I think that having applications in the cloud edit what I wrote is wrong on many levels.
Obviously, trust in the robot provider is the biggest issue here.
Then, bugs in the robot implementation run the risk of completely destructing our data (which is balanced somewhat by Wave’s unlimited undo/playback, IIUC).
Then, this makes auditing content much more difficult, since robot contributions aren’t clearly marked (unless all robots use some common annotations…). Wave proposes the playback feature to discover the trail of changes that led to the current state, but I think that’s a bad idea: I don’t want to watch a possibly very long animation just to see who wrote what (Note: all of this is based on my limited understanding of Wave, which I haven’t tried yet.)
Then, there’s the engineering question of what happens to the robot-generated content when I edit what I wrote. The robot will have to reparse the text, and update its inserted content. But I may have moved, edited, etc, that content by purpose or accidentally. I think that’s just a big stupid mess altogether.