AI Coding Is Just Coding Now
It happened in two ways: gradually and then suddenly
I’ve been avoiding writing about coding with LLMs. There’s just so much noise in the space that it’s easy to sound like you’re an AI maximalist (whatever that means). But I think that a few months ago we crossed a line which closes this debate.
Things are really polarizing these days and I know some people who have put themselves on the AI=BAD corner since the start. And that makes me want to write this.
We don’t say that we’re researching online for how to fix a bug. It’s just implied that you’ll use the internet for that. AI coding is heading the same direction. Having code written by AI in some capacity will be the defacto way of doing it.
Slowly and then suddenly
I’ve been experimenting with these tools since the faraway 2021 when you’d write a comment describing what you wanted and hope the model would fill in the function body. But in the last six months, the majority of my coding has been done with LLMs. Don’t get me wrong. I still manually write a lot of it. The thing is that when I count personal projects + code at work, my total output has increased by a LOT and that’s mostly due to AI.
The biggest jump for me was Claude Code, about six months ago. I try to keep my finger on the pulse, because though I could see how these tools could be useful I also knew they weren’t good enough yet. Over the years I tried Cline, Aider, Cursor, Windsurf, Gemini CLI then finally Claude Code + Sonnet 3.7. That was it. Things started to click.
With claude it was now the first time I could see how the agent’s tool calling worked. This transparency changed everything. Instead of mindless prompt engineering loops, I could look at the agent’s thinking traces, understand what went wrong when it failed, and actually optimize my workflow rather than trying to force a square peg into a round hole. Alongside that, the new models have unlocked a really compelling and useful workflow that at this point is completely pointless to argue against.
The point
Rogers Innovation Adoption Curve
I have a lot of thoughts on agentic coding that I’ll write about eventually. But one thing is clear: it’s not going anywhere. It’s not a fad. The way we use them is definitely still going to change, but when you cut through the noise from some people in the industry, and you give this an honest try you realize that this is just another way of coding and it’s here to stay.