AI can draft a screen — heuristics still catch the flaws
Published 10 September 2026 · Source dated 1 January 1995
Nielsen and Molich's ten usability heuristics are nearly 30 years old but remain the fastest sanity check for anything an AI agent generates for you.
Jakob Nielsen and Rolf Molich's ten usability heuristics have been the go-to sanity check for interface design since the mid-90s, and NN/g still teaches them as the fastest way to catch obvious problems. That hasn't changed just because an AI agent generated your screen instead of you.
If anything, heuristics matter more now: tools like Figma's agent or Make can produce a plausible-looking layout in seconds, but "plausible" isn't the same as "usable." Visibility of system status, error prevention and consistency catch the most AI-generated slip-ups — missing loading states, no undo path, inconsistent button styles across screens.
This isn't a new framework to learn. It's a 30-year-old one that's suddenly more relevant.
Coach
Next time an AI tool generates a screen for you, run it past 3 heuristics — system status, error prevention, consistency — before you ship it.
Jakob Nielsen's ten usability heuristics have been the backbone of UX critique since 1994. They're worth relearning properly, not just skimming the list.
AI tools can generate screens fast, but they don't validate usability — that's still on you. Classic heuristic evaluation is more relevant, not less, now that production speed has gone up.
More first drafts are arriving via AI agents, but the fundamentals of a good design critique haven't moved. IxDF's literature on critique still holds regardless of who — or what — made the first version.