I think we're past the age of building fancy models. After about a decade or so of seeing where models could make an impact on decisions being made, I think we're entering a phase where operationalizing these models is more important than building bigger and more powerful models. The field of data science is evolving yet again, this time with a big offshoot branch with the name "Machine Learning Engineering".
The key problem here is that most data scientists don't possess the software and engineering skills to build and operationalize machine learning systems for colleagues. There's a big opportunity here to make machine learning systems work. The key here is to design machine learning systems the way software engineers design software systems: modular, with clean contracts, minimizing entanglement, etc. etc.
There are very few resources that I know of that can help us learn what needs to be done, but here are a few that I have encountered.