The real problem with AI coding

The real problem is the Ready, Fire, Aim! approach to developing that starts with writing code first. As the article states, successful teams begin with architecture: define the components of the solution and how they interact, along with identifying data flows and data management. You can use AI to assist in that effort - and capture the result in an architecture diagram: preferably using a standardized modeling language such as ArchiMate that we can feed back into an AI in the future.

The same process can be applied to design. Again, designs can be captured in a modeling language such as UML that we can feed back into an AI in the future.

With those models in hand comprehending the code is relatively easy - and it doesn't matter whether the code was written by a person, an AI, or a combination of the two. But when you just jump straight to coding? That's problematic even without an AI.