A note on AI in architecture, and the discipline nobody is teaching anymore.
The office has stopped drawing. It generates.
Every brief produces a thousand images before lunch. Endless variations on the same. The walls are covered in options. Nobody in the room can tell you which one is right.
Not because the images are bad.
Because design does not mean generating. Design means deciding.
And the hard part of design has always been knowing what to choose.
Choosing is a discipline nobody seems particularly interested in selling anymore. Look at what is being offered to architects today: AI for design options, AI for renders, AI for site analysis, AI for construction documents, AI for specifications, AI for presentations, AI for workflows.
Complete the course. Get certified. Become faster. Spend less time on repetitive tasks and more time designing.
It sounds irresistible.
If, of course, we know what we mean by designing.
Because sometimes the way we talk about design makes it sound like the pleasant part of the process. The creative part. The artistic part. The “oh, this is my style” part. As if everybody were waiting to fall in love with our personal taste, our line, our fashionable visual brand — and everything else were merely difficult, repetitive, technical, frustrating, boring and therefore avoidable.
As if the schedule were not design.
As if site analysis were not design.
As if resolving a code problem were not design.
As if listening to the client were not design. Or reading the brief. Or trying to understand what sits behind the brief and go beyond it.
As if discovering that the beautiful AI-generated model does not actually respond to the client’s needs — and may be considerably more expensive to build — were somehow outside design.
As if the fourth, tenth or twentieth revision nobody wanted to do were merely an inconvenience, rather than sometimes being the exact moment when the project finally became better.
We have started separating design from many of the things that historically taught us how to design.
For what?
Speed?
Since when has producing more generic possibilities simply because doing so is cheap and possible automatically been an advantage?
And once you have hundreds of them, who is going to digest that volume of information intelligently? Wisely? Professionally?
Who is going to choose?
The Tedious Part
Mastery in almost any discipline is built partly inside repetition.
Any art requires work. Hard work. It has happened throughout every human era: painting, sculpture, literature, architecture, medicine, music, drawing, engineering.
You do something badly.
Then you do it again.
Then slightly less badly.
You encounter resistance. Something does not fit. The idea fails. The detail refuses to work. The dimensions do not add up. The beautiful image collapses when confronted with gravity, legislation, money, climate, material and use.
So you return to it.
Again.
And again.
Until something happens that is very difficult to package into a seven-day certification programme:
you begin to know.
When I was a piano student, my maestro repeated one sentence to me constantly:
FIRST TECHNIQUE, THEN SPEED.
You must dominate the technique first. Then you can speed up.
I was a boy, but I still remember that advice.
It is deeply human to be seduced by speed. Faster feels better. Faster feels more advanced. Faster feels as though we have overcome something.
But speed without understanding and judgment is not simply useless.
It can be dangerous.
We need to give ourselves the time to learn from our mistakes. We do not need speed first. We need judgment first.
Technique first. Judgment first.
Then speed becomes powerful.
And judgment cannot be downloaded. It cannot be purchased in a course, faked through presentation or replaced by confidence.
It accumulates slowly, through exposure, correction, frustration, repetition and consequence.
Of course, some tedious work really is waste.
Nobody needs to romanticise copying room names between drawings or manually repeating tasks a machine can execute perfectly well.
Automate them.
Please.
But some tedium is waste, and some tedium is where judgment is formed — especially for younger architects who are still learning what works, what fails and why.
Those are not the same thing.
And if we automate them indiscriminately, we may discover the difference much too late.
The Image Is Not the Decision
There is another strange thing about the current promise of AI in architecture.
Almost everything is demonstrated through an image.
Look what it can render.
Look how quickly it can change the material.
Look at the lighting.
Look at the atmosphere.
Look how many options you can produce.
Judgment does not screenshot well.
There is no dramatic before-and-after image for knowing that option 37 is better than option 463.
There is no spectacular Reel showing the architect quietly rejecting nineteen attractive solutions because the twentieth one resolves circulation, structure, orientation, budget and experience simultaneously.
There is no wow moment in deciding that the most photogenic idea is actually the wrong one.
We cannot rely on technological power alone.
Think about 4K. Or 8K.
Shit in high definition is still shit.
At some point, we need to stop staring at the container and start looking at the content.
AI is a phenomenal generator. It is extraordinarily good at divergence. It can produce possibilities almost without limit.
But design cannot remain divergent forever.
At some point it must converge.
One of my professors used to tell me:
Composition often means removing, not adding.
And removing means selecting.
Selecting means judging.
