I think the "what can we build with these models" is a data science question. "How can we do it" is an engineering one.

Anyway, I'm grateful for ML engineering, MLOps etc. It's a natural evolution of data industry.