If I reference some AGPL projects (without actually copying code) to write my own project, is that frowned upon in the FOSS community? My guess is no.

So to me the difference between the two sides, is about fair use. One side thinks AI training isn’t inherently violating copyright licenses, and the other side thinks it is. FWIW, I believe the US courts side with former.

I’m confused what you’re trying to say here. It sounds like you are trying to say that unix philosophy fails to address the difficulty of actually composing and integrating the simple primitives provided by GNU/Linux. But you could say that for any programming language.

Unless you are saying that the shell should have intermediate tools that are more specialized than the primitives. Like libraries or frameworks built on shell primitives. But I think for that sort of complexity, shell scripts are too slow and counterintuitive, and you should just use a proper programming language, like all the other complex specialized cli tools in the Linux ecosystem (nmcli, systemd, etc).

I see shell primitives as relics of the past, and I treat them like you might treat the buttons on a GUI. There’s a button for copying items called cp, a button for moving items called mv, etc. I rarely compose these tools into complex scripts, and I almost never write bash scripts longer than like 10 lines. For any sort of complexity I use Python or Go.

Defining a threat model means to qualify what sort of attacker you expect to face and what they are assumed to be capable of

Which is exactly what OP is talking about. 5 years ago if you were only trying to defend against script kiddies, using containers was good enough. But now you have to consider that script kiddies might find container escapes.