I’ve begun to earnestly dig into what it means to actually design for GenAI user experience, like a GenAI tool or feature (as opposed to designing with a GenAI tool like Replit or Lovable as a co-creator). I’m quickly realizing there’s a set of important concepts to understand, like how “fluid grid”, “breakpoints”, and “media queries” are vital concepts for responsive web app design.

Designing for GenAI isn’t just about adding a prompt box! 🤨

It’s about handling prompt latency, model probability, token limits, and elastic degradation, via clear interface & interaction design, while still creating experiences that feel coherent & trustworthy. Perhaps a bit dramatic to say, but we’re sorta designing in a world of “controlled chaos”. Patterns are fast emerging, but here's what to watch for…so far! 😅


✦ Update: I recently spoke at the AI Experience Salon in June 2025 sharing my framework on the “Craft of AI” involving four vectors of shaping effective AI experiences. No Zoom recording sadly, but I published a Substack post explaining all.

image.png


⚖️ Part 1 — Design for trustworthy AI tools

Let’s fully acknowledge it — yes, there’s tons of gender/racial bias and stolen/unpaid content for various LLMs. But I’m not talking about that here, in that broader socio-political-economic context. Much smarter folks with legal & political credentials can argue that. 🙃

Instead, I’m primarily focused on how to even convey “trustworthiness” to the user in a structured, legible, reliable manner.