Why is a generic bus protocol even needed?

Using http over unix domain sockets, or even making a simple protocol over unix domain sockets with a size header + json payloads is trivial in most languages. Unix domain sockets work with standard linux permissions, you can access them with curl, you can forward them with ssh, you can mount them in containers, etc etc. One socket per service.

D-Bus is an overcomplicated mess all over the place. You have services, interfaces, paths, _and_ methods to classify messages yet messages are still only partially qualified and rely on one connection per message flow. There are return message ids, but some services use polling instead. And even if you have D-Bus you still need to know application-specific protocol details to make any sense of the messages. Application protocols leak wire details into messages (i.e. video access requests duplicate random dbus identifiers in payloads) which makes handling dbus messages generically for e.g. proxying impossible anyway.