In architecture school, I had a different kind of professor. His name is Carlos Puente, and in 2022 he received Spain’s Gold Medal of Architecture.
Back in those university days, we all talked about Foster, Rogers, Koolhaas. Steel and glass buildings, structures that looked impossible, magazine covers. The geniuses of the moment.
Puente would turn off the lights and start the projector. Real slides, from a carousel projector, in black and white. One day, a path of stones in Athens appeared on the screen. No skyscrapers, no big cantilevers. Just a floor, some benches and a temple at the end. And he told us that this was also architecture. And the good kind.
I was fascinated. Not because Foster and his contemporaries were wrong, but because suddenly architecture was much bigger than I had thought.
Today I often think about those slides, but for a different reason. I believe that path explains very well how we should design with AI. Because AI, like the author of that path, does not invent a single stone. It works with fragments of what already exists. The question is not whether the result is original, but who decides where each piece of that pavement goes.
A path made of leftovers
The path was the work of Dimitris Pikionis. Between 1954 and 1958, he designed the routes that climb to the Acropolis and to Philopappos Hill, right across from it.
To see it, Mayte Piera’s photographs on Metalocus (opens in a new tab) show it better than I can.
What I find interesting is where the material came from. Athens was tearing down many neoclassical houses to build new apartment blocks. Pikionis collected the remains: marble slabs, stones, pieces of cornices, bits of ceramic. With them, he paved the paths.
But he did not leave it to improvisation. Pikionis drew the grid that generated the layout. He did not draw every stone on a plan (not even in Figma): he drew the system. Then he spent his days on site with the stonemasons, choosing piece by piece where each one went. The result is a floor where almost nothing repeats, but that reads as one single work.
This drawing illustrates Pikionis’ philosophy: turning the architectural walk into a sensory experience. Instead of imposing a rigid grid on nature, he used geometry as an invisible tool to orchestrate the visitor’s eye, the views of the historic landscape and the rhythm of their steps along the route.
Drawing of the generative grid of the path, 1954-1957. © Benaki Museum Athens / Modern Greek Architecture Archives, Dimitris Pikionis Archive. See the drawing on Arquitectura Viva (opens in a new tab) · open the image (opens in a new tab).
Someone else’s material, his own result. Nobody who walks there thinks they are stepping on rubble.
Everything had its moment
What Puente wanted us to see was not the pavement. It was the context.
Every part of the climb had its moment. A place to rest in the shade. A point where the path opens up and, suddenly, the temple appears. A narrower stretch that makes you slow down. Some steps that change the rhythm just before you arrive.
Nothing was there just to be there. A stone that works in one corner would not make the same sense ten meters further up. The value was not in the stone. It was in knowing where to put it, and that is only possible if you know the place: the slope, the views, the sun, the people who walk up.
Pikionis was not designing a floor. He was designing an approach: a route toward a specific goal.
AI also builds with rubble
A language model does not invent from nothing. It learns from everything that already exists and recombines it. Generative UI works in a similar way: the agent does not create new components, it combines the ones it already has in the design system.
In that, AI is very much like Pikionis. It works with fragments.
The difference is that Pikionis knew where he was. He knew every curve of the hill. An agent, if we don’t give it context, knows nothing about the place. It doesn’t know who the user is, what just happened, what they are in a hurry to do, or where they need to stop and think. So it puts correct pieces in any place.
In a recent article, I called it the Empty Neighborhood Problem: everything meets the rules, but nobody thought about the people who would live there. With fragments and no context, you get an empty neighborhood. With fragments and context, you can get a Pikionis.
That is why context is not a technical detail. It is the place where we are going to build.
But context alone is not enough. Pikionis also had his grid, and the rules did not take away variety: they gave coherence.
That is what an AI-ready design system should be. Not just a catalog of pieces, but the rules for combining them according to the place. With pieces, rules and context, the agent can build something that reads as one single work.
The orchestrator is also the site manager
Pikionis did not place every stone himself. The stonemasons did. But he was there, watching, correcting, deciding. Without him, the stonemasons would have made a good pavement. With him, they made a path.
The stonemasons brought their craft, what we would call skills today: knowing how to set a slab, how to fit it, how to finish it. Pikionis did not have to explain it every time. He had to decide where.
With agents, the designer’s role looks more and more like that. We no longer draw every screen pixel by pixel. We prepare the place, give the context, set the rules and review what comes out. We are orchestrators.
But orchestrating is not delegating and forgetting. An architect who only works in the office is not a good architect. The judgment is still ours: which piece goes here, which one is left out, where to stop, and where to prepare the space to look at the temple.
AI can move a lot of stones very fast. Deciding where each one goes is still design work.
This is also design
Today, the big AI launches look like the Foster and Koolhaas covers from my student days. Spectacular demos, a lot of noise, everyone looking in the same direction. It is good that they exist.
But what will really make the difference is less visible. It is the work of someone who knows the “place” and decides where each piece goes.
Maybe in a few years, a professor will turn off the lights and show their students an interface made by agents. And they will say that this is also design. And the good kind.
Whether it is good will not depend on the stones. It will depend on whether someone thought about the path before placing the first one.