Earlier quoted context omitted.
Gimp is a great example. If you look up screenshots from the late 90s of gimp 1.0 you think "Hey wow, that looks pretty great! I know where the buttons are, I can quickly scan them and it's clear what they do! It isn't a grey on grey tiny soup, they are distinct and clear, this is great. When is this version shipping? It fixes everything!" Apparently almost everyone agrees but somehow we're still going the wrong way,…
Krita matured nicely over the years and last time I found it quite easy to use. UI is hard. It got replaced by "UX", but nobody agrees what that really is. So it boils down to whatever impracticality designers dream up. When UI was easy, there were real research, data backing up claims of improvements and laid down rules to enforce some consistency. This became "unfashionable" and was removed.
It's not at all anymore about presenting consistent mental models, it's solely about the ease or difficulty with which particular isolated tasks can be performed.
It's also not automatically the goal to make all tasks as easy as possible. Instead, discoverability and "friction" are often deliberately tuned to optimize some higher-level goals, such as retention or conversion rates.
This is why we have dialogs where the highlighted default choice is neither the safe one nor the one expected by the user, but instead the one the company would like the user to take. (E.g. "select all" buttons in GDPR prompts or "go back" buttons if I want to cancel a subscription.
You can see that quite often in browsers as well, often even with good intentions: Chrome, for a time, still used to allow installing unsigned extensions but made the process deliberately obscure and in both Chrome and Firefox , options are often deliberately placed into easy or hard to discover locations. (E.g. a toggle on the browser chrome, vs the "settings" screen, vs "about:config", vs group policies)