What separates designers who do not use AI from those who do?

Utility, not usability. In the survey Nielsen reports, only one of ten barriers separated the 84 holdouts from adopters: 40% of holdouts found AI insufficiently useful against 16% of adopters, while ease of use scored 12% against 10%. The holdouts are unconvinced it does the job, which in the Deliver phase is a fair verdict on a tool with nothing written to check it against.

What is in a failure inventory row?

The worst plausible output in the product's own words, who gets hurt, what it costs, whether it can be recovered, and one of five responses: prevent, catch, disclose, recover, or accept. Plausible means a normal Tuesday with the system working as designed. Ten rows, three people, ninety minutes, and the list goes in the ticket.

Why not wait for better finishing tools?

Because finishing is already where the hours are. Figma's design chief reported 17x more code and 1.3x more releases, which means generation stopped being the bottleneck and deciding what is good enough to ship became it. A team that writes ten rows per feature now runs AI in Deliver this quarter. One that waits is measured in 2028 the way the holdouts were in 2026.

Why Designers Use AI Less in the Deliver Phase

Why do designers use AI less in the Deliver phase?

THE SHORT ANSWER

Because late errors ship and nobody wrote down what a late error is. Sepideh Tajarmakan and Khashayar Hojjati Emami's survey of 443 designers, as Jakob Nielsen reports it on October 19, 2026, found AI use at 72% in Discover and 41% in Deliver, with only 35% of adopters using it in all four phases and nobody starting with Deliver except two. Nielsen's reading is that early errors are free and late errors ship, so AI is startware until vendors build finishware, which he predicts by 2028. My reading is that the Deliver phase opens the day design writes the failure inventory: ten worst plausible outputs, each with prevent, catch, disclose, recover, or accept. The catches become eval cases, and a generated edit that passes them ships with the same check a hand-made one gets. Ninety minutes, three people, this quarter.

The curve in the survey is real and Nielsen's seven-word explanation of it is right. The wait he prescribes is not, because what makes a late error shippable is not a trusted tool, it is the absence of a written list of what a late error looks like, and that list takes ninety minutes.

The rewrite, with the survey's numbers and the ten rows, is in AI Is Startware, Says Nielsen. It Finishes the Day the Failure Inventory Exists.. The artifact itself is the handbook chapter The Failure Inventory(coming Nov 25), and the number that says finishing is the bottleneck now is in 17x the Code, 1.3x the Releases. Design's Job Is the Spec Now.

Last reviewed 2026-10-20 · 1 min read

Why do designers use AI less in the Deliver phase?

SOURCES

THE LONG VERSION

RELATED ANSWERS