Thinking about it in pragmatic terms, the system will need to distinguish between thoughts and normal brain activity and commands, intended actions.
So, the interface will require a trigger, to know when to listen to commands, this is an interesting challenge.
On many other interface types, there is always a feedback mechanism, to assure the user the action was completed, gesture was recognised, etc
With thought control, what would be the feedback mechanism? Sound? Will there be one? Will the person trust the system blindly and just fire away actions?
We're talking about abstract actions here, moving an arm provides direct visual and perhaps sensor feedback, but taking a picture, sending a text is a bit different.
If feedback come into play, thought interfaces will be, at least initially disruptive, you won't need to move, but you will need to concentrate in it and friction will perhaps be enough to distract you from your main task.
Another use case for thought interfaces is the passive monitoring of the users thoughts, thinking about how good yesterday's Affogato was? Well, Facethought will let your friends know you'd love them to bring you one while you're inside Small Batch Coffee.
Cultural barriers are obvious with this latest option, people won't feel comfortable with this kind of urges being shared and perhaps don't want to manage them because of their nature.
Another thing to consider is a hybrid interface, technology like Disney Research' Touché could provide a way to reliably denote intention in a very frictionless way such as touching your chest twice, blinking - which perhaps can be easily identified by the thought interface as well.
So, here's to imagination.