Game NPCs are usually built around scripted behaviors and predefined dialogue, which can make it difficult for players to communicate with them in a natural way. This talk introduces the MindSand Behavior Engine, a real-time Generative AI framework that allows players to give voice commands to NPC teammates and receive more context-aware responses. The system combines live game state, world knowledge, teammate personalities, and in-game perceptions to interpret a player’s intent. An LLM then turns that intent into a coordinated plan for the squad, including both dialogue and executable actions. The game’s existing systems continue to handle movement and other low-level mechanics, allowing the AI to focus on tactical reasoning rather than replacing the underlying gameplay framework. Using Starship Troopers: Extermination as a case study, the presentation will show how this approach can support natural communication, squad coordination, environmental problem solving, and dynamic positioning. It will also cover how these capabilities can be added to a commercial game without requiring a complete rewrite of its existing systems.