Show GN: civStation – Civilization VI를 전략 수준에서 제어하는 computer-use VLM 기반 에이전트(Human in the...

civStation demonstrates how Vision Language Models can decompose high-level strategic intent into automated UI-based game actions without API integration, revealing both the potential and limitations of autonomous agents in unstructured visual environments. The Human-in-the-loop architecture enables real-time human intervention during execution, moving beyond pure automation toward collaborative human-AI systems. This approach offers crucial insights for designing autonomous agents that must operate solely through visual interface interaction.