Descriptions:
Kathryn Grayson Nanz, senior design and developer advocate at Progress Software, delivers a talk at the AI Engineer conference on the distinct UX challenges that arise when building AI-powered applications. Drawing on a background spanning design and front-end development, she argues that the knowledge gap between AI developers and average users is wider than with any previous technology — and that closing it through deliberate UX design is now a competitive necessity, not a nice-to-have.
The talk covers several concrete patterns: showing users an agent’s action plan before execution (a flow already common in Claude and ChatGPT) to build trust before irreversible actions are taken; clearly labeling AI-generated content to manage polarized user reactions; designing for non-deterministic outputs that can vary widely in quality and format; and calibrating the level of abstraction shown to users based on their familiarity with AI concepts. Nanz draws an analogy to the evolution of the Macintosh UI from System 1 to System 6, positioning AI interfaces today at roughly “System 3” — past the stage of the most literal metaphors, but not yet able to assume expert users.
A through-line of the talk is that poorly designed AI features do active damage: each time a user tries an AI feature and gets a sub-par result, they are less likely to engage with it again. For teams shipping AI features to mainstream audiences, the talk provides a practical framework grounded in established UX research and real-world AI deployment patterns.
📺 Source: AI Engineer · Published July 18, 2026
🏷️ Format: Deep Dive
