This comment and the grandparent both ring true for my and my experience. When using Teams as basically an IRC replacement it worked fine and didn't have any serious issues, when folks started using it to try to host meetings across sites and share interactivity it really struggled unsuccessfully.
I agree with Jacques that all the pieces (except for high speed networking) were in place in the last century I expected this to develop more than it has.
Given the stuff that ML can do TODAY, it should totally be possible to give the "presenter" a view of the audience where you use a combination of the pre-computed model for people's faces, and a client doing pose estimation from the web cam. If I'm looking at the presentation on my screen, that information should allow the presenter's view to see me looking at them. If I'm looking away it could show me looking down or something. If I leave and go to the bathroom or something it could show my seat being empty. Similarly, raising a hand should present on the speakers screen as my raising my hand in the audience, calling on me with the mouse should unmute my microphone until the presenter moves the mouse/pointer off my face in the audience.
Early, EARLY, on in the Java group the idea had been worked on to provide a "visual environment" for doing things, and of course "Snowcrash" came out at the same time and everyone was doing their own "metaverse." I was looking at the problem and realized that to be successful we'd have to emulate some of the physics that keep us sane in group settings today. (like everyone can't hear everyone else talking in a room usually). David Rosenthal did a study for NIST, as I recall, on the network implications of something like the metaverse and concluded that it would take a terabit of network connectivity to "host" a live concert in the metaverse with the same "feel" as being there in person. (so everyone gets a view from their POV, voices travel 10 - 30' in a cone from the direction of the mouth, sound from the stage follows the acoustics of the modeled room etc. And then that analysis is followed by trimming things out to reduce bandwidth until you reach the "present day" bandwidth numbers and describing that reduced experience. He got back to Zoom like meetings with something like 1mbps/10gbps for the participant/network ratio.
So summarizing; Often people approach the problem without realizing the complexity of what they are trying to do, in part because meetups/communication in "real life" has so much of the complexity taken care of by physics and the fact that you learned what worked as a toddler before you could think so it just "is" in your brain. Stepping outside of that mindset requires deliberate effort. There are likely opportunities here that are not being exploited because of this mindset.