I also feel like the Data Science profession will closely mimic Software Engineering. The difference will be the "stack" in data science starts at pure mathematics and theory, and builds up to AutoML solutions. There is always going to be a need for full stack data scientists who can build custom solutions in assembly to fit in some special firmware. There also be a market for cheaper, less specialized "AutoML Analysts" who don't understanding the underlying methods, but can still build sensible models that bring value to the company.

At the end of the day, most companies will benefit immensely from fast and easy model building / analytics. For example, most companies are interested in finding more customers. Building a geographic targeting model is _not_ hard, and can be done with AutoML.

The stigma behind not knowing the theory will slowly start to disappear. After all, how many programmers know _exactly_ how their instructions are being executed by the CPU? Not many, unless you are pretty specialized.