Everyone talks about how AI lets you ship faster. More code, more features, more output. I want to talk about the morning I sat down and deleted 825 lines of code, and why that felt more productive than anything I had added.
Yesterday I spent a few hours on cleanup. The numbers: +387 lines added, -1,212 lines removed. Net: -825. The codebase is smaller, simpler, and does exactly the same thing.
Here's what I keep coming back to: every line of code is a liability. Not just technically, cognitively too. Every abstraction is something the next developer has to learn. Every class is a surface area for bugs. Every file is something that has to stay in sync when the rest of the system evolves.
AI has made this harder to notice. These tools are genuinely incredible, I use them every day, and I wouldn't build without them. But they optimize for output. Ask an AI to implement something and it will. What it won't do is tell you when the right move is to delete what you already have instead of adding more.
And the danger isn't just maintainability. It's comprehension. When a production bug hits at 2am, you need to move fast. If your codebase is full of AI-generated layers that nobody fully understands, including you, that's not a velocity advantage. It's a liability waiting to surface at the worst possible moment. Making code a machine can execute is the easy part. Making code that a human can reason about under pressure is the hard part.
I wrote about this dynamic back in 2025. AI is a tireless pair programmer, but it's not taking the wheel. That's still true. What's changed is that as the tools have gotten better, the gap between output and understanding has gotten wider. The collaboration is more powerful than ever, and so is the responsibility to stay on top of what you're actually building.
The future of software engineering isn't humans vs. AI. It's humans keeping AI honest. Knowing when to build. Knowing when to simplify. Knowing when the most productive move is to make the system lighter.
So next time you wrap up a morning's work, ask yourself: is the codebase easier to understand than when you started? Because the best sessions often end with less code than you began with.
