Kathryn Grayson Nanz is a Senior Design and Developer Advocate at Progress Software who specializes in making complex software understandable, accessible and useful. Her background in graphic design, front-end engineering and user research informs a distinctive approach to AI products: powerful systems need interfaces that let people verify their outputs, control their actions and apply their results.
From graphic design to developer advocacy
After earning a fine arts degree in 2013, Nanz began her career in graphic design before moving into front-end development. At one startup, she split her responsibilities between design and engineering and helped introduce usability testing without a dedicated UX specialist.
Her interests expanded from interface implementation into the organizational challenges surrounding software design.
In August 2021, she joined Progress as a developer advocate for KendoReact, moving into technical education and developer-community work. Reflecting on her first year in developer relations, she described the challenge of measuring work that includes teaching, community conversations and helping developers feel more confident.
Her writing spans React, design systems, accessibility and collaboration between designers and engineers. Her guide to accessibility in React addresses semantic HTML, the accessibility tree, component rerendering and single-page applications. In 2024, she examined the designer-developer handoff, arguing for earlier engineering involvement and shared ownership throughout product development.
Designing AI around human agency
At AI Engineer World’s Fair 2026, Nanz organized her approach to human-centered AI design around trust, clarity, control, transparency and meaningful benefit:
Familiar patterns, adapted for AI. Conversational interfaces can borrow messaging conventions, but they also need citations, interruption controls and clear mechanisms for agent actions. Nanz compares this gradual introduction of unfamiliar capabilities to the evolution of early Macintosh interfaces.
Verification before trust. Generated responses should link to inspectable sources, identify AI-created content and make proposed agent actions visible before execution. Her framework for trustworthy AI interactions treats verification as more valuable than unsupported promises of accuracy.
Control proportionate to the stakes. Users need prominent stop controls, reversible actions, targeted edits and, when appropriate, version histories or checkpoints. Her approach to user control distinguishes simple conversations from consequential document editing and autonomous workflows.
Granular, revocable permissions. People should know whether an agent can read, send, modify or delete information; whether authorization is temporary or persistent; what it remembers; and how to revoke access. Transparent AI permissions should also disclose anticipated time and cost.
Outputs that become useful work. Streaming responses, highlighted changes and visible progress keep users oriented; suggested next steps and integrations help them turn generated material into documents, messages, code or other practical outcomes.
