> This comes up every time, I just don't understand the argument. You can fork AOSP when that happens

Hacking on the Linux stack is several orders of magnitude more practical than AOSP though. AOSP's build system is massively unwieldy for individual contributors to work on. You can't easily work on any of the system components without dedicating >100 GiB of storage for a build, for example. Everything is scattered across hundreds of repositories tied together with a bespoke repository management tool.

This all works for a company equipped with entire build infrastructure teams, not for a grassroots project trying to keep up maintenance with (at best) limited funding. It's why Nura (fka postmarketOS) is a thing whereas the various AOSP forks barely ever modify the core experience.