Curators, not vassals

A few days ago I published an essay on project management in this same space. I wrote it with an artificial intelligence assistant: I dictated my stories to it as voice notes, out of order, full of filler words and half-finished sentences, and it turned them into a clean, well-structured text with references. In one paragraph it wrote that some payments had been held back “because of the change in the entity’s presidency”. The sentence sounded right, fitted everything around it and had exactly the tone of the rest of the essay. It was also false. The change wasn’t at the entity; it was the country’s president who changed. I spotted it once the text was already published, reading it calmly, and we corrected it that same day.

Nobody else would have noticed. There was no way to notice from the outside: the error wasn’t in the grammar or in the logic, it was in reality, and I was the one who had lived that reality.

That same week the opposite happened. I remembered a contract as having lasted six months, and that’s what I told the assistant. It went to look at the documents and showed me it had been close to nine, with an extension along the way. My memory was wrong and the tool was right.

I keep the two scenes together, because between them they take apart the two easy positions on artificial intelligence: the one that says the machine makes mistakes and is therefore useless, and the one that says the machine knows more and must therefore be obeyed. Neither describes what happened. What happened is that there was a person who knew things the tool couldn’t know, a tool that found things the person had forgotten, and someone who had to answer for the result. That someone was me.

01 — The eyeHaving the tools is not having the judgement

A few months ago I came across the work of Kakeru Taira, a Japanese artist who builds settings that are hard to describe: metro stations, building corridors, waiting rooms. Ordinary, almost bland places that look like lucid dreams in his images. There’s a blur that keeps you from seeing quite what object is there, yet you understand perfectly where you are. You recognise the place and at the same time feel that something doesn’t add up.

On his channel he had videos showing how he worked, which tools he used and how he had brought artificial intelligence into his process. Those videos are no longer available. What remains is the work.

Since then I keep telling my team something: I have access to practically the same tools he does. I could install the same programs, use the same models and spend years iterating. And I still wouldn’t arrive at anything with that force. Not because the tools stop me, but because I don’t have his sensibility. He knows when an image works, which detail changes the atmosphere, what is too much and what is missing. I could generate a thousand images; he knows how to recognise the one worth keeping.

That his videos have disappeared seems to me, in a way, the best illustration of what I mean. The process can be hidden, copied or made obsolete within a year, when the tools change. The eye can’t.

02 — The bridgeProducing is not judging

The same thing happens far from art, and there the consequences are less poetic. Today I can ask an artificial intelligence to calculate the loads on a bridge, propose materials and program a simulation. It will hand me something convincing: formulas, charts, reasonable explanations. But I’m not a structural engineer. I wouldn’t recognise a badly framed load condition or a wrong assumption. Worse, I could be steering the tool towards the error with my own questions, and we would both end up convinced that everything was fine. The bridge doesn’t care how confident we were. It falls down all the same.

This isn’t just a hunch. In 2023, a group of researchers from Harvard and other universities ran an experiment with hundreds of consultants from an international firm. When the tasks were within what the tool did well, those using artificial intelligence worked faster and to a higher quality. When the tasks were just outside that boundary, the opposite happened: those using the tool got it right considerably less often than those working without it. The researchers called it a “jagged frontier”, because you can’t see where it ends. On the inside, the tool makes you better. On the outside, it makes you worse and doesn’t warn you.

That, I think, is the confusion of this moment. We are treating two very different capacities as if they were the same: the capacity to produce and the capacity to judge what has been produced. The first is becoming accessible at a speed we have never seen. The second still demands knowledge, experience and, above all, someone willing to answer for it.

03 — The wordTo curate is to take care

I chose the word curator on purpose, and I know it has a problem: it sounds like someone who picks between options, like a person choosing which paintings hang in a gallery. If that were all, it would be a poor word for what I mean, because choosing among what a machine produces is the easiest part of all. But the word comes from the Latin curare, which means to take care, to take charge. The original curator wasn’t the one who chose, but the one who answered for something that wasn’t theirs.

