How Agent-Driven Creative Trace Mining Reshapes Design Behavior in a Deployment Study of Intent-Aligned Multimodal Design

Designers continuously capture inspirations across media, yet these creative traces remain underutilized in computational design support. We present DesignJournal, an agent-driven system that mines multimodal creative traces and transforms them into intent-aligned image generation. The system consolidates artifacts into a unified workspace and employs an agentic pipeline for structured grouping, belief-guided prompt refinement, and iterative intent updates. We developed a framework that classifies sequential design actions, deriving a hierarchical action taxonomy from interaction logs. We used it to examine how agent affordances reshape design behavior. A 10-day deployment study with eight professional designers compared baseline and agent-supported conditions across 5838 classified actions. Agent support significantly reduced the frequency of spatial repositioning actions, particularly small-distance movements, while sequential transition structures shifted from extended repositioning chains toward relational encoding sequences. Awareness of agent interpretation shaped upstream recording behavior, and agent-mediated collaboration extended designers’ creative reach beyond independent exploration.