The Seat and Shape of AI UI: Exploring How UI Affects Designers’ Workflow and Trust in AI

Abstract

As artificial intelligence (AI) becomes increasingly integrated into a designer's workflows, its effectiveness may depend not only on algorithmic performance but also on how it is embedded within the user interface (UI). This thesis explores how the design of AI user interfaces shapes designers’ workflows and trust when collaborating with AI tools. While recent advances in generative AI have rapidly expanded designers’ access to AI capabilities, less attention has been paid to how the interface mediates this collaboration. Through a diary study and contextual interviews with ten professional User Experience (UX) designers, this study examines how designers integrate AI tools into their day-to-day design processes and how different interface patterns influence their interactions with AI. Using reflexive thematic analysis, the study identifies five common types of AI interfaces that designers encounter in practice: conversational, canvas, contextual, embedded, and one-click interfaces. The findings show that designers actively experiment with AI tools and strategically choose or switch tools based on perceived strengths, memory capabilities, and organizational constraints. However, designers consistently value AI features that integrate seamlessly into their existing workflows, as frequent tool switching introduces cognitive overhead. The results also reveal that different interface types support different levels of task complexity. Designers prefer iterative interactions for complex tasks that require context sharing and refinement, while quick and lightweight interactions are favored for simple tasks. The study further shows that designers rarely treat AI outputs as final work. Instead, they use AI as a thinking partner, where trust develops gradually through ongoing interaction, transparency, and opportunities for user control. Overall, this research highlights that the user interface is not merely a surface layer for AI capabilities but an active mediator of human–AI collaboration. By identifying common interface patterns and their relationship to designers’ workflows, this work provides insights for designing AI systems that better support effective and trustworthy collaboration between designers and AI.