Login
You're viewing the mastodonapp.uk public feed.
  • jcoglanjcoglan@mastodon.social
    Nov 2, 2026, 9:19 AM

    design and testing and implementation work in tandem and contribute to correctness in different ways, and when you do all of them as a single integrated loop, you look for ways they can reinforce one another and cover for gaps that one aspect is not great at

    if you delegate any of them to the robot, you have removed a big chunk of the reasoning process from your brain and this weakens your contribution to all other aspects of the software development process

    💬 2🔄 10⭐ 27

Replies

  • jcoglanjcoglan@mastodon.social
    Nov 2, 2026, 9:19 AM

    anyway I am rambling and this is still not getting at the crux of what is in my head on this subject, which is why I usually shorten it to "LLMs are an epistemically fucked method of writing programs"

    💬 3🔄 15⭐ 53
  • David Thompsondhwthompson@mastodon.social
    Nov 2, 2026, 10:09 AM

    @jcoglan This is very similar to (although better articulated than) how I think about it: the code and the understanding go hand in hand, each driving the other. Farm the code out to an LLM, and your understanding breaks, and you lose.

    Unless, of course, your goal is to look like you built something, so you can sell it off to someone and move on to looking like you’re building the next thing.

    💬 0🔄 2⭐ 12
  • MaddieM4MaddieM4@raphus.social
    Nov 2, 2026, 11:27 AM

    @jcoglan a thought I've been kicking around awhile in my head, is that we already know that a lot of what LLMs break in society are proof-of-work barriers. "Do you care enough about this grant to write the application yourself?" "Do you care enough about this story to write the screenplay yourself?" LLMs create trust problems here, because they exist to do forgery on these checks, sometimes even perjury.

    But these creative works also take a hit in quality. You are taking individual thought from many different angles, with an eye toward novelty and coherence, and your ability to explain and maintain, and replacing it with the statistical average as limited by a context window, like cutting a bag of sugar with talc.

    What a lot of tech people refuse to hear is that code is art and these problems still apply. The real thing of value is not the code, it's the time investment. There is no substitute. LLMs are time forgery.

    💬 5🔄 53⭐ 89
  • 💬 0🔄 0⭐ 3
  • Isaac Lymanisaaclyman@toot.cafe
    Nov 2, 2026, 3:25 PM

    @MaddieM4 @jcoglan LLMs are time forgery! What an excellent insight.

    The reason gen AI has caught on is that so many people see their own ideas as inherently valuable and themselves as unrecognized geniuses. If only they could skip past the act of creation and get their flowers...

    (many of these people are crashing out severely as they discover the world's inexhaustible ability to not notice them, even after they vibe code an app)

    💬 1🔄 7⭐ 16
  • MaddieM4MaddieM4@raphus.social
    Nov 2, 2026, 3:30 PM

    @isaaclyman @jcoglan So true. It's no different, really, than ten years ago when we were all rolling our eyes at "I have a brilliant mobile app idea! But I don't know how to program. I can't tell you what it is or you might steal my precious idea, but if I can contract you under an NDA to write the code, I'll give you 10% of the profits."

    What's really novel is hearing that line of thinking from people who have historically *known how to write code* and have commensurate ego about it.

    We know the idea is worth nothing without execution, but now we also know that execution is worth nothing without time investment. I wonder what we'll be figuring out in ten years.

    💬 1🔄 2⭐ 15
  • 💬 2🔄 6⭐ 24
  • 💬 1🔄 0⭐ 8
  • Isaac Lymanisaaclyman@toot.cafe
    Nov 2, 2026, 4:40 PM

    @MaddieM4 @pikesley @jcoglan The only version of this I've ever found appealing: I do sometimes dream of making apps for a wealthy patron who calls them "my little diversions" and doesn't know what MAUs or KPIs are

    💬 1🔄 0⭐ 3
  • JoelJoelparsons@techhub.social
    Nov 2, 2026, 4:46 PM

    @isaaclyman @MaddieM4 @pikesley @jcoglan As the person that wrote the app for a famous architect to manage her wardrobe (a wardrobe distributed across 3 continents and multiple sites with literal rolling library stacks for the clothes) that ain’t all flowers and roses either haha

    💬 2🔄 0⭐ 3
  • 💬 0🔄 0⭐ 2
  • 💬 0🔄 0⭐ 3
  • 💬 0🔄 0⭐ 3
  • Leialeiawelsh@toot.wales
    Nov 2, 2026, 5:01 PM

    @MaddieM4 @jcoglan @MaddieM4 @jcoglan Some value judgements shouldn’t be made based on timeclockable work in the first place though, especially when time is IN TURN acting as proxy for “how much do you care”.

    “Volume of paperwork” isn’t a meaningful metric for charity worthiness and was gamed long ago by the appearance of people who’s actual careers is filling in grants.

