For most of my career, the limiting factor in product work was the cost of making things. That constraint has collapsed, and most design organisations haven't noticed what it took with it.
AI generates interface, flow and copy faster than any team can. A capable model produces in an afternoon what used to take a sprint. The common response has been to bank the productivity and carry on, as though design were the same discipline with a faster pencil. It isn't. When execution becomes abundant, value moves to whatever stays scarce, and what stays scarce is judgement.
i.What judgement actually is
Taste feeds judgement but isn't the same thing, and judgement certainly isn't opinion, which anyone produces for free. It's the ability to make three calls correctly and keep making them: which problem is worth solving, whether the assumption under a feature actually holds, and what a system should never do. Each call is cheap to state and expensive to get right, because each is only as good as the evidence behind it.
That last part is where most of it lives. An opinion checked against nothing is just a confident guess. Judgement earns the name when it's tested, continuously, against real users in real contexts. Which is why research doesn't matter less as AI absorbs execution. It becomes the fuel for the one thing that's still hard. When the cost of producing a screen falls to nothing, the cost of producing the wrong screen doesn't. It compounds.
ii.The layer, not the department
I've stopped describing Product Design as a function that produces artefacts, and started building it as a judgement layer: the part of the organisation that decides what is worth making, holds coherence across far more generated output than any team could review by hand, and carries insight through the resistance a shipping organisation naturally puts up. The artefacts still exist. They're just no longer the point, the way typing was never the point of writing.
This changes what you hire for, what you measure, and where a designer sits when a decision gets made. A judgement layer sits upstream of generation. It writes the rules a system generates within, rather than redrawing whatever the system hands back. It's accountable for outcomes it can defend with evidence, not for volume.
iii.What this asks of design leaders
Three things, none of them comfortable.
Dismantle the production identity before the market does it for you. A team measured on screens shipped is a team measured on the thing that just became free.
Build the evidence loop. Judgement without calibration quietly decays into people being trusted because they're senior, not because they're right.
Learn to say no with a mechanism attached. Not “this feels wrong” but “this fails the constraint we agreed, and here's the user evidence that set it.” The first is an opinion. The second is the job.
The discipline that does this well will matter more in the AI era than it did before. The one that keeps polishing output will find out that output was never the moat. ✳