A director of UX research described his world to me recently. Since AI coding tools rolled out across his company in January, the line between design, product, and engineering stopped existing. He called it the wild west. Engineers building screens, shipping them, and calling them designs because they looked like designs.
He wasn't worried about the screens. Most of them were fine.
He was worried about what stopped happening before them.
The constraint moved
When production was expensive, the bottleneck was making the thing. Every process we built sat on top of that constraint. Reviews, handoffs, specs, gates. They existed because rework was costly.
Production isn't the constraint anymore. Anyone with a prompt can put something shippable in front of you by Thursday.
Making it is abundant.
Knowing it's right is not.
That question belongs to research. And it's the one input AI cannot manufacture, because you cannot synthesize a customer. You can synthesize something that talks like one. The gap between those two will cost you a quarter before you notice it.
In the blur, research becomes the load-bearing discipline. It's the only function whose output can't be generated.
Three lines
That only holds if someone draws lines. Three of them, specifically.
Between the researcher and the participant
No synthetic users standing in for real ones. AI in the workflow, never in the room. The moment a model speaks on behalf of your customer, you have a very confident feedback loop pointed at nothing.
At the judgment layer
AI proposes, people decide. AI does the volume. Design judgment does the taste. Ten directions in an hour only helps if someone can tell you which one matters.
At the bar
The standard has to be human-owned, written down, and enforced in review. Not vibes. When generation is cheap and everyone can produce, a written standard somebody defends is the only thing keeping an experience coherent.
What he actually wanted
What that research director wanted from his next leader was someone who would argue for guardrails and keep the team from handing its thinking to the tooling.
That isn't process nostalgia. He works in an org with a real aversion to process, and he knows how the argument sounds when he's the one making it.
He's right about where the risk sits. The failure mode of cheap production isn't bad output. It's confident output nobody interrogated.
The blur is real and it isn't reversing.
Design leadership in this era means protecting the questions that come before the making.
And keeping someone accountable for the one AI can't answer for you.