The challenge, as you illuminate it, comes with any new frontier. It is the classic production puzzle, not a machine that decides, but humans struggling to make choices. A lot is said about the machine learning part, not so much about the practical engineering of the use itself. We've been here before. In early computing days, the promise of automation did not erase the necessity for process. Similarly, these days we see a race towards the use of this new technology, but without the necessary process thinking. The rush to market overlooks the basics. It might be beneficial to bring up experiences taken from prior projects, even relatively simple ones.

Consider a case where the focus has been on structuring the use-case and what steps were involved, as exemplified here: https://medium.com/dev-genius/its-not-only-about-prompting-but-planning-llms-196110865cf7

It is crucial to have clarity. That clarity needs a plan. There was never a time without order. So why should we expect that this machine would need less order than any previous attempt in bringing change and novelty.