This is a very interesting interface bordering on brilliant in the development of focus and flow. As several others have noted this is not really an OS but something more than a GUI. It could certainly be built on Linux initially.
More in my domain, I’m both interested and concerned about how this would evolve and scale. We know the desktop interface is a legacy becoming a relic as only Microsoft is seriously working on it and, given the just announced (kind of) abandonment of the unified Windows environment, is recognizing that the base concept is no longer relevant.
The Mercury concept would seem to be excellent for AR using glasses with gestures and voice input. The module and flow plus design concern for messiness and limitations of distractions seems to be equally valid for this type of interface.
And, as mentioned, the module and flow structure can be partially or completely replaced by AI components. That is, I think, where this needs to be taken. Many of the criticisms listed in responses here could be addressed by fuller automation even with simple auto scripting.
I will certainly watch for further development of Mercury.