Comment by knivets
2 days ago
> Over the past six months, there hasn’t been a single day where I’ve checked the HN Best RSS feed without seeing a post about how AI “writes bad code,” “introduces bugs,” “creates technical debt,” or something along those lines.
because it's true
> Users don’t care whether the code was written by AI or by hand, or which framework you used. They care that the product works.
How can you guarantee that it works though? You can verify, but it would be at the same speed as before the AI, or even slower.
> By then, enough real-world feedback would have surfaced to identify the major issues, and tools like Claude Code would make it possible to fix and ship version 2.0 at an incredible pace.
By then you have a blackbox of a codebase which is unmaintainable, or in a worst case scenario you end up losing your data or get hacked or both.
Boggles the mind that people can't see this. You gotta keep the guardrails pretty tight to succeed with LLMs and code in my experience.
>because it's true
I mean, it's just as true that most teams “write bad code,” “introduce bugs,” “create technical debt,” or something along those lines. There is this amazing thing that happens where when we talk about AI code, it's always compared against some idealized infinitely capable professional following every best practice, purposefully and carefully crafting every line.
Which is fantasy. Most of you are creating absolute slop. That's just truth. Clutch your pearls and say it ain't true, but there's this weird reality that when people inherit other people's code, it's declared 100% slop, almost all of the time. It's the reason we are constantly tossing projects in the bin and starting anew.
AI is absolutely imperfect, just like teams of human developers.
>By then you have a blackbox of a codebase which is unmaintainable
Have you ever used Claude Code to work on a project? In what reality is the code unmaintainable? How is it a "blackbox"? Are you getting it to write LLVM IR or something?
In my experience, and with careful guidance and oversight, it makes spectacularly maintainable code. Better code than any human developer I have ever worked with. It helps to occasionally do cycles of refactoring as you've built out the foundation and the core becomes more evident and clear.
>or in a worst case scenario you end up losing your data or get hacked or both
This one is amazing. The industry is awash with garbage, insecure, exploited code. We were making that garbage code by the data centre full long before LLMs joined the scene.
Yes, there have always been bad programmers. The only difference is that now thanks to AI anybody can be a bad programmer. You've got people out there contributing sloppy code like only a bad "10x engineer" could do before. Good code is still hard to write, and from what I have seen in 3 companies so far, the people who write good code with AI are pretty much the same ones that were writing good code before AI.
> AI is absolutely imperfect, just like teams of human developers.
The old "nothing's perfect therefore everything is equally imperfect" fallacy. It's not a binary. While everything is flawed, somethings are more flawed than others. Welcome to the world, it's complex.
> when we talk about AI code, it's always compared against some idealized
Have you ever seen a development mailing list? Seems like when human code is scrutinized it's held to a high idealized standard. "Technical debt" is a concept that originated from looking at human code. How then can it be true that applying it to AI code is setting a higher standard? It's setting the same standard. These things existed and were applied to human code before AI exploded.
The whole "many eyes" thing can be quite brutal. We don't always get the many eyes but when we do... wars are waged. It's brutal out there for anything getting scrutiny. Currently, AI code getting a lot of eyeballs. That's a good thing. Don't wish it away just because you're butthurt about your new pet tech being held to a standard. That's how it gets better.
The "problem" of AI being held to a supposed higher standard isn't a problem. It's a free pony.
What this looks like it practice: enterprise SaaS vendor introduces bug, I report the bug, get a canned AI response because literally nobody cares, and it never gets fixed. The product continues to deteriorate.
No, it’s not just one vendor.