New guy gets a task: "Add a thing to the thing". One liner. Simple enough.

Commits code written by chatGPT to an unrelated branch, proudly screenshots the resulting screen and posts to work chat - "My first task!! What is next?"

I don't think I want to work in this field anymore.

Sounds like an enthousiast newcomer that just needs some guidance. Invest!

Yup, I'm with @retoor.

If he can modify/explain it, I wouldn't even be worried about it being from chatGPT

@retoor Agreed. I've been putting on brakes at work with AI for juniors, because if they don't understand what the LLM is doing in the code, it will be really bad for them in the long term. I want them to really understand what they are doing and be conservative with just trusting AI-generated code.

One shouldn't just blindly accept LLM code as good enough. I press it to provide better options and always modify it to make it more readable by humans. Recently I had to fix so much shit code generated by an LLM...

@int32 true true true. But we also accept that python devs don't need to learn c for what really happens, so why is it different for llm usage? Hypothetical question. I also think they need real deal first. The only thing I have to say to some who is convinced that gpt coding is the way: you'll become a better prompter if you know shit under the hood. We have to come with good arguments to fight the new mindset. But in general: you get stuck with prompting very quickly. Every prompter will find out at certain moment that it's time to study.

@retoor I have used Claude extensively with Angular and ColdFusion (vomit), because I don't care for learning the inner workings of those frameworks. So yes, use it for working with unfamiliar stuff such as languages, especially if the documentation is shit and there's nobody else to help you out.

@int32 yeah, for me same but with css/html. I just don't care :p