The Architecture of Judgement

A question raised by AI

What does an architect contribute when AI can generate images, options and written descriptions in seconds?

That question became part of the development of the new SCA website. The task began with practical requirements: rebuild the practice’s public presence, clarify the work, improve the copy, organise the projects and make the site easier to use.

As the work progressed, the website became a useful test of AI itself. The process showed where the technology could support architectural thinking, where it became unreliable, and where judgement still had to come from the practice.

This was not a theoretical exercise. The website had to describe real work. It had to explain projects clearly, organise them coherently and make the practice’s position legible to clients, collaborators, students, press and a wider design audience. Each page had to carry a relationship between what SCA makes and how SCA thinks.

AI became useful within that process, but never as an authority. It produced possibilities. The practice had to decide what those possibilities were worth.

Why speed is not enough

Much of the current discussion around AI is built around speed. Faster images. Faster options. Faster workflows. Faster presentations.

These gains have practical value. A tool that can accelerate research, generate alternatives, reorganise information or test visual directions can widen the field of work. It can make early exploration faster. It can bring several possible routes into view at once.

The difficulty begins when speed is treated as evidence of progress.

A project can now produce more material in a shorter period of time. That wider field of material creates its own pressure. More options mean more decisions. More visual output means more need for criteria. More fluent language means more risk that weak thinking will be hidden by smooth phrasing.

Architecture has always involved selection as much as production. A drawing, model or render only becomes useful when it can be tested against the brief, the site, the structure, the material system and the experience it is meant to create. AI increases the quantity of available material. It also increases the importance of knowing what should be carried forward.

The central question is therefore not whether enough ideas can be produced. The question is whether the right ideas can be recognised, tested and resolved.

Testing AI through the website

AI was used throughout the website process. It helped draft copy, test alternative structures, explore phrasing, organise project descriptions and produce supporting imagery. It allowed a broader range of options to be examined than would have been practical through a conventional editorial process alone.

Some outputs were useful immediately. Many were wrong. Some sounded close to SCA but said very little. Others produced fluent language with the wrong centre of gravity. The writing appeared composed, but the thought behind it was weak.

That became an editorial problem rather than a technical one.

Every paragraph had to be checked against the work. Every claim had to be earned. Every phrase had to answer a direct question: does this make the practice clearer, or does it only make the sentence more polished?

The same issue appeared in the images. AI could generate atmosphere, landscape, context and visual direction with remarkable speed. It could suggest a setting, a mood or a surrounding condition. It could also alter the architecture itself. In those cases, an image might contain something useful while failing at the point that mattered most.

A landscape could be retained. A building might need to be replaced with a controlled render. A composition might suggest a useful direction while misrepresenting the project. The task was to separate contribution from distortion.

The website process therefore became less about using AI to produce finished answers and more about using it to expose questions. What belongs to the work? What is only stylistic? What makes the practice clearer? What distracts from the architecture? What survives comparison with the actual projects?

What had to be judged

The most useful part of the process was not the generation of material. It was the repeated act of judgement around that material.

Architecture develops through generation, testing, correction and resolution. A reference is identified. A system is developed. A form is tested against structure, material, movement, climate, programme and public experience. Ideas are removed as often as they are added. The final work depends on what is rejected as much as on what remains.

AI made that part of the work easier to see.

For SCA, this question matters because concept design is central to the practice. It is built through research, comparison, consequence and experience. If a machine can generate persuasive concepts in seconds, the value of that experience needs to be examined carefully.

The website process clarified the issue. AI could clearly produce ideas. The harder question was whether those ideas had architectural value.

A proposal may be visually persuasive and still have no reason to exist. A phrase may sound convincing and still fail to describe the work accurately. A concept may appear complete before it has been tested against place, brief, structure or use.

In architecture, the image has never been the whole discipline. Drawing is a medium. Modelling is a medium. Rendering is a medium. AI is another medium. The discipline lies in deciding what the work is being asked to do, which references matter, which systems can carry them, and when the result has become resolved enough to stand.

Judgement begins before form. It asks what question the project is really answering. It asks whether a cultural reference has architectural force. It asks whether a structural idea genuinely organises the building. It asks whether a material decision changes the experience of the work. It asks whether an option should be developed, paused or abandoned.

Reference, system and resolution

The website process connected directly to SCA’s principle of Form Follows Meaning.

A project begins with a reference: cultural, historical, environmental, material, structural, social or site-specific. That reference is translated into a system. Structure, movement, space, envelope, material and public experience are developed together. The final form is reached through that process.

The website followed the same logic.

The reference was the practice itself: the body of work, the recurring questions, and the methods that had shaped theatres, pavilions, special structures, public infrastructure and speculative concepts.

The system was the website: project structure, navigation, imagery, language, page hierarchy and editorial rhythm.

The resolution was a clearer account of what SCA is and how the work should be understood.

This made the website more than a communication exercise. It became a method of self-definition.

The process revealed a consistency in the work that had not always been fully articulated. Across different scales and programmes, the same pattern appeared: a meaningful origin, translated into architectural logic, resolved through form. AI did not invent that pattern. It produced enough alternatives to make the criteria more visible.

That is where the tool became useful. It helped expose the structure of judgement already present in the work.

The architectural consequence

AI can expand exploration. It can accelerate movement between options. It can support research, reorganise information, test tone, generate visual directions and reveal patterns that might otherwise take longer to see. Used carefully, it can make parts of the design and communication process more expansive.

Architectural responsibility remains with the architect because the consequences remain architectural.

A prompt can be changed instantly. A building cannot. Decisions made early in a project affect cost, structure, planning, public experience, environmental performance and the way a place is understood. The ease of producing an option increases the importance of choosing well.

This may be the real shift.

For a long time, the visible effort of architecture was tied to production: drawings, models, images, documents and presentations. AI changes that relationship. It makes some forms of production faster and more accessible. The value of the architect becomes less easily measured by the number of options produced and more clearly measured by the quality of the decisions made.

The architect’s role becomes more explicit. The work lies in recognising the idea that can survive development.

What will this decision do to the brief? What will it do to the structure? What will it do to the experience of arrival, movement, light, material, climate, cost and public meaning? What will remain true when the image is gone?

These are not questions a tool can settle on its own. They require responsibility for consequence.

What deserves to remain

The development of the new website changed SCA’s understanding of AI. It also clarified something about architectural practice.

The generation of ideas is only one part of concept design. The harder task is evaluation: recognising which ideas have depth, which have only appearance, and which can survive the pressure of development.

That is the architecture of judgement.

It is the work that happens between possibility and resolution. It is the discipline of deciding what deserves to remain.