The usage of artificial intelligence (AI) techniques in digital games is currently facing a steady need of improvements, so it can cater to players higher and higher expectations that require realism and believability in the game environment and in its characters' behaviours. In order to ease the fulfillment of these goals, software engineering techniques and methodologies have started to be used during game development. However, the use of such techniques and the creation of AI middleware are still far from being a generic and flexible enough tool for developing this kind of application. Another important factor to be mentioned in this discussion is the lack of available literature related to studies in this field. This dissertation discusses the research effort in developing a flexible architecture that can be applied to diferent game styles, provides support for several game AI functionalities and serves as basis for the introduction of more powerful techniques that can improve gameplay and user experience. This work presents: design issues of such system and its integration with games; a study on AI middleware architecture for games; an analysis of the state-of-the-art in the field; and a survey of the available relevant literature. Taking this research as starting point, the design and implementation of the proposed AI middleware architecture was conducted and is also described here. Besides the implementation itself, a study on the use of design patterns in the context of the development and evolution of an AI framework for digital games is also presented.