I’ve been building software for nearly two decades. For the past year, AI tools have been part of my daily workflow. Here’s what actually changed, what didn’t, and what surprised me.

What changed

I prototype 5x faster. The gap between “I have an idea” and “I can show you a working version” collapsed. Things that used to take a weekend now take an evening. This changes how you think about building. You try more things. You throw away more things. You converge on the right answer faster because you can explore more of the solution space.

I write differently. Not worse. Not better. Differently. I think in higher-level chunks. Instead of “how do I implement this sort algorithm,” I think “what’s the right data structure for this access pattern.” AI handles the plumbing. I focus on the architecture.

I read more code than I write. This was unexpected. Because AI generates code quickly, the bottleneck shifted to reviewing and understanding what it produced. Reading code carefully became the critical skill.

What didn’t change

Judgment. Knowing what to build is still the hard part. AI can’t tell you whether a feature is worth building. It can’t tell you that your users don’t actually need what they asked for. It can’t tell you that the real problem is organizational, not technical.

Taste. The difference between good software and great software is still about the thousands of tiny decisions that require context, empathy, and experience. AI generates plausible solutions. Shipping great products requires knowing which plausible solution is the right one.

Communication. The hardest part of building products is still getting humans aligned on what to build and why. AI doesn’t help with that. If anything, the speed increase makes alignment more important, because you can now build the wrong thing even faster.

What surprised me

Junior developers need different skills now. When I trained teams this year, I noticed juniors who grew up with AI tools have different gaps than juniors five years ago. They’re faster at producing code but sometimes struggle to debug it when it breaks. The fundamentals still matter, but the path to learning them changed.

I’m more ambitious. Because the cost of trying something went down, I attempt things I would have dismissed as “too much work” before. Some of those ambitious attempts turned into the best things I’ve shipped recently.

The net effect: AI made me a better builder by letting me spend more time on the parts that matter most. The thinking, the deciding, the connecting with users. The code is just the medium.