20 January 2003. Give or take a day.
I’m catching a train to my sister’s eighteenth birthday. The train is late, the doors close too quickly, people start moving around, and one of those things happens that, twenty-three years later, feels almost scripted — although at the time it was simply painful.
I fall from the train and break my right hand.
For the first time.
I’ve told that story before, but I’m coming back to it today for a different reason. Because of a strange historical echo triggered by a sentence I’ve been hearing more and more often over the past few months.
At the time, I had just finished my Urban Planning II studio with three other girls.
Anyone who studied Architecture back then will know exactly what I mean. For everyone else, imagine four months spent bent over enormous maps, colouring them with pencils, Pantone markers and the whole analogue arsenal that still carried an aura of unquestionable academic nobility.
I probably don’t need to explain that the purpose was not to revisit kindergarten — or to make up for having abandoned it rather early myself because I found it boring.
By January, studio classes were over and it was time to finish the drawings for the exam. We had divided them between us.
Minor problem: my right hand was in plaster.
So much for receiving the usual compliments on my elegant ink stippling and watercolours.
Necessity being the mother of invention, I decided it was time to learn Photoshop properly.
I was already a CorelDRAW fan. I had fallen in love with it when I was twelve, after discovering it in one of my father’s computer magazines. Photoshop, on the other hand, still felt rather hostile to me — mainly because there was one thing I wanted from it above all else:
automation.
The exam was only a few days away, and I needed to redo all my drawings from scratch, digitally, preferably using just one hand.
My left one.
So I read manuals, hunted down guides at Feltrinelli, tried to understand batches and procedures, experimented, failed, tried again.
I had also read Bruno Munari’s Da cosa nasce cosa, and perhaps something about method had already begun to settle in my head.
But mostly, I had a broken hand and an exam to pass.
If there was a smarter way out of the situation, I had to find it.
And somehow, it worked.
I recreated all my drawings using my left hand and a mouse with the buttons reversed. By the end, I was rather proud of the result.
My classmates, slightly less so.
The professor, considerably less.
At Quaroni, more than twenty years ago, group exams normally meant one grade for everyone.
The professor made an exception.
Three 30s and one 29.
Guess who got the 29.
More importantly, guess why.
“Your drawings were made by the software. Not by you.”
Does that make you smile?
Good.
Because I brought you into this story to take you straight back out again — into another one, much more recent and considerably less funny:
“You didn’t make this. AI did.”
Still smiling?
Because twenty-three years later, I recognise exactly the same mechanism.
When we can no longer see the effort, we begin to question the value of the person who produced the result.
Of course Photoshop and generative AI are not the same thing. I think we can all agree on that.
And that is precisely why the question matters even more today.
An LLM can generate, compare, write code, suggest alternatives and, occasionally, tell you something completely wrong with breathtaking confidence.
Trying to ask, “How much did you do, and how much did the AI do?” means attempting to weigh on a tiny set of scales something that actually emerges from the relationship between a person, a tool and a method.
A few days ago I used a Stradivarius as an example with my colleagues.
Suppose I placed one in your hands today.
What would you be able to play?
I would, in all likelihood, produce an appalling and extremely expensive noise.
A great violinist would obviously produce something else entirely.
And nobody listening to Paganini would say:
“Well, yes, but you didn’t really play that. The violin did.”
It would be equally absurd, however, to claim that the violin does not matter.
The violinist matters.
The Stradivarius matters.
And above all, what matters is what that violinist is capable of doing with that Stradivarius.
I recently read about mathematician Ernest Ryu at UCLA using GPT-5 on a problem that had remained open for years.
The model also generated arguments that sounded plausible but were wrong.
The value lay in Ryu’s ability to recognise them as wrong.
The very same LLM someone else asks how to lose weight.
Different person.
And this is what we underestimate when we say that “AI does everything now”.
The difficult part is not some mythical talent for prompt engineering.
It is knowing what to ask, which information actually matters, when an answer is merely seductively plausible — and when you need to stop talking to the AI and switch your own brain back on.
Then comes the part I find even more interesting:
turning a good answer into a good process.
Throwing a mountain of data into Gemini and asking, “Make me a report,” may produce something beautiful.