That is the sense I’m interested in. A curator, in this age, is someone who takes charge of what the tool produces: who frames the problem well before asking for anything, who recognises a mediocre solution even when it is impeccably presented, who knows which side of the frontier they are standing on, and who signs.

The vassal is the opposite. Not someone who uses artificial intelligence; we are all going to use it. It’s someone who hands over their judgement to it. Who accepts the first answer because it sounds sure of itself, who stops asking whether that is what the problem needed, who ends up answering to the tool instead of answering for it.

And there is something more, which seems to me the most important part. Knowledge isn’t only there to keep us from getting things wrong. It’s there to take us further. The small stuff, the manual work, the things that ate up our time and attention, is falling away, and what remains is open ground for deciding where to go. An engineer who truly understands structures can use these tools to test designs they used to rule out for lack of time. An artist with an eye can explore atmospheres that would once have taken months. Experience stops being only what keeps us safe inside familiar territory and becomes the starting point for leaving it. Without it, the tool just has us going round in circles faster in the same place.

04 — The paradoxSo how is the eye formed now?

This is where the objection that troubles me most comes in, and I’d rather put it on the table than dodge it. If the judgement of the artist, the engineer or the project manager was formed precisely by spending years doing the small stuff, how will someone starting today form it, if the tool does the small stuff for them?

It isn’t a rhetorical question. The same jagged-frontier study found that the consultants who improved most with the tool were those with the weakest previous performance. Artificial intelligence raises the floor, and that is good news. But raising the floor is not forming the eye. Someone can deliver correct work for years without ever developing the ability to recognise when something correct is wrong.

I don’t have a settled answer. I have a few ideas that seem sensible to me for those of us who lead teams:

• Ask for the why, not just the result. Every deliverable should come with the decisions behind it: what was ruled out, what was checked and against what.
• Have new people critique what the machine produces before they produce with it. Finding the error in someone else’s text trains the eye faster than writing your own.
• Keep room for doing things by hand. Not out of nostalgia, but the way you train a muscle: handmade work still has value and, in some cases, will become more and more desirable.
• Create zones where mistakes are cheap. Prototypes, pilots, internal exercises: places where failing doesn’t cost a client’s trust or anyone’s safety.

If we don’t do something like this, we run a quiet risk: a generation of very productive professionals who never had the chance to become experts.

05 — The fearThe age of safety is over

I understand that all this is frightening. And I don’t mean a reasonable unease that a course can fix. I mean real terror, the kind felt by someone who spent twenty years mastering a technique and watches a tool carry it out in minutes. Uncertainty is deadly, and for many people failing is not an option: there are families, debts and people who depend on everything going well.

Not all fear is resistance to change. Sometimes it is a very clear-sighted reading of the risks, especially when organisations use these tools to cut rather than to grow. And not everyone has the same margin for getting it wrong. That needs saying, because it is easy to preach experimentation from a comfortable position.

But I also believe the age of safety ended a few years ago, and that pretending otherwise is the riskiest way to live through this moment. Mastering a technique no longer guarantees decades of stability. What we can do is stop measuring our worth by how long it takes us to draw up a plan, write a report or draw an image, and start measuring it by what we are able to see, decide and stand behind.

What better moment than this to get things wrong, to find out how far we can go. Let the chains fall, let the blindfold come off, even if what we see frightens us. I’m not talking about experimenting with other people’s safety or failing through carelessness. I’m talking about recovering something many organisations took from us over the years: permission to try something without knowing whether it will work.

I go back to the moment I found the error in my own essay. There was nothing heroic about it: a paragraph, a changed word, five minutes of correction. But it left me thinking about how many sentences like that go round every day, well written, coherent, published, that nobody will correct because the only person who could have noticed didn’t read them carefully, or because there was no longer anyone with the experience to notice.

I don’t know how the professions will reorganise themselves. I’m wary of anyone who says they know. What I do know is that in every craft there are errors only someone who really knows it can see, and successes only that same person can imagine.

What is the error in your work that only you could see? And what could you try now that you wouldn’t even have considered before?