Bounded Delegation: How Student Design Teams Keep Strategic Control over Generative AI

Generative AI is increasingly embedded in team-based design, yet little is known about how teams negotiate its role over extended projects. We conducted a four-week situated study with nine graduate design teams (54 students) developing complete murder mystery Live Action Role-Playing games. Using domestication theory as a sensitizing framework, we analyzed 104 human-AI collaboration events, project outputs, and 27 semi-structured interviews. We identified three collaboration modes: Task-Based Empowered Creation, AI-Augmented Creative Co-creation, and Progressive Trust-Informed Collaboration. A fourth mode, Ceding Executive Trust Collaboration, was theoretically possible but did not occur. Across teams, AI was delegated tactical tasks such as content and asset generation, while humans retained control over narrative direction, quality standards, and core mechanics. We conceptualize this pattern as bounded delegation within a strategic-tactical hierarchy, offering an alternative to linear autonomy-progression models and informing AI tools that support rather than supplant human strategic control.