Serious games are pedagogical tools used in many domains and considered a viable solution to professional training. Core elements for the construction of effective serious games are the Non-Player Characters (NPCs) which, depending on the quality of the Artificial Intelligence (AI), can make the training scenarios more realistic and engaging. One of the main issues with existing games concerns the limited possibility to edit their content. These tools are usually delivered as closed products with no, or minimal, customization features. The modification of the virtual environment is strongly constrained and the behavior of NPCs is fixed to a predefined set of actions. Therefore it becomes hard for trainers to adapt the training according to different requirements. In this paper we propose a solution for the customization of AIs, aimed at allow non-technical users to edit complex behavioral models through a new high-level approach. The Actors' Programming Environment (APE) has been developed on the base of a user-centered design and a preliminary evaluation was conducted with eight experts in computer science and interaction design, serving as evaluators. The results suggest that the approach is suitable for this context and might prove to be a good basis for future developments.