Workflow-Net: Toward Understanding Designer Workflows in Generative AI-Driven Systems through Comparing Node-and Prompt-Based Interfaces

Abstract

Background Generative artificial intelligence (AI) is increasingly integrated into design practice, yet how these generative AI-driven design support system interface frameworks shape the designers' workflow remains underexplored. To address this gap, we formalized and compared three generative AI-driven interfaces: Agentic Prompt-Based (AB), Programming Node-Based (PB), and Creative Node-Based (CB). Studying their influence requires methods that capture workflow dynamics beyond micro-level actions. Existing approaches such as linkography or workflow graphs (W-graphs) focus on words, concepts, or artifacts, limiting the analysis of high-level actions and cross-user patterns. Methods This study introduces Workflow-Net, a novel evaluation method that uses large language models (LLMs) to cluster structured protocol data based on semantic intent to map high-level design actions and to aggregate individual workflows into a comprehensive, weighted directed graph. A within-subject user study was conducted with nine designers, where participants performed three distinct design tasks across all three interfaces to capture multi-user, cross-interface behavioral patterns. Results Findings show that interface frameworks do not merely support design but strongly influence designers' behavior by structuring the cognitive arc. AB supported initial ideation but limited refinement due to text abstraction gaps and a lack of iterative detail control. PB offered precision and granular control but enforced rigid linearity, extreme convergence, and the highest cognitive load. In contrast, CB best supported the creative process by enhancing designer agency and satisfaction, effectively balancing exploration with refinement through automated process traceability. Conclusions This study establishes that the interface framework is a structural determinant of the creative workflow. While AB and PB interfaces impose significant tradeoffs through abstraction gaps or high cognitive load, the CB interface emerged as a balanced model that fosters higher designer agency. Beyond evaluation, the Workflow-Net methodology offers a foundation for developing future hybrid, agent-assisted co-creative partners that adapt to the fluid dynamics of the designer's workflow. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http:// creativecommons. org/licenses/by-nc/3.0/), which permits unrestricted educational and non-commercial use, provided the original work is properly cited.