It will be driven down to the cost of having a good project and product manager effectively understanding what the customer wants, which has been the main barrier to excellent software for a good long time.

And not only understanding what the customer wants, but communicating that unambiguously to the AI. And note who is the "customer" here? Is it the end-users, or is it a client-company which contracts the project-manager for this task? But then the issue is still there, who in the client-company decides exactly what is needed and what the (potential) users want?

I think this situation emphasizes the importance of (something like) Agile. To produce something useful can only happen via experimentation and getting feedback from actual users, and re-iterating relentlessly.