How Can I Learn While Using AI?

Recently I had a problem with my sound in Linux. Although Linux desktop had progressed immensely in the last several years, still sometimes I find myself trying to understand what went wrong. This time, instead of diving into the system logs, I let an LLM model do it. Quickly I got the root cause and a solution which was to turn off the unnecessary Pipewire plugin. That led me to a sad realization.

LLM models, together with an agentic harness like OpenCode, have become really good in debugging my system problems. So good, that when I have problem in my system, it usually takes me one prompt and maybe a few minutes to sort it out. Before it, I might have spent hours and sometimes even more, trying to understand what is wrong and how to fix it.

Having all my computer problems sorted out is really nice, but it also made me notice that I don’t learn as much about my computer system anymore. Sure, I get the surface details from the LLM output but without delving into the system I miss all the details along the way that I would have usually learned.

Unlike programming itself, which I find much more important to maintain the hands on experience, I feel like I can’t justify spending that time on fixing random problems that the LLM will fix quickly. Will my employee understand the need for it? How can I myself differentiate between a detour that will give me a good lesson and a random bug that will just waste my time?

I don’t know what I will do about it, but I do know that I miss that part. My blog also contains random bits of information I learned along the way. Those posts were not about my core work that I did, but about the systems that I’ve used and small problems that I’ve stumbled upon.