The current generation of AI coding tools is genuinely impressive. Give them a clear spec and they'll write production-quality code faster than most junior developers. That's real.
But here's what I keep seeing: the output is only as good as the input. And most teams are feeding these tools requirements that were vague to begin with.
Garbage in, garbage out just faster.
The real bottleneck was never how quickly code gets written. It was the clarity of what needs to be built before anyone starts writing. Requirements that drift. Architecture decisions made too late. Scope that wasn't defined tightly enough to survive contact with a sprint.