I don't disagree, but whether you're even aware of the XEPs and how it's presented to the user, is a critical factor in viewing it as "confusing". Gaim for example only even tells you about XEPs if you dig into the server settings, and then it shows a very good job of listing all XEPs from either the server or client and noting which are supported by each in a table if you're far enough down the rabbithole that this info is useful. But for a regular user they just log in and it Just Works (tm).

It's overwhelmingly more of an outsider talking point than an actual issue in practice. There's a category of people that just says any extensible protocol must fundamentally have massive amounts of incompatibility, and brings it up every chance they can... and while that is technically always possible, it only happens in practice if clients diverge greatly. XMPP clients mostly work together much better than Matrix clients, from what I've experienced, as long as they've been actively developed at some point in the last decade. Which is by far most clients in use.