On the Joy of Claude - Pastebin.com

4 min read Original article ↗
  1. Much digital ink has been spilt by proponents of LLM coding and its detractors. I'd like to spill some more by addressing comments by a particular subset of LLM users, on the spectrum closer to vibe coders than tab-completers, who say that tools like Claude Code and Codex have restored their love of programming, that coding sparks joy for them once again. HackerNews is filled with comments of this nature, and invariably one level of indentation beneath them depend dozens of replies explaining that what they're doing isn't actually programming, that AI is using them and not the other way around, and maybe they actually never liked programming in the first place, because programming is in the details.

  2. I count myself among the detractors of AI and I believe that exposure to it leads to cognitive decay and social homogenization, isolation, and enshittification, but I have also caught myself enjoying using it (or being used by it) to create some programs that I would not have otherwise created, similar to those commenters I've read on HN. Hypocritical feelings deserve further examination. Are the HN replies right, do I even really like programming? I think I do. I'm a professional programmer, and after my workday is done I program more (by hand) in my free time. So what am I getting out of AI, when my philosophy tells me I should hate it?

  3. I came to realize that at the beginning of my programming journey, understanding how the computer works in order to make it do *what* I wanted it to do was what got me hooked. But the more experienced a programmer I became, the more opinionated I became on *how* a thing should be done. With exposure to different programming languages, libraries, frameworks, paradigms over time, I came to see their flaws more and more clearly. Essentially, I grew to have a strong sense of taste. My sense of taste grew stronger until unfortunately it became a pain to actually make anything, because I saw how much I could improve my tools.

  4. This leads in some cases to analysis paralysis, but for me it led to yak shaving. The project I've been working on (by hand) for years now is a compiler for a new programming language, which I'm making in order to make an interpreter for another new language for a game framework that I will make, so that I can make a game.

  5. The need for these things is real, yes, and I love the project I'm working on. But over time, I've allowed myself to get further distanced from the fact that what I actually wanted to make was the game.

  6. And then I understood the spark of joy that I got from using an LLM to code. The thing got me over my yak shaving, because the language, paradigm, and tooling became irrelevant; all I had to sacrifice was my understanding, and tell it what I wanted. It gave me the same feeling as if I actually created something useful, which I haven't had in a long time, because I've only been creating tools; all I had to sacrifice was my own creativity and growth. The process of creation, now bloated and elaborate, had been getting in the way of the product so much that getting rid of the process entirely seemed to be progress.

  7. All of a sudden, I understood the people that said Claude put the joy back into programming for them. It's not that they didn't like programming, it's that they, or at least some of them, were paralyzed by taste to the point where they couldn't make anything. It feels good to actually make something again.

  8. While I sympathize with these people, I'm not willing to sacrifice my understanding or creativity or personal growth for a product, and I won't be using LLMs for my personal projects in the future. More important than the product I'm making is the fact that *I* am the product of these creative efforts; the process is the product. But the realization that the psychological value of LLMs is actually to free one from yak shaving has itself freed me from the fear that maybe I do need to give in and use these tools in order to feel that joy again. Instead, Claude taught me a lesson that I knew at one point long ago but had since forgotten: use the tools that exist today, and make the damn game.