Understanding underlying math is what allows people to create improved algorithms. If it's just "implement scikit" then you barely even need a developer.
Understanding why things work is still important and why software companies still routinely test on algorithm design.
If you're a beginner at ML, the first course don't really need to be so thorough on math, especially for simple things like regression.
I'm not saying math is dispensable, I'm saying that it could be used much less because code allows it in many situation.
Ultimately, if what you're teaching is going to end up in software, why use math at all? Use code or pseudo code. I don't think it's bad to just give the working algorithm without having to prove the math.
Really how many students will end up being computer scientists anyway, and research and write about new methods of doing AI, and do the actual math? So few. I guess that's a simple criticism of academics.
It's just easier to work with code than mathematical notation most of the time, in my view. You can't replace math, of course, but when things are simple enough, it could be avoided. It's a matter of making math accessible to the most people.
Code is amazing because a computer can check to see if it works. A computer doesn't understand math.