Login
You're viewing the functional.cafe public feed.
  • Nov 8, 2026, 11:21 AM

    Observation on AI: A growing number of people in software development seem to be on the mental fast track from "human in the loop" to "spec conformance suite in the loop". Take for instance this bit I just stumbled over (never mind the origin, not about fingerpointing):

    "(...) That’s certainly understandable, especially if you’re still reading, reviewing, and debugging the code yourself. But if you’ve gone largely hands-free, then what attachment do you really have to the choices you made back when you were typing code line-by-line like an animal?"

    While this does not literally endorse taking the human out of the loop, the way it's written ("if you're still reading (...) the code yourself", "typing code line-by-line like an animal") I can't help but wonder if the author of these lines hasn't already internalized the idea that they are done with the "animal era of coding" (to paraphrase the author's interesting choice of word there).

    My two cents on this: What you unleash on your own machine is your business and problem. As a software developer you have a responsibility to others though. If you don't understand the difference between unintentionally shipping bad code (mistakes) and deliberately distributing entirely unread code (neglect) you shouldn't work as a developer (or any kind of engineer, for that matter).

    💬 2🔄 6⭐ 0

Replies

  • Nov 8, 2026, 11:54 AM

    @freebliss The irony of this is that, when I had a psychotic break earlier this year, I was talking about cubes and animals as kinda logic vs emotion (but not really) and posited that humans using AI were talking to cubes in animal language, while we were before speaking in cube or cube-animal language, as cube-animals.

    It's funny to see that language applied in the exact reverse of how I see it. It is much more animal to speak to computers in natural language imo.

    💬 0🔄 0⭐ 0
  • Nov 8, 2026, 12:08 PM

    @freebliss i'm on that tipping point, but i disagree with the "bad code" premise, my current pondering is on human review really being an effective way to catch issues, especially with the kind of code AI is writing (it looks more than fine, and it's very competent at the micro level, at least with the gidelines and tooling we use), and if there are better ways, more fit to the current situation (fuzzing, property testing, others?) how do we go forward not just reaping short term speed benefits at the expense of safety and correctness, but benefit in these areas as well.

    I do believe it's possible, but i want to be cautious and intent about it, and i don't think i'm alone in that.

    (comments to the tune of it being impossible and a fools' errand, not to mention a fundamental harm, without serious back up, will be ignored, along with insults)

    💬 2🔄 0⭐ 0
  • Nov 8, 2026, 1:40 PM

    @tshirtman If you're on that tipping point (no judgement) I'd simply advice to not go over it – problem solved. :)

    As for the other comments, I've really just brought up a simple moral code: Act in good faith, make mistakes sometimes – all good. Defer accountability to 100% automated systems – bad idea. I hope we can agree on this.

    💬 2🔄 0⭐ 0
  • Nov 8, 2026, 1:48 PM
    @freebliss @tshirtman Seems in addition, it is also betting on human review not being necessary ever again (which is another dimension to it no longer now).

    Like, that seems to me to be an unreasonable risk, but maybe that's not much different than having your software dev needs subcontracted.
    💬 1🔄 0⭐ 0
  • Nov 8, 2026, 2:51 PM

    @samae yeah I'd say subcontracting is very often maliciously used as an accountability sink already :) so maybe not so different but doesn't help either haha.

    💬 0🔄 0⭐ 0
  • Nov 8, 2026, 1:50 PM

    @freebliss that's the thing though, i don't think accountability goes away, it didn't when we started submitting generated code, with or without disclaimer, we still own and should be able to explain, if we don't look at the code anymore, we still need to know why we trust it, and take the responsibility seriously. But i don't want to pretend that we had it figured out before llms, we did ship buggy code we thought we understood, and had these "how did it ever work" moments, so the bar is not being flawlessness, it's doing better, and it irks me when people blame any and all bugs on GenAI now, as if we only just started to make bugs.

    So i won't jump over it blind, but i'm certainly peering, GenAI-based error detection systems i experimented with do find heaps of issues worth adressing in real-world systems built before them, so if quality is the aim, the tech can certainly help with that.

    💬 1🔄 0⭐ 0
  • Nov 8, 2026, 2:48 PM

    @tshirtman Ok you know what let's just defer the argument: Send me an update about how it's going with AI accountability when the next model "goes rogue" and attacks a hospital, social security provider, or whatever. That worked flawlessly so far (Altman or someone is already on trial for criminal neglect, no?), so I'm sure you'll be able to perfectly illustrate your point the next time around. I'm looking forward to it.

    Until then let's please leave it there, thank you.

    💬 0🔄 0⭐ 0
  • Nov 8, 2026, 4:09 PM

    @tshirtman @freebliss words like safety and correct don’t just encompass code, software development, and data. They have larger implications.
    Can one use AI safely at all? You have to consider the environment, economics, the effect data centers have locally (heat, noise, water usage), etc… before you can call it safe to use.

    My point is, it might write great code, but what are the other impacts?

    💬 1🔄 0⭐ 1
  • Nov 8, 2026, 4:35 PM

    @automattack @freebliss these are important questions, about anything we do, and i think we all make choices, we know not to be perfect. AI Data centers are impactful, and i hope we'll end up not needing too many of them, using more local models as they get better, but for now i consider that they are less of a problem, than many things we have normalized long ago, like animal agriculture, and fossile based energy production, a renewable based grid, alleviates some, not all, of the harm, but we have to favor that anyway.

    The question though, as i understood it, was about software practice, and the outcomes on software quality, and it seems than rather than defending the initial claim, we are now turning to different ones, that are worth entire discussions, and as you can see i can also broaden the discussion / what about it, which is unlikely to help us conclude about that.

    💬 2🔄 0⭐ 0
  • Nov 8, 2026, 5:26 PM

    @tshirtman @automattack feel free to discuss everything else that's morally bankrupt about AI right now (plenty to go around :)) but please remove me from the thread if you do. thanks!

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0