This article describes an ongoing project that aims at developing a strong AI-based artificial opponent for some wargame-type simulations. It describes the two approaches taken so far. The first one relies mainly on an incremental and automated learning of the complex knowledge necessary to devise sound tactics, but needs a hi,a-quality coaching which itself requires in-depth knowledge of the domain. The second approach is an attempt at modeling the decision making process that goes on at the various levels of a entire army as some form of complex and knowledge-intensive problem-solving. Future work will look at the possibility of hybrid methods as well as at some alternative methods such as case-based reasoning.