭐ 20
  • Sep 18, 2026, 3:15 PM

    @cwebber “I was on the plane and the internet wasn’t really working, so I opened up the project and it has hard at first, but good and I had a great time refamiliarizing myself and hacking on it without the bot … I don’t know that I have the discipline to not use the bot though”
    (from memory, not an exact quote) my sibling-in-suffering, you don’t know that you have the discipline to do the work that you fucking enjoy instead of handing it to a fucking word extruder?! You are your own boss, too? Fucking hell, these things are so goddamn toxic.

    💬 0🔄 5⭐ 24
  • 💬 0🔄 0⭐ 1
  • 💬 1🔄 0⭐ 2
  • Sep 18, 2026, 5:11 PM

    @Nicovel0 correct. LLMs can't reason. But they will confidently give you a statistically possibly right but probably wrong explanation furthering your confusion, sending you on dead end paths, and destroying any sort of debugging skill you had in the process.

    @cwebber

    💬 1🔄 2⭐ 7
  • Sep 19, 2026, 10:14 AM

    @dacmot @cwebber look I’m not a programmer, last time I did anything of the sort was to draw a circle in Pascal. But it seems to me that asking the statistical likelihood engine to explain the statistically likely code it spat out is not a sound engineering practice.

    💬 0🔄 0⭐ 2
  • Sep 18, 2026, 3:40 PM

    @cwebber You're being far too gentle on the latest scourge of humanity. These same folks, a few of whom will _get away_ with vibe coding this or that, will go on to inflict it on the world. There's already enough shitty software. A whole generation thinks needing to periodically reboot a computer is "just how computers are." We don't need more shitty software faster. #Vibecoders are management aspirants, DHHs in the making, people that never wanted to learn how the sausage is made, and still never want to be in the kitchen where it's made. They just want an easy button with which they can "make" something declaratively, by fiat, the same way #AgentOrange claims to be a "builder" by throwing currency at actual builders.
    Make stupidity expensive again.

    💬 1🔄 3⭐ 13
  • 💬 0🔄 1⭐ 4
  • Sep 18, 2026, 3:47 PM

    @cwebber Sounds exactly like drugs. Seems to be helpful, puts most users ever deeper into trouble.

    The ones that might be able to mange it don't need it.

    💬 0🔄 2⭐ 14
  • Sep 18, 2026, 3:49 PM

    @cwebber I've seen the same thing happen to software not written by agents, the only difference is the speed.

    I think practically it takes years for one or more humans to write themselves into a corner, with agents you cut down that time to months or weeks, maybe even days.

    In both cases this is probably avoidable, or at least something that can be mitigated. Although I think there's also a camp now of people who no longer care about human understanding of things.

    💬 0🔄 1⭐ 0
  • Sep 18, 2026, 3:57 PM

    @cwebber Unfortunately, a lot of FOSS now is career checklisting.

    Just like high school students know to add volunteering to their college application -- to look like they want to help others, show leadership, and that they're exposed to different life experiences...

    CS grads know to have a GitHub account with activity -- to look like they have passion for the field, do something clever, and have that added experience.

    That started before AI, and AI homework cheating culture makes it worse.

    💬 1🔄 3⭐ 14
  • Sep 19, 2026, 11:15 AM

    @neilvandyke

    Ohh that makes sense. Only reading this now do I realise I had been harbouring a vague "but what's the point?" about the flurry of low-quality LLM-generated contribs.

    💬 0🔄 1⭐ 1
  • Sep 18, 2026, 4:00 PM

    @cwebber on one have this is awful, was completely preventable, and now the commons is riddled with sloprot including many critical projects that will take decades to rebuild and this hurts *everybody*

    on the other hand, [nelson from the simpsons voice] HAW HAW

    💬 2🔄 4⭐ 33
  • Sep 18, 2026, 4:03 PM

    @cwebber one fun things my adventures playing around with 9front the last few weeks has taught me is that i've got the skills to fucking clean house now. the slop crackers have *nothing*

    💬 0🔄 3⭐ 25
  • 💬 0🔄 0⭐ 3
  • Sep 18, 2026, 4:19 PM

    @cwebber I uh ....

    I kinda went the other direction.

    I'm taking my time engineering the shit out of the simplest piece of software ever: a contact form server. I was burned the fuck out. Now I'm having fun again :)

    I'm still baffled and hurt that the people who thought xor with hard coded "password" is "encryption" were seen as more valuable than me, but I'm enjoying this again and enjoying the progress I am making.

