Judd's Objection

There is nothing in the artillery sheds at Marfa that can be justified by a requirement.

A hundred boxes in mill aluminum, each the same overall dimension, none of them the same inside. Two long buildings Donald Judd reworked with quarter-vaults and windows down both sides, so the light moves across the aluminum all day and the boxes stop being objects and become a condition of the room. You walk the length of it. Something happens that no description of it survives, including this one.

I wrote recently that a tool earns its place in the studio if it shortens the loop between intention and evidence. I still believe that. But I keep running the boxes against it, because they are a standing objection: what happens when the intention cannot be stated in advance, and the only available evidence is a body in a room?

Three witnesses

Judd is the sharpest of them because he was also a writer, and a merciless one. He rejected the word sculpture, called the things he made specific objects, and cared enough about permanence to build institutions so the work could stay put in the light it was made for. He was not producing images of ideas. He was producing presence, and presence is not the kind of claim you can check.

Barnett Newman published an essay in 1948 titled The Sublime Is Now. The title alone is the argument. His paintings are sized so that reproduction fails them; you have to stand close enough that the field exceeds your peripheral vision and the vertical band does something to your sense of where you are. That is a spatial effect achieved with paint. Any architect should be interested in a man who got that from a flat surface while we struggle to get it from a building.

And James Turrell, who trained in perceptual psychology before he made anything, and who has spent fifty years shaping a volcanic cinder cone in Arizona so that a person inside it will attend to their own act of seeing. There is no image in a Turrell. The material is light and the subject is your perception. You cannot photograph the work; you can only photograph someone else's turn at it.

Three artists, one shared commitment: the encounter is the work. Not the image of the work, not the description, not the intention. The encounter.

None of it is checkable.

What gets optimized gets substituted

Here is where this stops being an art appreciation essay and becomes a professional problem.

Our work is already dense with things that can be checked. Energy use intensity. Program conformance. Cost per square foot. Egress widths. Certification points. Each of those is legitimate and I have spent years of my life defending them. But measures have a way of migrating from evidence to objective. You begin by measuring what you value, and you end up valuing what you can measure. And the further along a project gets, the more the checkable criteria dominate the conversation, because they are the ones you can win an argument with at four in the afternoon in front of a client.

A building can satisfy every one of those measures and be nothing to stand in. We all know one. We have all, if we are honest, helped make one.

Now introduce a system whose entire competence is applying criteria continuously and tirelessly. It will not invent this bias. It will amplify it, because it makes the checkable half of the work cheaper while leaving the unmeasurable half exactly as expensive as it always was. Any time you reduce the cost of one input and not the other, you get more of the cheap one. That is not a prediction about software. That is just what happens.

There is a second-order version of the same problem. Much of what turns out to be good in a project was never in the brief; it was discovered somewhere around the third scheme, and the brief got rewritten afterward to make it sound intentional. A tool that faithfully holds you to your stated requirements is, in that light, a tool that holds you to your earlier and stupider self. Conformance is a virtue right up until the moment the project needs to become something nobody asked for.

The steelman, which I still believe

And yet. I do not think this is an argument against rigor, and I want to be precise about why.

The unmeasurable part of architecture is expensive. It costs schedule, it costs fee, and above all it costs credibility. You only get to insist on the thing you cannot justify if you have been demonstrably right about the things you can. Nobody grants the poetic exception to the team that missed the budget. In every project I have worked on where something genuinely good survived to construction, it survived because the surrounding discipline was airtight enough that there was room left for one unreasonable thing.

So the loop matters. Tighten it, not to industrialize judgment but to buy back the hours and the standing that judgment requires.

That is a claim about sequence, and it puts a burden on us rather than on the software: if the checkable work gets cheaper and we do not spend the surplus on the unmeasurable, the surplus just evaporates into more checking.

The Leverkühn bargain

Thomas Mann wrote Doctor Faustus in Pacific Palisades, across town from my desk, in a house he had commissioned from the modernist J.R. Davidson in 1941. The novel hands its composer, Adrian Leverkühn, a rigorous rule-governed method, twelve-tone composition, worked out with Adorno's counsel. Then it asks the question our profession is currently talking around: is work produced by system still art, or is the system what you traded for it?

Mann is too good to answer that cheaply. Leverkühn's method is not a fraud. It produces real work, and the discipline of it is genuinely beautiful. The price is not competence and it is not talent. It is warmth: the capacity for the human address the work was supposed to be for. That is a far more precise warning than the one usually on offer about automation. Nobody is about to lose the ability to produce accomplished buildings. The risk is producing accomplished, rigorous, entirely defensible work that has quietly stopped being addressed to anyone.

I find it hard to ignore that he wrote that book in a house he chose for its modernism, in the city where I practice, and that the house escaped demolition to become a place for arguing about exactly this.

Treat the model as material

So what posture does that leave? Jenni Pasanen is the best answer I know. She makes work by way of machine learning, and what makes it hers is not the model's output but her authorship through it: selection, iteration, refusal, a hand trained on a process rather than on a surface.

That is the difference between an instrument and a material. An instrument executes your intention. A material has a grain, resists you, and teaches you something in the resisting. Every architect already knows how to work that way, because that is our entire relationship with concrete.

The failure mode is treating these systems as vending machines, which produces exactly the averaged, precedent-shaped, technically-compliant work you would expect. The alternative is to treat them as a material with a grain: find out where they resist, work against that, and hold authorship over the result. Pasanen is not being served by the model. She is arguing with it.

Two questions instead of one

So I want to amend what I wrote. One test was not enough.

The first still stands: does it shorten the loop between intention and evidence? That is the question about competence, and competence is not optional.

The second is the one Judd asks: does it leave room for what cannot be checked? Because the room worth standing in was never in the requirements, will not appear in the model, and cannot be validated. It has to be argued for by someone willing to be unreasonable in a meeting.

That part isn't delegable. It's the job.