AI drafts fast — usability testing still the check
Published 5 September 2026 · Source dated 30 August 2026
As AI agents speed up how quickly screens get produced, the risk is shipping untested assumptions faster than before. Core usability testing method hasn't changed — it's more necessary, not less.
Nothing about a five-user hallway test is obsolete just because an agent generated the first draft in ten minutes instead of two hours. If anything, faster generation means more untested variants reaching stakeholders before anyone's watched a real person try to use them. NN/g's long-standing guidance on usability testing fundamentals — small samples, think-aloud, iterative rounds — is exactly the discipline that keeps AI-assisted speed from turning into AI-assisted guesswork. This is established practice, not a trend: the method itself predates AI tooling by decades, and it's precisely why it still works as a check on it.
Coach
before you ship your next AI-drafted screen, test it on 3 people who've never seen it — even informally.
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.