LLMs write worse code than any human's code. No logic. Fuck generative AI. I fucking hate it.

I actually do not agree. GPT writes many times exactly what I would've done but even does validation on top of it. And, personally, I never write validation for private projects since it's just not written for donkies. I see some cases that AI fails, but the things I use it for, it does better work than I do. But that's not very weird, I let it only do things what I don't like, like templates and css.

But I also genere complete modules for my projects. I have a example class of mapper, model, and service and the __init__ files it Hass to update and then example template files. I have that as one prompt. Below, I ask it to refactor that module to something new. Using my own vibe coding tool it automatically generates like 9 source files exactly the way how I've written it.

Thing is, don't let it do stuff. Invest time in some examples and then it'll do magic and exactly what you want. AI is lazy, but not THAT lazy. AI does now exactly as I want with some exceptions.

Hit and miss in my case

@netikras

This exactly

@retoor for boilerplate it's fine, but for more complex business logic, it just makes more bugs.

I use it just to add a bit of polish which it does amazing at, plus to spot tiny errors. For things like cleanup it does really good at, ofc deeper problem solving can be an issue.

Unless you ask for something extremely specific, it just doesn’t work.

Agreed. Most of the time it suggest codes that is not even part of the documentation (suggests functions that doesnt exist in cases of API integrations)

Instead of saving time, you spend more time. Its only good for simple queries.