The AI-code writing miracle has this tiny problem: you can't responsibly generate and deploy more code than what you can read. And you can only read code quickly if you write a lot too. I think it does boost productivity in certain areas, but not nearly 10x.
Very nice blog post based on a theory of the computer scientist Peter Naur - resonates a lot with me.
Programming is about building a theory, the code is an artifact. Junior developers reflexively accepting LLM-generated code don’t go through the mental struggle of building the programming model and only see codebases balloon with theoretically orphaned implementations.
link: 👇
Question for senior developers:
What helped you get "good taste" in how you think and approach your code, aside from time spent?
Books? Training? Mentors? A team/project? ___?
Asking because I've been thinking for a while on how to accelerate this "good taste acquisition"
@rockthejvm Having ownership of a project over many years and facing the consequences of my less good decisions. Having people who criticised my work (and really listening with an open mind to criticism). Seeing other people's cleaner code and trying to reconstruct their thought process.
Excited to announce our paper "Precise and scalable metagenomic profiling with sample-tailored minimizer libraries". https://t.co/Q8JNXuJbcT #kraken2#metagenomics
There's a way of programming that I've been doing for decades and I realized recently it doesn't come naturally to many folks, so here's a pro tip:
Write to the interface you'd like to have. That is, write the code you want, even if it doesn't work, then make it work.