    I don't need AI to spam reviewers with working and fully tested code

    💬 0🔄 0⭐ 5
  • Sep 18, 2026, 4:42 PM

    @cwebber I made something for work using Blender, and a higher-up suggested using AI.

    sure, take away the most fun parts. Why don't you kick my robot dog

    💬 0🔄 0⭐ 4
  • Sep 18, 2026, 4:48 PM

    @cwebber I've had luck with getting the robot to only work on a tiny section of the code and not look anywhere else. It helps in the sense that the robot can type faster than me and runs tests automatically. I'm not sure what more folks expect from it.

    💬 1🔄 0⭐ 2
  • Sep 18, 2026, 6:56 PM

    @nisha @cwebber I have made similar experiences, but LLMs "love" emitting pointless tests, so beware of that. Running the tests does not buy you anything if the tests are complete garbage. I'd rather have no checks than wrong confidence.

    💬 1🔄 0⭐ 2
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 3
  • Sep 18, 2026, 5:46 PM

    @cwebber I'm so glad I never let an agent anywhere near any of my actual code

    I wish I never decided to ever try communicating with an LLM instead of a damn plushie about whatever issue I happen to be stuck on

    and I am so fucking glad I stopped when I did

    💬 0🔄 0⭐ 11
  • 💬 0🔄 0⭐ 1
  • Sep 18, 2026, 6:03 PM

    @cwebber tech debt is bad enough and burns you out quickly enough when you're the one doing the actual architecting, can't imagine how bad it'd be letting stochastic parrots take the wheel but i guess it sounds pretty dang bad

    💬 1🔄 3⭐ 28
  • 💬 1🔄 0⭐ 12
  • 💬 0🔄 0⭐ 1
  • Sep 18, 2026, 6:27 PM

    @cwebber tbf if they're burnt out I'd suggest the LLM doing the fun stuff for them isn't the problem. If it was the response would be boredom not burn out imo. Being burned out sounds like not knowing how to take a day off to me. I believe I see this a lot in engineers who don't know how to do anything else (this was me at one point)

    💬 0🔄 1⭐ 1
  • Sep 18, 2026, 6:51 PM

    @cwebber

    A widely used FOSS project I've been contributing to started to overly rely on LLM, especially for fixing bugs on systems they're not familiar with (e.g. FreeBSD).

    When I offered to help (without LLMs of course), they simply replied "no thanks, but you can review the (LLM-authored) code if you like".

    I don't know what to say anymore, or if I should keep contributing to or even use that project. Sure, maintaining a FOSS project is cumbersome, but it doesn't have to come to this...

    💬 1🔄 1⭐ 11
  • 💬 0🔄 0⭐ 0
  • Sep 18, 2026, 6:52 PM

    @cwebber It is a bit more nuanced. I personally dislike the programming itself. I review the code, I use agents as a "more capable" grep or for bulk edits. I use them to bootstrap small projects and monitor the process. They have their uses _in programming_.

    But software engineering is much, MUCH more than just programming. And people who are eager to outsource their brains were already dangerous before, but nobody had handed them the tools to destroy.

    💬 0🔄 0⭐ 1
  • 💬 4🔄 1⭐ 5
  • 💬 0🔄 2⭐ 11
  • Sep 18, 2026, 11:31 PM

    What if you care more about the building being insulated, than about whether the walls are full of asbestos @skavish?

    What if you care more about travelling to a distant city, than about the huge carbon output of the transport?

    What if you care more about the slinky dress for tonight's party, than about the child labour used to manufacture it and the enormous landfill encouraged by fast fashion?

    What then?

    @cwebber

    💬 1🔄 1⭐ 3
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 3
  • 💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 1
  • 💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • Sep 19, 2026, 4:06 PM

    @cwebber @skavish I know, right?

    It's interesting that advocates on one side or the other are so emphatic about other people changing their development practice to improve productivity or reduce defects.

    If one thinks that using, or not using, AI is so much better, why not keep that clever optimization to oneself and reap the benefits of a competitive labour (and software) market? Why give away the secret recipe?

    💬 1🔄 0⭐ 0
  • Sep 19, 2026, 4:38 PM

    @evan @cwebber @skavish As a person who does not like AI when I say it shouldn't be used I'm not trying to give sloperators productivity tips, I'm trying to 1) keep software I depend on from degrading in quality and 2) stop the foss ecosystem from being completely beholden to a small number of very powerful (and highly immoral) actors

    💬 0🔄 1⭐ 3
  • 💬 0🔄 0⭐ 0
  • Sep 18, 2026, 8:15 PM

    @cwebber The worst part of this is that these projects will inevitably end up eating into the tiny resource pools (contributors, donators, translators, etc) of the greater Open Source world that has been struggling to stay alive forever.

    💬 0🔄 3⭐ 10
  • Sep 18, 2026, 8:35 PM

    @cwebber related predictable situation - you became dependent on this entire workflow and it stopped being subsidized and you're now stranded because this workflow isn't [yet?] backwards-compatible with coding by hand.

    💬 0🔄 0⭐ 3