Yes, it really is that bad. It's based on XML, which in theory would be possible to do right, but all the actually existing XML libraries insist on validating everything at the drop of a hat (probably fetching schemas over the internet) and a horifically overengineered model of namespaces where you can't do anything until you've precisely defined all your namespaces.
The vanilla XMPP standard doesn't permit a server to deliver the same message to more than one client for the same user. This is utterly wrongheaded and falls apart as soon as you start logging in from multiple devices (e.g. phone + computer, or even just two different computers). This was obviously terrible and has been obviously terrible for decades, but it seems impossible to change it. There is an extension to work around the problem, but extensions take forever to get approved and twice forever to get actually adopted in real clients.
The groupchat UX was always awful, partly because groupchats were never properly federated (they have to live on a single server) and there was no way to do groupchat history for people who didn't keep all their devices turned on all the time (other than, you guessed it, an extension that got stuck in limbo for a decade or so) but mostly just because all the clients sucked. It really isn't hard to do groupchats right: all you have to do is alert when the user is mentioned but not alert on ordinary messages to the group. But somehow everyone managed to screw it up.
XMPP was objectively terrible (or at least, all the XMPP clients were objectively terrible). Worse than AIM. But there were people who were (understandably) blind to its faults because it was open-source, and this anti-synergized with the traditional-unix luddite tradition that rejects anything other than plain text in messages, and the good old standards-first approach that brought us XHTML2.0 and CSS2.1 and a decade with no HTML standard (i.e. trying to standardize everything before any implementation, which never works - what works is allowing clients to add extensions and then standardizing what's been shown to work and become popular, a la WhatWG).
The real issue is that if you try to build an XMPP client that defines something "non-standard" you'll get torn to pieces by the OSS fans. Many of the niche instant messengers are probably running XMPP under the hood, but they don't want to expose it or support federation lest they get accused of trying to "embrace, extend, extinguish".
The vanilla XMPP standard doesn't permit a server to deliver the same message to more than one client for the same user. This is utterly wrongheaded and falls apart as soon as you start logging in from multiple devices (e.g. phone + computer, or even just two different computers). This was obviously terrible and has been obviously terrible for decades, but it seems impossible to change it. There is an extension to work around the problem, but extensions take forever to get approved and twice forever to get actually adopted in real clients.
The groupchat UX was always awful, partly because groupchats were never properly federated (they have to live on a single server) and there was no way to do groupchat history for people who didn't keep all their devices turned on all the time (other than, you guessed it, an extension that got stuck in limbo for a decade or so) but mostly just because all the clients sucked. It really isn't hard to do groupchats right: all you have to do is alert when the user is mentioned but not alert on ordinary messages to the group. But somehow everyone managed to screw it up.
XMPP was objectively terrible (or at least, all the XMPP clients were objectively terrible). Worse than AIM. But there were people who were (understandably) blind to its faults because it was open-source, and this anti-synergized with the traditional-unix luddite tradition that rejects anything other than plain text in messages, and the good old standards-first approach that brought us XHTML2.0 and CSS2.1 and a decade with no HTML standard (i.e. trying to standardize everything before any implementation, which never works - what works is allowing clients to add extensions and then standardizing what's been shown to work and become popular, a la WhatWG).
The real issue is that if you try to build an XMPP client that defines something "non-standard" you'll get torn to pieces by the OSS fans. Many of the niche instant messengers are probably running XMPP under the hood, but they don't want to expose it or support federation lest they get accused of trying to "embrace, extend, extinguish".