The value
in the loop.

The floor is moving

AI is rewriting how product design gets made, valued, and bought, all at once. The derivative work the last two decades rewarded, wireframes, variants, generic pattern-matching, best-practice production, is exactly what generative tools now do faster and cheaper than any team. The economics of paying a human for it are already gone. That is not the scary part. The scary part is mistaking the disappearance of that work for the disappearance of the value.

What the machine can't do

A model can generate a serviceable interface from a set of principles. It cannot decide what a person needs to see, in what order, to make a high-stakes call quickly and stand behind it. It cannot hold a position against a room of stakeholders who each want the product to work their way. It cannot tell you which recommendation is worth acting on and which is noise dressed as certainty. Judgment, taste, and accountability for a decision are the parts of the work that do not commoditize.

We are the value in the loop.

Especially where the stakes are real

This matters most in the products where a wrong call costs something. An operator triaging AI-flagged signals. A clinician checking what an algorithm surfaced before signing the report. An analyst deciding whether a score is worth acting on. The interface is not there to be expressive. It is there to make the human faster, more confident, and more right at the moment of decision. Designing that moment, the handoff between what the algorithm produced and what the person decides, is where I have spent the sharpest part of my career, and it is going to define product design for the next decade.

What that asks of design

It raises the bar instead of lowering it. When the machine takes the derivative work, what is left for the human is harder: information architecture that keeps a sprawling product coherent, workflows built around cognitive load instead of screen count, design systems that outlast their authors, and the research and business judgment to know what should exist before anyone builds it. Less pixel-pushing. More deciding what is true and what matters.

The last era taught us to make interfaces disappear. The next one asks us to make the decisions inside them trustworthy. That is the work I want to keep doing.