Generating possibilities is only one part of design. Design is also the act of eliminating possibilities.
Sometimes ruthlessly.
A designer is not valuable because they can imagine ten thousand things.
A machine can already do that.
The designer is valuable because, among ten thousand plausible things, they can recognise the one worth pursuing — and explain why.
That distinction becomes more important, not less, when generation becomes almost free.
When producing one option took a week, scarcity imposed discipline.
When producing one thousand takes an afternoon, scarcity disappears.
And another problem appears:
selection.
Which one?
Why this one?
According to what criteria?
What are we optimising?
Who decided those criteria?
What are we willing to sacrifice?
What must survive?
What matters?
Those are not rendering questions.
They are judgment questions.
Welcome to the Gallery
So imagine the office a little further into the future.
The screens are extraordinary. Or perhaps there are no screens anymore. Holographic architecture floats around the room. Three-dimensional models appear in X-ray transparency. Every surface is alive with possibility.
The system can produce another option before you finish asking for it.
Change the roof?
Done.
More glass?
Done.
Less glass?
Done.
Timber. Concrete. Curved. Orthogonal. Japanese. Mediterranean. Minimal. Biophilic. Iconic. Quiet. Cheaper. More dramatic.
Done. Done. Done.
Another hundred alternatives appear.
The gallery becomes infinite.
Psychologists have a name for this: the paradox of choice — more options does not make a decision easier, it makes it harder.
And at some point the whole thing becomes paradoxical.
Every decision carries the suspicion that a slightly better option might be one prompt away.
Why stop now?
Why choose this one?
Why not ask for another?
And another?
And another?
This is the industrialisation of “Next!”
The future on offer is therefore not necessarily one where architecture becomes better.
It may simply become infinitely faster to generate, and progressively harder to decide.
A thousand options.
Nobody left who was trained to choose.
What Is AI in Architecture Actually Teaching Us?
This is the part that interests me most.
Not whether AI will replace architects.
That question is already becoming boring.
Of course tools will replace tasks. They always have.
The more important question is:
What happens to the architect when the tasks being removed were also part of the architect’s education?
Look closely at the language surrounding these new courses and certifications.
Prompt better.
Generate faster.
Automate workflows.
Produce presentations.
Create renders.
Reduce repetitive work.
Save time.
Fine.
But save time for what?
More clients?
An aperitivo?
More time posting and Instagramming?
We constantly describe time as our most valuable resource. So perhaps we should also ask what deserves that time.
Is time spent learning your profession really the time we should be trying hardest to eliminate?
Architecture is a practice that takes years to learn. Years to understand. Years to master.
There is accumulated knowledge in knowing how a building meets the ground. In knowing that the section is telling you something the plan is hiding. In understanding when the client is asking for one thing but actually needs another. In recognising when an apparently simple decision will cause ten other problems later.
You cannot compress all of that into a shortcut.
And perhaps we should stop pretending that every shortcut is progress.
Because if we short-circuit the process through which architects learn, we may eventually short-circuit the profession itself.
The Future I Actually Fear
I am not afraid of AI because it can draw quickly.
I am not afraid because it can render better than many architects.
I am not even particularly afraid that it will automate large parts of architectural production.
Much of that will probably be useful.
What worries me is subtler.
We are beginning to teach a generation that difficulty itself is a design flaw.
That friction should always be removed.
That the slow part is wasted time.
That mastery can somehow be compressed into the ability to produce the correct prompt.
But sometimes friction is information.
Sometimes the thing that refuses to work is teaching you something.
Sometimes revision is not inefficiency.
Sometimes drawing it again is precisely how you discover that you did not understand it the first time.
Sometimes the tedious part was never standing between you and the design.
It was teaching you how to design.
Let me share something personal.
Most of the time, I start my work by writing, not drawing.
A large part of my time is spent counselling and guiding clients, not modelling.
I spend enormous amounts of time studying, reading, researching, testing possibilities, understanding constraints and trying to identify the real question behind the obvious one.
Not rendering.
The beautiful image, the fly-through video, the polished Instagram post — those are only the tip of the iceberg.
The real design work is everything underneath.
The invisible part.
The slow part.
The difficult part.
The thinking.
The conversations.
The mistakes.
The corrections.
The decisions nobody will ever see.
That is the foundation of the work.
Skip it, shortcut it, automate it before you have learned from it — and one day we may discover that beneath all those extraordinary images, there is nothing left.
Nothing underneath the image at all.
This is the second postcard. Read the first one here.