[throwawayjava]

Basically, in most college courses the problems are either 1) easy enough that implementing a program to solve them isn't super insightful, or 2) difficult enough that complete automation would mean "do research".

I'll limit myself to Math since that's the topic of this article:

Calculus sequence: CS 1 is not a pre-req. And there's not enough time to teach both CS 1 and Calc 1/2/3 in a single course. "Implement it" works well for derivatives but not integrals. You're not gonna teach Risch, and implementing integration tricks isn't particularly insightful IMO. The cost/benefit ratio explodes in Calc 3, and the physical intuitions become as important as than the calculations.

Everything past that is proof-based and now you're kind of in "your homework is an open research problem in combining NLP with theorem proving" territory. Maybe with the exception of particularly bad Linear Algebra courses and a bit of the early stuff in Algebra.

From a "pragmatic skills" perspective, this approach is still highly suspect. E.g. no one's going to invent their way to Risch by implementing integration tricks.

Point is, every field teaches useful life skills / knowledge, and programming gets in the way as often or more often than it helps.