Reflection on LLMs 01
(This is a self-reflection; I've had these thoughts floating in my head for several months, and I think writing them down will help me navigate them better.)
I'm usually working on a programming side project of some kind. I've approached them with different lenses over the years - varying from "I'm going to create the next big thing!" to "tool that will only ever used by 1 person a couple of times a year". And, as is tradition, most of these projects are now abandoned. I've mostly made peace with that.
I use side projects as a way to build mastery in a field that is ever-evolving. I enjoy gaining understanding of new technologies, tools, and patterns. Whether or not I finish a project (rare) is honestly irrelevant to me - as cliché as it might be, it's the journey that holds the most value.
In that context, I find LLMs valuable when trying to understand a new language or system, or as a sounding board when researching or thinking about a complex problem. A daunting knowledge gap that would have previously stopped a side project in its tracks can now usually be broken down, understood, and resolved quickly.
I also enjoy using these tools to remove what I consider tedium. For example, I recently used Claude Code to create a complex test harness for a tool I'm working on that would have otherwise taken me on a days-long tangent. I probably would have given up working on the project had that been the case. Claude basically one-shot it, and produced a very polished TUI on top, within a matter of 30 minutes of prompting.
However, I've learned that I don't actually enjoy vibe coding the meat & potatoes of these projects. The journey of struggling, researching and attaining enlightenment is what makes it all fun, and having Claude spit out hundreds or thousands of lines in a matter of minutes robs me of that experience. The speed-up gain is really not worth the technical growth in this context.