AI Helped Me Rediscover the Joy of Coding

I’ve written code since I was 19. Watching something come together from nothing, shaping an idea into a working system, that was always the part that kept me at the keyboard past midnight.

As Randoli grew, my own time shifted. More of the week went into architecture reviews, technical strategy, and conversations with customers, less into a terminal. When I did sit down to write code, I was usually the bottleneck: a branch would sit waiting on me for a day, or I’d touch something the team had already moved past while I wasn’t looking. Over three years, my hands-on contributions dropped close to nothing.

Mid last year I wanted to build something real again. I needed a system that could simulate production-like load and failure scenarios, so we could test how Randoli’s observability platform actually held up under the kind of mess real infrastructure produces rather than the clean demo path. Building an order-processing system to generate that mess gave me a reason to write code again, and I decided to seriously try AI-assisted coding while doing it.

I started with ChatGPT. It didn’t go well, and in hindsight the reason was obvious: I was pasting code back and forth in a chat window, which isn’t how I think about a codebase. Everyone serious about this had already moved to Claude, so I switched, first to Cline inside VS Code with Anthropic’s models. That workflow fit. The agent could see the files, make the change, run it, and I could stay in the editor instead of context-switching into a browser tab to copy code in both directions.

Even with the right tool, something felt off. I missed the older rhythm: carefully working out a method, optimizing a path, debugging for an hour and finally watching it click. That feeling is what kept me writing code since I was a teenager, and with AI doing more of the typing, it felt like some of it was being taken away from me.

It took me longer than I’d like to admit to see that the problem wasn’t the tool. It was where I was still looking for the payoff. AI-assisted coding wasn’t removing the craft. It was removing the boilerplate, the parts I’d quietly stopped enjoying somewhere around the tenth time I’d written the same kind of code. What actually needed to change was where I was looking for satisfaction, from the code block in front of me to the system that code was supposed to become.

Once I made that shift, the satisfaction didn’t disappear. It moved. Less time tuning a loop, more time on how the system behaved under a specific failure mode, where a design tradeoff would bite later, why a particular race condition kept resurfacing across three different services. That’s a different kind of work than the one I learned on. It isn’t a smaller one.

I’ve seen this shift before. Moving from C to Java and then Python didn’t make engineering easier so much as it moved the interesting problems up a level, away from memory management and toward the actual system being built. Higher-level abstractions didn’t remove engineering depth. They relocated it, and the engineers who kept their edge were the ones who followed it there instead of staying attached to the layer they’d first learned on.

The same relocation is happening again, and it isn’t really about the tools. Leaders who grew up writing code are going to have to find that same sense of craftsmanship somewhere new, at whatever the next layer of abstraction turns out to be, instead of assuming the satisfaction only ever lived in the layer they started in.