AI: to use or not to use as a programmer?

I don't really like the idea of using AI for programming and I'm definitely not a fan of a future where programmers are mostly managers, reviewers and orchestrators for AI-based tools. I'm one of those who ended up in this industry because we liked the craft, the logic puzzles that come with developing and the whole process of building something with blocks of code instructions.

Whether we like it or not, the AI industry has gained a lot of track in this field and nowadays it's hard to stay away from tools like Claude Code, with all their faults. As a modern developer, chances are you won't get a choice if you will use it or not. So I'm afraid the answer to the question in the title, "to use or not to use AI as a programmer", is "you kinda have to...". Which raises a new question. How should you use it?

I won't pretend I have a definite answer. After all, I don't even have that much experience with AI. But from the perspective of a programmer who wants to keep his skills sharp and also remain employable, I think there is one rule you should follow.

There is a crude saying that goes "don't sh*t where you eat". In my case, as a frontend developer who often deals with highly interactive interfaces (think editors) and reusable and highly customisable code / widgets / components across multiple clients and apps, I'm trying to use AI for things that are not in my area of expertise.

For example, I uses AI to successfully configure Docker images or to debug webpack / webmix / vite issues that would have otherwise been very time consuming for me. I remember spending hours in the past for such issues, especially during my hosting adventures. It was fun to investigate and go through trial and error with no real stakes and the experience itself it was very valuable. At the same time, I forgot most of the details.

I have configured multiple subdomains on my Hetzner VPS, played around with nginx configurations, manually installed SSL certificates and I still go back to my own blog posts when I have to deal with similar bugs. Because it's not something I do frequently so the information tends not to stick. So this is use case number 1 for me: time consuming issues in areas where I'm not an expert (and don't intend to become one).

Second, I've been meaning to make some updates to my personal website for quite some time. Some of the changes I did (and were fun!). But others are just tedious, like adding a navbar. It's the same old HTML and CSS structure and I just can't bring myself to do it. But if I can get AI to do it for me? I get a much needed update and not waste any real energy or time for it. So this use case number 2 for me: generic tasks where you already know to do them very well. It's good to still do such tasks occasionally, for that "muscle memory", the keyword being here "occasionally". We do have limited time after all.