If we could access the brain outputs that control the hands, why would we make robot hands and special input devices to be controlled by those hands to control the computers, rather than interface to the computers directly?

Another issue could be that repeating simplistic patterns over and over with your brain could generate a kind of "brain RSI" and you'd prefer the old school way to do things.

I find it infinitely entertaining to think about all the practical things you'll care about when a dream technology actually becomes available.

A major obstacle to developing a useful brain-computer interface using EEG is signal strength. The event-related potentials measured at the scalp are microvolts and do not play well with other electronics nearby. Direct measurement of the motor cortex solves this problem, but most people aren't eager for brain surgery. The other option is to move downstream to the muscle site where a clean signal is quite accessible. I agree, however, that this is a less than ideal solution.