From Structuralism to AI. How Code Became the New Grammar of Architecture. Is architecture a language? This question was asked in the mid-20th century not only by linguists but also by architects—especially those associated with European Structuralism and Japanese Metabolism. Today, in the era of algorithms, parametric design, and artificial intelligence, the answer seems even more affirmative: architecture is a linguistic system—primarily in a syntactic and generative sense, though not necessarily a semantic one.
Contemporary design is shifting the emphasis from “designing form” to “designing process.” Instead of manually defining a finished, closed object, architects are increasingly creating sets of rules that allow a structure to “grow” and evolve. This trend, known as Rule-Based Design, is not entirely new. It is, rather, a digital renaissance—and an exponential acceleration—of ideas that were born in the 1950s and 60s.
This essay attempts to trace this path—from analog linguistic concepts and design patterns, through the parametric revolution, to my own experiment (a digital illustration) with an urban generator, which serves as a bridge toward AI-assisted design.
ACTIVE TIPS: 2
From Concrete Grammar to Pattern Language
To understand what is happening on computer screens today, it is worth going back to the roots of systems thinking. The foundation here is the linguistic distinction between the abstract system of rules (grammar, langue) and the concrete, individual use of this system in practice (speech, parole).
In the 1960s, architects from the Team 10 circle (including Herman Hertzberger) translated this into the concept of so-called “support structures.” The architect designed an overarching framework (the system of rules), which users were meant to interpret and fill with their own arrangements over time. A classic example is Hertzberger’s Centraal Beheer office building—a repetitive concrete matrix providing support but leaving room for individual expression, where the user does not so much design as actively adapt the structure.
This idea was developed in a more humanistic approach by Christopher Alexander in his fundamental work, A Pattern Language. Instead of a rigid grid, he proposed a set of “patterns”—proven solutions to specific spatial problems. These patterns are nothing more than procedural rules or design heuristics from which designers and residents can “assemble” unique buildings. Alexander proved that relying on rules does not have to lead to monotony; on the contrary, it is a path to creating organic spaces.
The Wall of Complexity
If these ideas were so potent, why did their development stall? In the 1970s, Structuralism—as well as Alexander’s early attempts at a systemic approach—collided with the barrier of computational complexity. The problem was not only the immense number of relationships between elements but, above all, their non-linearity and dynamic nature. Manually predicting every possible permutation of a system on paper exceeded human cognitive capabilities. Architects could imagine complex systems, but traditional tools did not allow for their effective simulation.
Code as the New Grammar
Here we enter the 21st century. The renaissance of structural thinking is strictly correlated with the development of computational power. Computers solved the problem that stifled architects half a century ago. Today, code is becoming the medium that implements the grammar of rules (algorithmic logic). Algorithms allow us to define logical dependencies and then generate thousands of variants. Contemporary parametricism is a digital continuation of Structuralism—not a simple repetition, but an evolution that has built a bridge between the theoretical dreams of Modernism and the impending era of artificial intelligence.
Case Study: “Digital Metabolism” in Practice
Inspired by this historical continuity, I created an experiment in a 3D environment (WebGL/Three.js). The goal was to design a system based on Agent-Based Modeling.
The script utilizes a growth algorithm in which autonomous agents move through a virtual space. They scan their surroundings, and when they encounter an existing structure, they make a decision to add a new module, while strictly adhering to boundary parameters (constraints):
Gravity: Elements must have support—the system does not allow levitating blocks.
Adjacency: New elements must fit their neighbors according to material or functional logic.
Topography: The structure must react to the terrain’s morphology, adapting to slopes.
Initially, the system generated heavy, quasi-brutalist masses. However, by modifying variables in the code—reducing density and swapping blocks for a set of basic geometric shapes—I achieved the effect of a lighter volume. The presented script is a working model of the phenomenon of emergence—showing how, in the absence of a central plan, a complex global form emerges bottom-up from simple, local rules.
Conclusions: From Designer to Curator
Observing the forms generated by the script, we see a paradigm shift that is already standard in other industries. One need only look at gamedev. Creating gigantic, open worlds in video games would be impossible without procedural generation (PCG). There, programmers also create a “space of possibility,” and the computer fills it with content.
This way of thinking—from structuralist rules to parametric scripts—is becoming crucial in the face of the growing role of Generative AI. Working with AI rarely involves issuing a single simple command. It is a process of defining boundary conditions and principles within which the algorithm is to operate.
Thus, the role of the architect is changing fundamentally—we cease to be the exclusive creators of closed forms and become system designers. Our task is to define the problem and impose constraints. The final result is the effect of a process in which the machine generates variants, and the authorial decision consists of curatorship—selecting and polishing what has emerged from the calculations. This resembles the role of a director or gardener more than that of a traditional sculptor.