I feel like most modeling is going to shift to generative contexts where the application provides robust descriptions to the user's client and the client rebuilds a representation of the object that the client is capable of rendering. These descriptors could also be sent and rendered/streamed from remote devices too.