Automated testing scales easily, but properly reviewing code/proofs/technical prose really doesn't. Lecturing scales even more easily, though maybe not if you want students to be asking questions in class.