I think the point is that it builds on existing infrastructure that almost everyone already has in their life. Whether you self-host, use a niche provider, or a mainstream provider, email provides a pretty interoperable starting point which isn't a walled garden and doesn't have to be owned by FAANG.

Are there some compromises? Of course. But creating new accounts, or using new services isn't one of them. Email-speed instant messaging is good enough for me.

And to be honest, as much as I like Matrix, it has plenty of compromise of its own - including speed/reliability when using their own servers.

Only because we choose it to be.

When you use XMPP or Matrix you know your message will be delivered instantly. When you use email you don't. But that's only because we haven't defined an extension that tells the client whether the server can deliver messages instantly. It's not really that inherent to the system. Surely you've had at least one real-time conversation by email because both of you happened to be online at the same time and nothing went wrong necessitating a message delay.

Yes, it's theoretically possible to adapt many protocols to many use cases, but that doesn't mean it's a good idea. Email and XMPP are just geared towards very different use cases.

In particular, Email has a lot of assumptions about acceptable delivery delays, bidirectional (non-)reachability etc. baked in that would be very hard to globally undo.

As programmers, is it not our natural instinct to create a ChatSystem and MailSystem on top of a GenericMessageDeliverySystem? When did we stop doing that? Creating a brand new GenericMessageDeliverySystem for each purpose comes with substantial costs. We don't reinvent the Internet to create WhatsApp.

What generic message delivery system are you talking about, though?

SMTP is optimized for the use case of email; XMPP is optimized for one-to-one instant messaging; Matrix is optimized for "Slack-like" chat.

I highly doubt that these three use cases are similar enough to warrant introducing an additional common protocol/layer.

Optimized how?