Well, the phrasing stops just short of saying that Fortune 50 companies are actually using the platform in all these application areas. I'm curious whether these companies are using g-speak for critical work or just trying it out in pilot/research deployments.

I think that we're not giving these guys an entirely fair hearing, in that most posters seem to be tying all their claims to the interface, when the platform that they are hyping has networking and interop components. I'm hoping that their "large datasets" claim is pointed in this direction, as the zoomable spacial visualizations that they show in the demo are 1) pretty easy replicated in a traditional desktop environment and 2) not really efficient ways to interact with complex datasets.

I think that the user interface is, as many have said, of limited use. For most tasks, including most spatial tasks like CAD work, freespace manipulation seems very slow compared to current methods, not to mention more demanding of precision motor control. Looking closely at the demo, it appears that not much actual freespace manipulation is going on anyway. The tasks displayed mostly composed of pointing (albeit multi-surface pointing) and a gestural command language.

I think it would probably be most effective for collaborative use, which is one area where today's dominant metaphors are least effective. I also think that without some kind of haptic feedback, it will never be transformative, even for the tasks that it is suited for.

Some if their key premises, such as "Gestural input is measurably more efficient at performing complex navigation, sorting and selection tasks." seem untested. I'm interested to see what the empirical human computer interaction community thinks about this system.