An architect described a design intent to an AI system. A word got misread in the transcription and it wasn’t caught.

The building that came out the other side wasn’t wrong exactly. It was just different. Subtly, persistently, and expensively different in ways that didn’t show up until the steel was already in the ground.

The AI did exactly what it was told.

That’s the problem.

The gap wasn’t in the technology. It was in the assumption that the instruction was complete, that the intent had transferred cleanly, and that what was said and what was understood were the same thing.

They weren’t.

Experienced architects know that drawings can lie. That the gap between design intent and built reality is where most of the hard work lives. They’ve built practices around catching the drift before it becomes permanent, reviews, coordination meetings, and someone whose job it is to ask whether what’s on paper matches what was meant.

AI doesn’t close that gap. It accelerates through it.

The faster the system moves from instruction to output, the less time there is to notice when something got lost. And in a medium where a misread word can mean the wrong wall in the wrong place, speed without verification isn’t efficiency.

It’s just faster regret.