Now there's an angle that's definitely interesting. I wonder, though, at what point we will have to start "training" our AI's to do things for us in a completely new manner. For example, the only training AI's today (in the form of bots as they exist today, at any rate) is just the coder trying to catch all the possible decision trees that a user might run into, coupled with the ability to navigate obstacles and areas depending on where the pre-programmed static "task points" are located.

Now, what about tomorrow when AI really is a form of AI and not just some shambling conglomeration of pattern-matching batch files? That, I think, is where things get really interesting. And, probably by nature, much more difficult to spread around for other people to use.

Of course, it remains to be seen how fast we advance AI's as I can see quite the backlash from the first few successful ones being trained to trade stocks, for example. :P