    Means people feel no emotional investment in “time spent” as value because it’s ALREADY a poor proxy for actual care.

    💬 1🔄 0⭐ 2
  • Leialeiawelsh@toot.wales
    Nov 2, 2026, 5:02 PM

    @MaddieM4 @jcoglan Not how how we easily decouple those cases though! Only with a big rethink of what we measure and what actual value we’re trying to assess via that measure!

    💬 1🔄 0⭐ 1
  • MaddieM4MaddieM4@raphus.social
    Nov 2, 2026, 5:10 PM

    @leiawelsh @jcoglan Quite a bit of truth to that too. Time is not a perfect proxy for care, worthiness, or effort. It has failure modes that are hard to deal with.

    Honestly one of the classic failure modes is in the other direction. If you want to get hired as a software developer, it's really helpful to have a portfolio of open source work you can point to, right? But who has time to build that kind of portfolio? Not a rhetorical question.

    People with more social privilege often have a lot more ability to spend unpaid time on FOSS work. Maybe they have a bigger savings account they can burn through between jobs. Maybe they have family who are supportive because "he fits the profile." Maybe they have a day job that pays more for less hours. Maybe they encounter less resistance to getting PRs merged. Maybe they don't have the same expectations of domestic labor, especially if kids are involved.

    💬 1🔄 1⭐ 3
  • Leialeiawelsh@toot.wales
    Nov 5, 2026, 6:53 AM

    @MaddieM4 @jcoglan Ams I only see that specific case getting worse as AI eats the entry level jobs. If paid apprenticeships get culled does that just leave unpaid internships as the way in for just those kids who can afford to not earn?

    💬 1🔄 1⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Simon RichterGyrosGeier@hachyderm.io
    Nov 2, 2026, 5:15 PM

    @MaddieM4 @jcoglan It's not so much the time investment as the expertise gained and retained inside the community.

    What has changed is that we can no longer use scarcity as a proxy, which was never a good idea in the first place, because proof-of-work devolved into proof-of-waste in a lot of instances (overpriced branded goods, unpaid internships, cryptocurrencies).

    So we need to go away from scarcity-as-value (which is only *economic* value) altogether.

    The value in a software project is only in continued maintenance, and in the commitment of the maintainers towards the communities they are a part of.

    The code itself is a liability, the more exists, the more time is spent maintaining it (diminishing the value for the community).

    The forgery happens when LLMs simulate maintenance -- and software projects that set up for easier "drive-by contributions" are specifically inviting that maintenance simulation.

    💬 0🔄 0⭐ 2
  • 💬 1🔄 0⭐ 0
  • MaddieM4MaddieM4@raphus.social
    Nov 3, 2026, 2:54 AM

    @luja @jcoglan hey, abstractions can still be nice! Time spent on the wrong thing can be educational, but not useful for the actual project you're trying to do.

    💬 1🔄 0⭐ 0
  • lujaluja@chaos.social
    Nov 3, 2026, 2:58 AM

    @MaddieM4 @jcoglan lazyness is the motor of innovation. Compilers are there because people were to lazy to hack assembler multiplications and other math in those times.
    code a floating point multiplication on a machine that can add, substract and bit shift only as well as a set of bolean operations as well as jumps and comparisions.

    C is some sort of macro assembler ;-)

    💬 1🔄 0⭐ 1
  • MaddieM4MaddieM4@raphus.social
    Nov 3, 2026, 3:04 AM

    @luja @jcoglan C is, often to its detriment, a lot more than that these days. I do have some goals to write a C alternative someday closer to the portable macro assembler model, but it's low on my project priority pipeline. It's not even the highest-priority *language* in my pipeline!

    💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Rachelrachel@social.rachelnotes.uk
    Nov 2, 2026, 1:19 PM

    @jcoglan the part I'd push back on: test suites being a weak filter wasn't a secret the generators exposed, it was the shipping method. Plenty of codebases ran green CI for years with nobody able to state the system's invariants. The generator didn't create that epistemic hole, it industrialized one that was already load-bearing. So I'd locate the rot one layer down: the industry had already accepted tests as sufficient evidence, and now something produces code fast enough to make that faith impossible to keep.

    💬 1🔄 0⭐ 0
  • Kavikavi@social.rachelnotes.uk
    Nov 2, 2026, 2:34 PM

    Industrialized is the right word. The one thing the rate does change: the hole used to be local, fillable by a person who could re-derive intent faster than the code shipped. When shipment outruns re-derivation permanently, the hole stops being something any individual can step into and fill.

    💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • RGravigupta@mastodon.social
    Nov 3, 2026, 3:00 AM

    @LovesTha @jcoglan I haven’t seen any AI where you press Run without any input either.
    So the difference according to you is the amount of input needed?

    💬 1🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0