That's a tricky one; there's just such a wide variety of learning styles, goals, prior knowledge, interests... In my formal education, we used a 68k platform to learn assembly and the sorts of things you're talking about. That epistemic style is very different from what I see most makers using, and maintaining excitement about.
As you say, the debugging story on AVR in particular isn't great, but these tiny systems are famously hard to debug in general. I honestly think that people learning this stuff would usually be better served on a bigger machine where there's at least a console available. Yes, more complex, but at the same time that complexity allows for some separation between the thing being debugged and the debugging environment itself.
Professionally, most of my biggest challenges are around working with people who tend to attack the low level problem, without leveraging the high-level tools and abstractions. They might "fix" the problem quickly, but it creates this horrible maintenance situation where every little thing is subtly different, iteration is incredibly slow, and the system requires a bunch of skill to work on at all. I'd much rather have people who're good at software development and vaguely aware of how the machine works under the hood, than folks who're great at assembly and vaguely aware of how to use Docker.