Coding with LLMs

I have been using LLMs to work on stuff way outside my comfort zone lately. Like recently I wanted to build something in Rust, which is something I absolutely know nothing about. To see what the hype was all about.

So instead of spending time to learning the language as a good developer would do, I was lazy and used an LLM to help me with my lack of knowledge. I would describe what I wanted to achieve and it would help me understand the code, compiler errors, memory management and all that systems stuff.

It is pretty good for this kind of thing. You can build a project that would normally take a while of learning and actually get somewhere in a few days. Normally I would object in using LLMs for this but for a throwaway project it is pretty good.

There is a funny thing I found while browsing social media “vibe coding” this is a recent term going around, which basically means letting an LLM write all the code while you sit back and accept it without much involvement. Personally I am not a fan of that approach. The idea of blindly copying and running code doesn’t sit well with me. What I found out with using LLMs is that it often seem to suggest things that look clever or impressive but aren’t actually always practical or appropriate for the task at hand. So I always need to take the time to review, validate, and refine their output rather than just running with it.

But still it is kind of weird (and honestly cool) how you can suddenly work in languages you barely understand. But if I were building something serious I would really want to understand the language and write the code myself. LLMs can assist but they should not be a replacement of understanding the code/language.. especially in a professional setting.