KI-Codierung ist mittlerweile überall. Aber nicht alle sind überzeugt.
https://www.technologyreview.com/2025/12/15/1128352/rise-of-ai-coding-developers-2026/
KI-Codierung ist mittlerweile überall. Aber nicht alle sind überzeugt.
https://www.technologyreview.com/2025/12/15/1128352/rise-of-ai-coding-developers-2026/
13 Kommentare
AI is bullshit
Is everywhere and it crash everywhere
ai doesn’t think in a coherent manner, it just follows the strongest pattern match regardless of if it makes sense overall
thats why the code is broken as hell in multiple places and never works, it doesn’t really ‚know‘ what its doing at a higher conceptual level to tie everything together
F u c k A I
“Raw milk is everywhere, but doctors advise caution”
Linus Torvalds thinks AI coding is the future and he approves AI generated commits to the Linux kernel, as long as they’re properly reviewed.
It is neither the Messiah nor the Devil. It’s just a tool, and up to the user to use it properly. It still requires you to know what you’re doing.
Nobody is convinced. The only people pretending to be convinced are the executives beholden to their board that’s full of investors that dumped their money into AI and now insist on everyone downstream using it as much as they can.
Count down the days to when the US bails out OpenAI.
>These tasks can be tedious, and developers are typically glad to hand them off. But they represent only a small part of an experienced engineer’s workload. For the more complex problems where engineers really earn their bread, many developers told MIT Technology Review, the tools face significant hurdles.
Not sure this article is really adding much to the discussion that hasn’t been said a billion times already but this seems like a reasonable take.
I don’t think I agree with some of the later comments about agent based solutions though, I’ve found them to be pretty terrible at anything that goes beyond very basic and small problems. Especially the part about the models improving every few months – I’ve seen *claims* that models have improved but the actual differences in the real world seem negligible at best.
>Instead, he advocates for “disposable code,” where each component is generated independently by AI without regard for whether it follows design patterns or conventions. They are then connected via APIs—sets of rules that let components request information or services from each other. Each component’s inner workings are not dependent on other parts of the code base, making it possible to rip them out and replace them without wider impact, says Worden.
This also seems like a *horrible* take and I would hate to be the support engineer doing root cause analysis on any complex system that took this approach.
It’s a blast figuring out why our servers take a shit every time someone vibe codes an algorithm that looks correct and compiles but actually leaks memory or runs in n^2 time.
I really like doing that please keep it coming.
The new software abstraction
I’m a software engineer. I find it’s only really good for things that are _so common_, it’s borderline derivative. At that point, it’s not helping with “coding”, it’s effectively just helping with typing
It _can_ solve difficult problems, but only if given extremely specific directions. At that point, you might as well just code it yourself
Let me know when AI can write a powershell script that actually works
Cursor IDE hooked into your preferred LLM is insane and I’ve been a dev for 20 years.
Full stack development and easily establishes knowledge on an existing codebase.
As with any tool it works best with people who know how to use it.