I think this is challenging because there’s a lot of tacit knowledge involved, and feedback loops are long and measurement of success ambiguous.

It’s a very rubbery, human oriented activity.

I’m sure this will be solved, but it won’t be solved by noodling with prompts and automation tools - the humans will have to organise themselves to externalise expert knowledge and develop an objective framework for making ‘subjective decisions about the relative value of things’.