AI is turning developers into pull-request machines, the numbers are real, the stories are everywhere. But shipping code and building software aren’t quite the same thing. Something in the math doesn’t add up.

AI is making developers faster, the dashboards say so, the line counts confirm it. But writing code was never where the time actually went. Somewhere in that speed, something got lost. Nobody put it on the dashboard.

The demo was impressive. The slides were polished, the use cases relevant. But the moment questions started, there was nothing behind it. Producing content and having knowledge are not the same thing, and the gap between them may be wider than we think.

Developers have never been so divided over a technology. But the fault line isn’t where most people think it is. It was already there, long before AI arrived.

When a technology is hot, everyone feels compelled to weigh in, and it’s always amusing to look back at the bad takes. Rather than laughing at others, here’s a chronicle of my own wrong predictions about AI, from dismissing ChatGPT to underestimating AI coding tools.