A newer problem
If AI got you to a demo and then stalled
AI tools are good at producing something that looks like working software. That is a real advance. The gap is everything a demo does not have to handle: the business rules nobody wrote down, how P21 actually stores the data, and what should happen when a record is wrong, a file is late, or a connection drops.
Each new requirement strains a design that was never planned, which is why these projects tend to break more the further they go. What closes the gap is not a better prompt. It is knowing the system, the business, and the ways it fails before the first line is written.
Why AI projects stall, and what finishes them