In my previous blog, I said don’t vibe code your apps. Today I recognize I was wrong.

90% of the code I’m writing is now AI generated. Not one-line auto-complete — full features, tests, refactors. The models have gotten significantly better in the last few months.

But here’s the thing: you still need to understand what the agent is generating. It might look good, compile fine, and pass your tests. But if you know the domain, you’ll catch the slop. AI doesn’t replace expertise, it amplifies it.

I was wrong about the tool. I wasn’t wrong about being an expert.