A short-stacking bot should be an achievable goal. Playing short-stacked really simplifies the decisions: fold, or push all-in. Once you hit an all-in win, leave the table. Perfect strategy for a bot to use.
The problem is that it is also one of the most detectable patterns for pokerstars staff to catch onto.
(damn it... reading this post has made me want to play a sit-and-go instead of returning to my django project...)
It's really not as simple as you say. 20BB (BB = big blinds), the usual minimum starting stack size, is too large for a fold/shove strategy to be profitable. You need to do a lot of limp-reraising and squeezing pre-flop, and you also need to not be terrible post-flop, since, e.g., after a raise of 3 BB with one caller, the pot is 6BB and you still have 17BB.
I don't think this guy knew enough probability/stats/game theory to write a profitable poker bot. Note that he was enumerating a lot of manual decisions in Excel. That's not going to scale, you want a probabilistic model and a game state evaluation function (ideally one trained on many past games).
He got a heads up bot working, which is a totally different animal than a full table ring. U of Alberta has a group discussing poker as an open AI issue : http://poker.cs.ualberta.ca/ NLHE is not checkers, it's genuinely difficult to crack. That's one of the things that makes it fun.
It's worth noting to the ethics committee that PokerStars is technically breaking the law and should have ceased accepting deposits like PartyPoker did after the UIGEA was passed.
The problem is that it is also one of the most detectable patterns for pokerstars staff to catch onto.
(damn it... reading this post has made me want to play a sit-and-go instead of returning to my django project...)