• xep@discuss.online
      link
      fedilink
      arrow-up
      11
      ·
      edit-2
      4 days ago

      In my experience it doesn’t even take a few features, a medium-size non-trivial script by Claude would not pass review by me without human cleanup.

      • Phegan@lemmy.world
        link
        fedilink
        arrow-up
        16
        arrow-down
        1
        ·
        5 days ago

        It won’t be cheaper for long. It creates a ton of bugs and outright removed features. Without real devs to fix it, companies who replace devs with AI will be left with products that simply don’t work in a matter of months.

        • queermunist she/her@lemmy.ml
          link
          fedilink
          arrow-up
          12
          ·
          edit-2
          5 days ago

          Absolutely, but cheaper now is all that matters to the Dunning-Kruger c-suit that runs every business. Besides, they might have already moved on to a different company by the time it becomes a problem, so why should they care?

          • Mirshe@lemmy.world
            link
            fedilink
            arrow-up
            2
            ·
            4 days ago

            Pretty much this. The business model now is create your killer app with AI, get it to where it JUST works, then sell the whole company to a bigger fish that has devs that can fix things. Rinse and repeat.

      • IcyToes@sh.itjust.works
        link
        fedilink
        arrow-up
        7
        ·
        5 days ago

        You tried adding stuff to a spaghetti code base or tried fixing bugs? Half the time you need to rewrite it. It’s not cheaper, you just pay more later.

        • queermunist she/her@lemmy.ml
          link
          fedilink
          arrow-up
          6
          arrow-down
          1
          ·
          5 days ago

          If it’s cheaper this financial quarter then they don’t care if they have to pay for it the next financial quarter. That’s a future problem!

    • MrLLM@ani.social
      link
      fedilink
      English
      arrow-up
      4
      ·
      4 days ago

      I wonder if some bolognese would make it a little bit better… I’ll see myself out

            • justgohomealready@sh.itjust.works
              link
              fedilink
              arrow-up
              1
              arrow-down
              1
              ·
              edit-2
              3 days ago

              The codebase is large, or you have files with thousands of lines of code? Because there’s a big difference. AI agents are perfectly able to work with large codebases, but they do have problems with large files (which usually are a sign of terrible code).

              • trolololol@lemmy.world
                link
                fedilink
                arrow-up
                1
                ·
                14 hours ago

                For an idea, we have over 150 gradle modules. Most modules range 3-20 kotlin files, with the older ones and new important ones reaching +100 files.

                Copilot often chokes doing grep when finding what files to work on. Then again when a minute change spans more than 20 files.

                My experience with my hobbies projects is much better, or when writing new code on big project that I don’t need to stick to existing code standards.

                It’s a very narrow scenario where it does what I need and I actually save time.