Well, why should we limit the range of possible applications of the programming-game concept to the examples given in the past days? Other possibilities are available, such as the genre of Real-Time-Strategy games.
The paradigm harvest-build-attack is simple enough to be mastered by our intelligent bots, also known as agents. Your possible objection may be based on the number of the units: if one has to assign a virtual CPU to each unit, their computational power must be very limited. In this case, however, I have in mind something different: each player will program the AI of his army as a whole, then the computational time has to be shared among several units. This means that a single thread, the general, will command a number of units and receives feedback from them, leaving some of them, during time consuming activities (harvesting), in a sleeping status. The working unit will awake only for short periods, for example to control the nearby environment for threats. The approach may seem complex, but it's not as difficult as some programming-games are and... no general says that an army is easy to master.