"What almost shipped instead?"
Nothing came back. Not a weak answer. No answer. There was no other version. There was no almost.
The candidate had just walked a hiring panel through the strongest banking onboarding case study its head of design had seen in months — clean problem framing, research artifacts, a decision log, before-and-after numbers, presented fluently. If the interview had ended at the presentation, they'd have been talking about an offer.
Then a colleague asked the first question: you sequenced identity verification before showing the user any value, most teams do the opposite — what made you choose that order? A pause, and then the candidate walked everyone through the screen already on the shared display, again.
The second question is where it collapsed. Not because the case study was fake — the interviewer didn't know how it was made and didn't need to. What the two questions established was narrower: the judgment they were hiring for was not in the room that day.
That scene is the whole industry in miniature right now, and it is the trap this piece is about. Everyone in design leadership has quietly agreed on the answer to the AI question, and almost nobody is asking whether the people they're telling to go get that answer were ever allowed to build it.
Last week this column tracked the reckoning between outcome and artifact — the argument that screens themselves are getting demoted as agents absorb the six-step "attach a photo, browse, select, confirm, send" grind into a single spoken sentence. This week the story moves to where the value floor landed after it fell: not on better artifacts, but on judgment and "product sense" as the new hiring criteria. That relocation is being sold as a rescue. I want to argue it's closer to an eviction notice with a kind word attached.
Start with why the artifact stopped carrying the proof.
For twenty years the design career ran on a division of labor: the portfolio, the case study, the deck were expensive to make, so the interview merely verified you'd made them. AI inverted that economics in roughly two years. A polished case study, complete with research narrative and before-and-after metrics, can now be assembled in an afternoon by someone who was never near the project.
So the artifact became the cheapest thing in the pipeline, and the conversation became the expensive one. The proof didn't disappear. It moved — into the live, unprepared moment where you translate your own judgment for a stranger, under pressure, with no artifact to hide behind.
This is where the rescue narrative gets specific and, on its own terms, correct. Helio Interactive lists the judgment layer AI hasn't touched: choosing the right problem, reading context nobody wrote down, taste as a point of view, designing for conflicting incentives. Juicero engineered a $699 juicer beautifully; per Helio Interactive, the packets could reportedly be squeezed by hand, Bloomberg demonstrated it on camera in under a minute, and the company folded a few months later. No AI tells you which situation you're in.
Fine. But how, exactly, does a person acquire that? Here the story turns, because NNGroup has done something unusually useful: it has defined the thing precisely enough to reveal who can and cannot get it.
Product sense, in NNGroup's definition, is not empathy and not creativity and not taste. It's "the ability to recognize when current problems match past successes or failures and reliably estimate how similar solutions will affect the desired outcomes."
And it is built exactly one way — by closing the full experimentation loop. Face problems. Choose solutions. Measure the outcomes. Reflect on the results.
This is the same loop that makes a fire chief able to read a burning building, a NICU nurse able to read a failing infant, a chess grandmaster able to read a board. Rich repertoires of past experience, built by completing all four steps. Not by consuming inputs. Not by doing teardowns. By running the loop until the patterns are earned.
Now hold that definition up against the last decade of actual design work, and the trap snaps shut.
The Head of Design in the E2/E8 account describes what those years rewarded, from inside the performance reviews she sat in. The conversation was "entirely about delivery." How many products the team touched. How quickly designs moved to handoff. Whether the design system was used consistently enough to speed up the next sprint.
Rarely did anyone ask: did we design the right thing? Did we see what was coming before it shipped? Did we prevent anything?
Her sentence is the load-bearing one. "We did not have language for prevention. We had language for production."
We did not have language for prevention. We had language for production — and the profession was too busy shipping to build the muscles that would have made shipping matter.
Read that against NNGroup's four steps and the arithmetic is brutal. Design got very good at step one and step two — face the problem, choose a solution, ship it by Thursday. Steps three and four, measure and reflect, were the sprint's first casualties.
You shipped and moved to the next ticket. You rarely owned the outcome; someone else read the dashboard, if anyone did. And reflection was the thing you promised yourself you'd get back to when things settled down. Things did not settle down. They never do.
So the capabilities didn't vanish because organizations stopped valuing them. In her words, they "atrophied because we stopped practicing them." The profession was "too busy shipping to build the muscles that would have made shipping matter."
Here is the part the rescue narrative refuses to say out loud. The velocity machine didn't just fail to build judgment as a side effect. It structurally severed the two steps that judgment is made of, because owning outcomes and reflecting on them is precisely what slows a sprint.
The greatest professional sin was being the bottleneck — the designer who wanted more research before the thing shipped. A decade of designers were selected, promoted, and paid for optimizing away the exact loop the new hiring bar now requires them to have closed.
That is not a skills gap. It's a manufacturing defect in the career itself.
And it makes "go develop your judgment" a near-empty instruction, because judgment in NNGroup's own account is not readable from books or teardowns — it requires facing real problems, choosing, measuring, and reflecting, again and again, until you've built a repertoire. Repertoires take reps. Reps take runway. The market is not offering runway; it's offering the second question, right now, in the room.
There's a crueler layer still. NNGroup notes that even a real repertoire only transfers in high-validity situations — ones that mirror where the patterns were learned. A designer who fixed several checkout flows can diagnose the next one. Put that same designer on a zero-to-one product with a long feedback loop and the past offers little.
Which means the loops that were closed for the lucky few were mostly closed on repeatable, incremental work — the very work AI now does. The reps that were available were reps in the domain being automated. The reps that would matter now, in genuinely ambiguous problems, are the reps almost nobody was staffed to run.
So the honest version of the advice is not "develop judgment." It's a scarcity question with no comfortable answer: who, in the next hiring cycle, gets the rare assignments where you own an outcome long enough to learn from it — and who gets told they lack a repertoire they were structurally prevented from building?
The candidate in that room may have had years of real practice, like the interviewer himself once did (carrying, by his own account, eight years of real practice) when he reportedly asked a VP what she meant by "design thinking" and watched the verdict close in a final round. The artifact survived. The conversation didn't. Rooms are not truth machines; they're pressure tests, and they fail people carrying real work all the time.
But the room doesn't care about your explanation. It asks what almost shipped instead. And a generation trained to ship and move on, never to weigh the version that didn't survive, is now being asked to produce the one thing the job description never let them keep.
That's not a bluff AI called. It's a bill the profession wrote, addressed to the people who had the least say in the terms.