But the less glamorous question is:
What happens next time?
Which data?
Which rules?
Which checks?
I find it far more interesting to use Claude to build a small program that takes the right data, applies the rules, generates the report in the required format and flags the exceptions.
Then the result becomes predictable.
Repeatable.
Standardised.
Controllable.
In one word:
Quality.
And this brings us to the biggest misunderstanding of all.
We are getting excited because AI lets us do in ten minutes what once took ten hours.
Wonderful.
I have spent much of my professional life looking for ways to do boring things better and faster, so I am probably the last person who will mourn those ten hours.
But if, after those ten minutes, the result is exactly as mediocre as the one we produced before, then what we have mainly invented is:
faster mediocrity.
No.
That is not enough.
If I have a better tool, I want a better product — not simply the same product made faster.
Otherwise, we have bought ourselves a Stradivarius simply to play the same scale more quickly instead of creating new harmonies.
What excites me is using the time AI gives back to us for all those things we used to sacrifice because they were too expensive:
checking, comparing alternatives, simulating, looking for the mistake before it becomes a problem.
Because, frankly, we have become rather tolerant.
Projects full of errors and gaps.
Reports nobody reads.
Models that are formally complete and practically useless.
AI can produce all of that much faster — and make it look magnificently professional while doing so.
The danger is that it may make mediocre things beautiful enough for us to stop noticing that they are mediocre.
And this is where our profession really begins to change.
For a very long time, the main cost was production:
drawing, writing, modelling, calculating.
Today we can produce fifty alternatives in the time it once took us to produce three.
Fantastic.
But someone still has to understand which one is good.
Suddenly, the scarce resource is no longer production.
It is judgment.
The ability to recognise that an elegant solution does not actually solve the problem.
That a convincing result rests on the wrong piece of data.
That the less spectacular option is, in fact, more robust and more verifiable.
The more the machine can produce, the more valuable the person who knows how to choose becomes.
Perhaps that is why that grade stayed with me for more than twenty years.
Not because of the missing point — I have collected plenty of 29s since then — but because of the criterion behind it.
My classmates had worked with pencils, and their effort was visible.
I had built a digital process that was somehow both innovative and anachronistic, and that effort had become invisible.
Seen from the professor’s desk, pencils were work.
Software was a shortcut.
If AI now allows me to do in two hours what once took me two days, how much is my work worth?
Two hours?
No.
If twenty years of experience, a method and better tools allow me to produce in two hours something that used to take two days — perhaps with fewer errors and a better result — I cannot see why its value should decrease.
If anything, the opposite.
The real challenge is making sure we do not become that professor.
Very soon, we will be the ones judging work created with tools we do not fully understand.
Teachers.
BIM Managers.
Team leaders.
Directors.
Clients.
Reviewers.
Perhaps people with thirty years of experience — and that dangerously reassuring feeling that we know pretty well how things are supposed to be done.
Someone will show us, in ten minutes, something that would have taken us ten hours.
And perhaps our first reaction will be exactly the same as it was in 2003:
It isn’t worth as much.
They didn’t do it.
The software did.
When that happens, I hope we will be present enough to ask better questions.
Why did you choose this approach?
How did you verify the information?
Where did you decide not to trust the tool?
And finally, the question we should have been asking from the very beginning:
Is the result better?
Sometimes I like to imagine walking back into that classroom.
The glossy prints spread across the tables in Via Gianturco.
Photoshop and AutoCAD open on my first laptop — a 17-inch Toshiba Satellite.
My right hand in plaster.
Those four grades still waiting to be assigned.
Except that this time, the professor does not try to determine how much of the work was done by the software.
He looks at the drawings.
At the decisions.
Perhaps he even notices things he does not like.
Then he looks up and says:
“Show me how you did it.”
Not to uncover the trick.
To understand.
Perhaps progress begins exactly there:
when we learn to suspect that there may be value in something produced through a process we do not yet understand — even when, at first glance, it looks worse.
Because sooner or later, someone will arrive carrying their futuristic Stradivarius.
They will do something strange, unconventional and new — in a fraction of the time — and it will look impossibly difficult to us.
I hope we will not stand there counting how much was played by the person and how much by the violin.
I hope we will still know how to listen.
