The author experiments with TypeSafe's Jev, a decision-making AI model, to play the classic game Manic Miner. Jev shows mixed results compared to simpler algorithmic approaches, struggling to understand spatial relationships from text descriptions of game maps despite receiving ASCII representations and factual information about objectives.
A developer describes an LLM-based system for creating intelligent NPCs in a 3D action game that responds in under two seconds. The system uses an event-driven behavior pipeline that converts game state into text prompts, leveraging long-term and working memory to generate short-term action plans. Inference speed is optimized through compact YAML prompts, summary folding of event logs, and streaming of action outputs.