Co-Constructed or Constrained? How AI Collaboration Tools Reshape UI Design Practice in a Time-Boxed Design Challenge

Paper Title

Co-Constructed or Constrained? How AI Collaboration Tools Reshape UI Design Practice in a Time-Boxed Design Challenge

Publication Info

• Topic area: The impact of generative AI tools on professional UI design workflows and outcomes.
• Keywords: Generative AI, UI design, UX design, FigmaAI, creativity support tools, design fixation, professional workflows, computational analysis, visual similarity, design outcomes.

Background and Problem

• Problem / challenge: While generative AI tools promise efficiency and inspiration in design, they may constrain creativity, reduce ownership, and lead to homogenized outputs. There is limited empirical evidence on how embedded AI tools like FigmaAI affect professional UI design workflows and outcomes.
• Significance: Understanding how AI tools reshape design practices is crucial for aligning them with professional values, fostering creativity, and ensuring their effective integration into workflows.
• Motivation and related work: Prior research highlights the potential of AI in creativity support but identifies gaps in adoption, risks of design fixation, and challenges in aligning AI outputs with professional needs. This study builds on these findings by empirically examining FigmaAI's impact on UI design processes, reflections, and artifacts.

Solution

• Proposed approach: A within-subject study comparing manual and AI-assisted UI design workflows using FigmaAI, involving 16 professional UX designers completing two design tasks under time constraints.
• Novelty:
  1. Empirical investigation of FigmaAI's impact on professional UI design workflows.
  2. Analysis of designer reflections on AI-assisted versus manual design practices.
  3. Computational evaluation of visual and structural characteristics of AI-assisted and manual design outputs.
• Procedure and key techniques:
  • Participants completed two tasks (manual and AI-assisted) with think-aloud protocols.
  • Data collection included NASA-TLX workload assessments, post-task questionnaires, semi-structured interviews, and computational analysis of design artifacts.
  • Computational methods included color composition analysis and LPIPS-based layout similarity evaluation.

Results

• Concrete findings:
  • AI-assisted workflows shifted from additive (building from scratch) to subtractive (refining AI drafts).
  • AI reduced mental demand (70.94 to 48.44, p = 0.021) and effort (70.00 to 41.56, p = 0.030) but did not significantly affect frustration.
  • AI-assisted designs showed greater visual similarity (e.g., Task B LPIPS: AI mean = 0.33, manual mean = 0.39, p = 0.0040).
  • AI outputs were less diverse in color and layout compared to manual designs.
• Advantage over baselines:
  • AI accelerated routine tasks and reduced workload but introduced risks of design fixation and reduced creative ownership.
• Experiments / evaluation:
  • Conducted with 16 professional UX designers using FigmaAI.
  • Tasks included designing a pizza customization screen and a language-learning app profile page.
  • Metrics included NASA-TLX scores, qualitative reflections, and computational analyses of design artifacts.
• Limitations and future work:
  • Small sample size, skewed toward early-career designers.
  • Limited generalizability due to the beta state of FigmaAI and time-boxed tasks.
  • Future work should include longitudinal studies, comparisons across GenAI tools, and evaluations of broader UX workflows.

Summary

This study investigates how FigmaAI, a generative AI tool, reshapes UI design practices. AI-assisted workflows shifted from additive to subtractive processes, reducing workload but narrowing creative exploration and ownership. Computational analyses revealed greater visual similarity and reduced diversity in AI-assisted outputs. While AI tools enhanced efficiency, they introduced challenges like design fixation and mismatched expectations. The findings highlight the need for GenAI tools that support divergent exploration, maintain creative ownership, and align with professional workflows, paving the way for more effective integration of AI into design practices.