Humankind is a series where we ask the questions every other AI conversation skips: not what changes in how we work or how we design, but what changes inside ourselves.
Every designer with enough years of experience in our industry carries a private catalog of paths not taken. I still think about the ideas I killed. Not because it was wrong, but because I was tired, the deadline was the next morning, and the other option was safer. The layout that almost survived the second review. The direction abandoned three days before the deadline, for reasons that still aren't fully articulated, only felt. This catalog rarely gets shown to a client or a stakeholder. It lives underneath the final decision, invisible, and it is exactly what makes the decision worth trusting.
Nobody arrives at their first design job doubting their first idea; they defend it, usually too loudly. Doubt comes later, after enough certainties collapse in front of others, and after enough obvious answers turn out to be the wrong ones. Research on design education has identified tolerance for ambiguity as a genuine skill, one that develops over time rather than being innate. It’s like a scar you're grateful for.
A generated answer arrives with no history of hesitation attached, though. It has no discarded version, no route abandoned three days before a deadline. It offers one tone for everything, the same fluency whether it is naming a city or inventing one. A 2025 study out of Carnegie Mellon found that AI systems remain overconfident even after performing poorly, unlike people, who tend to recalibrate their certainty once they see the results.
We tend to think authorship is about producing, but it’s really more about choosing. The person who made the design was never simply the one who generated an option; it was the one who sat inside several possible directions, uncomfortable, unresolved, and picked one for reasons they could stand behind. That discomfort was the work. Not a phase of the work. The work itself.
When a machine gives you options that are already assembled, fast, confident, and indifferent to which one gets picked, the temptation is to skip the sitting-with and go straight to the choosing. But a choice made without occupying the doubt first is only a selection. Nobody owns a selection the way they own a decision.
The instinct now is to resolve everything a little too quickly, because a fluent, confident answer is always one prompt away. I do this too. I catch myself reaching for the prompt before the question has finished forming. It's faster. It feels productive. I'm not sure it's the same thing as thinking. A designer who deliberately holds a question open a while longer, who refuses the first plausible answer even when a faster one is sitting right there, is doing something that looks like inefficiency and is actually the opposite. It is the last part of the job that a system cannot perform on someone's behalf because it requires, for a while, actually not knowing on purpose.
I wrote most of this article in one sitting, and I'm not sure I trust it yet. Which might be the point. Whoever still knows how to sit inside not knowing is the one still doing the design.