Writing Code for the Joy of it

These days, some people argue that we shouldn’t be writing most of our code: that coding agents such as Claude Code have significantly higher output, and that software developers should simply be the filter between Claude Code and production. I don’t think that always applies, though. I believe there are still times when you should write most of the code, if not all of it. One of those cases is “coding for the joy of it”.

Code as a form of art

Coding can be a form of artistic expression. The output can be expressive, but so can the architecture, the way your modules interact, your design patterns…

Take a side project, a passion project so to speak. If you enjoyed coding as a hobby before AI got so good at it, then you probably know what it feels like to truly own a project: to have this fully working thing and be able to proudly say “I made that” or “we made this”. Almost every line of code, every design decision, every tiny choice is yours. You end up feeling as proud of your creation as an artist who just finished a painting. Once you finish that huge project you always wanted to make, you can say, “This is my masterpiece, my Mona Lisa.”

Owning what you build

When you use Claude Code to write your whole project, or big parts of it, you don’t get that. I’d say most people aren’t even reading the code AI spits out. When speed and output become the number one priority, and you don’t care about the details of how the thing works, you probably won’t bother to check.

When I’m starting a big project, I want to own it, like a painter owns his masterpiece, like a car enthusiast owns his build, like an engineer owns his system. When I’m done and people can see it or use it, I want to be able to feel like “this is me”. The project is an output of me; in a way it represents a part of me, it has my soul in it. Not that I believe you can compare the “soul” of a web app to the soul put into a creative piece of media like a painting. But it can have soul in a different way, if you let yourself be creative.

AI is not creative; humans are. The outputs of an LLM are statistical and lean toward safe averages. Even when you instruct it to add your creative ideas, it will implement them in the most “average” way, if that makes sense.

Writing code to learn

Another time you should probably write 90% or more of the code yourself is when you’re learning. As a software engineer, you should always be learning, and over-reliance on AI will atrophy your skills. Take someone like me, who at the moment has never built an agentic LLM harness with other AI features. I could tell Claude to build one for me, and it would. But then I probably wouldn’t truly know how it works. The devil is in the details. Sure, you can look at the implementation and read it, but you have to get your hands dirty and understand all the parts of the system relevant to your role. Because when something in one of these huge vibe-coded codebases breaks due to an emergent failure no one saw coming, it won’t be the vibe coder who only knows the basics and how to prompt Claude who fixes it.

Final thoughts

While AI agents may significantly increase raw output, there is definitely still a place for writing your own code. If you enjoy it, I think you should do it more and not feel like you’re falling behind. The person who is willing to do the hard thing will have an advantage. I do find these tools useful, and I even use them myself for small things like frontend design (CSS), which I don’t enjoy as much as building the system or logic. But I’ll keep writing code by hand whenever I have the chance, and I think you should too.