Mu: (I agree with both of you).

1. Make a dependency Map of topics.
2. Have students tackle projects and learn what they need along the way.
3. Teach students A1 before teaching them A.

3 requires 1. 3 is not necessarily optimal, since it still implies a set curriculum. IOW, teachers teaching to students instead of students learning what they need. 2 is made more efficient by 1. If I'm trying to learn how to do a high-level thing like build a machine learning algorithm, normally, schools would require me to take calculus and statistics courses beforehand, however, I could just learn what a gradient is (and all the pre-reqs for that like partial derivatives).