There is an oversupply of people who only know how to train models and throw them over a wall to have someone else sort out the mess.
There is an undersupply of technically skilled people who have the knowledge and experience to sort this mess out and bring it into a production system (I'm talking about data engineers and machine learning engineers).
I think this will correct itself in the sense that those who only did some online courses will not be able to find a data science job so they will either retrain themselves to become a data engineer or they will pick a different field (maybe the field they came from initially. I suppose having domain knowledge AND some data science skills could be very useful).
There's also an undersupply of managers and executives that know the right way to build and manage these kinds of development efforts. They think trial & error of data products is something we can do via Scrum or somesuch. And they don't understand anything about what is being built or have the data literacy to question/examine/learn from the data insights of data products
Agreed, the engineering side is where valuable growth is/will be. Also from the limited sample I know of (people I know who are hiring managers) the candidate pool for data scientists lacks people with genuine analytical ability. Training models is easy, thinking the right way about data is hard.
There is an undersupply of technically skilled people who have the knowledge and experience to sort this mess out and bring it into a production system (I'm talking about data engineers and machine learning engineers).
I think this will correct itself in the sense that those who only did some online courses will not be able to find a data science job so they will either retrain themselves to become a data engineer or they will pick a different field (maybe the field they came from initially. I suppose having domain knowledge AND some data science skills could be very useful).