Generally, my feeling now is that Mosaic is a fantastic starting point, but that judging from the number of presentations which require either a modified browser or, at best, a kuldge-laden technique, it has a long way to go.
That WWW servers cannot contact WWW clients is probably the biggest problem facing interactivity. Many applications need a way for the server to update a client's version of a page. This theme was present in both the HCI session and strongly throughout the CSCW presentation.
Extensibility seems to be a great problem. Security, commercial transactions, synchronous collaboration, agents, and so forth, all seem to need ``little'' changes to the client, which is currently impractical. Perhaps what is needed is some sort of way to download simple scripts to a client, so that when a needed feature is absent, the server can recognize this and download any needed modifications. But there are an enormous number of technical problems with such a scheme. Nobody came out and really said that this is what is needed, let alone propose how to solve the problem.
An issue with these things is the question, ``is our installed base already too large as to preclude serious technical modifications?'' This is amazing to contemplete, since the Web is only about two years old now. We've seen this sort of difficulty before (MS-DOS, the Mac OS, Unix, actually just about everything that's become a standard), so it's not surprising that it would arise with Mosaic. But is it too late, or does the incredibly dynamic nature of the Web preclude these problems? It's something to think about, anyway.