[smdz]

> If you're doing _groundbreaking_ research in, say, AI or facial recognition or whatever, you may need people who are technical superstars.

No - AI and ML stuff does not need technical superstars (i.e. from a programming perspective). Usually it is a rinse-test-repeat kind of job. It needs people with patience and who can interpret data and who can be comfortable with uncertainty and then can make their bosses comfortable with uncertainty on timelines.

In most of the other cases, hiring superstar developers help you save costs in the longer run. It helps product (or SaaS) companies.

But for consulting biz that means less of the recurring consulting work. Superstar developers cost higher - means more work in fetching some consulting work. Consulting/service businesses will rarely value superstar developers. Because it does not help the topline as much.

[santaclaus]

> If you're doing _groundbreaking_ research

I think research is the key word here. If you are developing fundamentally new ML techniques (like the folks at Google), you'll probably need some serious technical chops.

> Usually it is a rinse-test-repeat kind of job. It needs people with patience and who can interpret data and who can be comfortable with uncertainty and then can make their bosses comfortable with uncertainty on timelines.

That sounds less like research and more like applying established tools (Tensor Flow or Keras or whatever).

[hobofan]

> If you are developing fundamentally new ML techniques (like the folks at Google)

Nope, still no superstars needed. The most technically challenging thing they've built is probably is what they are doing with the TPUs, but even there you "just" need people expierienced in the right fields. Most groundbraking research stuff these days is mostly on the mathematical side, not the implementation.

[maxxxxx]

"That sounds less like research and more like applying established tools (Tensor Flow or Keras or whatever). "

This is what most AI, ML and data science people do from my experience. It's really not that groundbreaking.

[trhway]

>...rinse-test-repeat kind of job... people with patience and who can interpret data and who can be comfortable with uncertainty and then can make their bosses comfortable with uncertainty on timelines.

somehow that isn't a definition of a technical superstar. I suppose a superstar is a DC/Marvel style character who do software engineering impatiently, skipping the test-debug-fix-repeat ... Well, save for the missing Batman mask it is basically the lowest productive engineers i've ever met, and probably you really don't need these superstars.