Just spitballing, but what if multiple AI agents played different parts and had to synchronize, harmonize and even improvise with each other? This might add the texture necessary to make the result more "listenable"?
Anyone out there trying to demonstrate this?