Industrial design felt safe from AI longer than graphic design did, because matter is expensive to iterate. Why that moat is drying up — and what happens when it reaches the most physical corner of design.
A while ago I started working with a manufacturer that has spent more than forty years making point-of-sale displays: the stands and units you see in any supermarket aisle, the ones holding a bottle at exactly the height of your hand. I arrived with my tools: a repository, deploys that take seconds, the freedom to get it wrong forty times before lunch without it costing me a cent. They came from another world. One where getting it wrong costs.
That was the first thing I understood standing in that workshop: not all design iterates for free. Atoms resist.
When you design on a screen, error is cheap. You try a grid, it doesn't work, you try another, and another, and none of those attempts left any residue. In the physical world it's different. You want to know whether a leg holds the weight of twelve bottles — you don't deduce it from a render: you cut, you fold, you assemble, you load it, and you see if it comes down. Every iteration is material, machine time, hours, money. The person who designs things that weigh something doesn't fail fast. They fail expensively.
The moat was matter
That's why, for a good while, industrial design felt safe from this wave.
AI tore through what could be iterated on a screen first — the graphic, the editorial, everything that lived in pixels. But faced with an object you have to cut, die-stamp, assemble, and that has to withstand a real force in the real world, it seemed to stall. The friction of matter worked like a moat. The person designing physical things could watch the fire from the far bank and think: this doesn't reach me, my work weighs something.
And they had a point. Generating a pretty image of a chair is not designing a chair that holds someone sitting on it. Between the sketch and the part there was a chasm of physical tests that no model spared you. That chasm was the protection.
But the moat is drying up
What changed isn't that AI now "understands" matter. It's that the expensive part — the iteration — moved from atoms to bits.
Generative design hands you hundreds of structural variants of a part from a few rules: how much weight, how much material, which support points. Topology optimization dissolves a solid form into a structure that uses half the material and holds the same load — those parts that look like bone or coral, the ones no human would have drawn by hand. Simulation tells you whether it breaks before it exists. And only at the end, once you've chosen, do you commit any matter at all.
It's the same move I lived through on the web, in a different field. The deploy collapsed the distance between imagining a system and having it running. Here, simulation and rapid fabrication collapse the distance between the sketch and the part. The chasm didn't disappear: it moved onto the screen, where iterating is cheap again.
The same gesture, in the most physical corner
If you read the first part of this series, this will sound familiar. The industrial designer stops being the hand that draws the part and becomes the one who directs the system that generates many, and chooses among them. Duchamp shows up again: the craft isn't in producing the form by hand, it's in the judgment to point at which one of all the forms that appeared is the right one.
And here's the point that matters most to me: if the redirection reaches all the way here — to the most physical, slowest, most expensive corner of design, the one with the deepest moat — then it reaches everywhere. Industrial design was the alibi of "AI doesn't touch me because mine is real." That alibi is over.
What doesn't get replaced
But be careful, because the easy conclusion would be the wrong one. In that forty-year-old workshop, no one is redundant.
Generative design throws you a thousand variants; not one of them knows that this cardboard absorbs moisture in a warehouse, that this corner will get knocked in transport, that the client is a major brand and the unit has to be assembled in thirty seconds by someone who's never seen it. All of that — the knowledge of the material, the process, the real-world use — is exactly what becomes valuable. It stops being manual work and turns into the judgment that steers the tool. The person who has spent decades knowing how it folds, how it holds, how it breaks, doesn't become obsolete: they end up at the wheel.
Matter still resists, just like it did forty years ago. The difference is that now someone who truly knows it can fight back with the whole twenty-first century on their side.
Designer of what? Of what weighs something, too.