Bounded Adaptive Agency in Intelligent Products: Designing Scope, Control, and Non-Intrusion in Human-AI Interaction

Everyday intelligent products increasingly use adaptive components whose actions are constrained by capability, context, uncertainty, and privacy. We introduce bounded adaptive agency, a design framework for making these constraints understandable and controllable. Study 1 interviewed 28 designers and related specialists, identifying four interdependent boundaries: capability, deployment, agency, and interruption. Study 2 tested a boundary-oriented design canvas in a controlled concept-generation task with 54 participants, examining how it shaped early decisions about scope, uncertainty, autonomy, recovery, and interruption. Study 3 evaluated paired adaptive-product scenarios with 462 valid participants, comparing a bundled boundary-oriented configuration with high autonomy while holding product goals and nominal sensing inputs constant. Across the studies, boundedness was most useful when it clarified capability, authority, timing, and recovery without sacrificing product value. These findings offer a traceable framework for calibrating adaptive behavior in human–AI interaction and intelligent-product design.