A designer shares a candid account of using LLMs throughout a complex visual comparison UI project. The LLM excelled at cross-domain research, enumerating failure modes, and rapid concept generation — producing 15 viable directions and 8 prototypes in days. However, the breakthrough came from a human teammate who reframed the core question from 'how do we compare two signatures?' to 'what does the practitioner ask themselves when they look at this?' The key insight: LLMs are powerful for divergence and research but cannot replace the human reframe. A practical 6-step recipe is offered: diverge wide with the model, generate 3-6 concepts, then converge with practitioner-experts to find the right problem statement before finalizing.
Table of contents
The Problem, in the AbstractWhere the LLM Was a Force MultiplierWhere the LLM Fell ShortThe Recipe I’ll Run Next TimeThe Takeaway294 Impressions