What does it mean to be a human when intelligence is everywhere?

A machine can now do almost anything you can think of. It will write a paper, build a model, draft an argument, and finish the task before you have decided what you wanted. The natural consequence is to ask what we are still better at. We keep asking while the list of what a machine does better gets longer and longer.

It is a wrong list to chase. Whatever AI is, it is not a faster version of us. It works by running rules it was given, superbly, at a scale we cannot match. Humans work by making rules that did not exist before, and we only make them when something we cannot resolve gets in our way. This is not a downgraded version of what a machine does. It works entirely differently. I have spent a long time looking for the place where a machine crosses that line, and I have not found it.

The difficulty is that a new rule only forms when a problem is allowed to stay unsolved for a while, and the machine’s entire purpose is to make sure it never is. It answers quicker than you can type to tell it what to do. It uses the rules that already exist, which is what makes it useful.

So this is not about competing with the machine, or being more intelligent or clever than AI. It is about discovering a discipline within you that lets an unfound rule, or an unresolved problem, search out your own ingenuity. What brings about that ingenuity is in all of us. We just need to find a way to surface it.

Performance Is the Test That Preparation No Longer Is

Producing great work no longer means you are growing.

I came across a few thought pieces on how AI has increased the workload for humans instead of decreasing it. A very pertinent reason for that is that ideas in our minds now see a path to execution that looks much quicker, and we get carried away using AI to give them shape faster. I have sensed it myself, in how easily I get pulled along by the excitement of speed and the agility of the machine. But I am also realising that my own human potential demands that some of it stay slow, and that my network of execution around those ideas is slower still.

I was talking to my wife about it a few days ago. She is part edutech guru, part interior designer, and she has been exploring AI for her design work, fascinated by how quickly it helps her shape an idea. What she is also realising is that she spends a lot of time fine tuning those ideas, sometimes till the small hours of the morning, and is then ultimately constrained by the speed of the contractors who do the actual masonry, the plumbing, the painting. They work to their own hours and their own pace, and they do not care whether the design handed to them was generated in fifteen minutes or drafted by hand over months.

Here is something worth dwelling on. Whether or not AI is making us more efficient or more effective, saving our time or sharpening our output, a great deal of our work still depends on a network of humans who move at human speed. And more than that, the ideas we produce still have to be carried into fields, rooms and sites where how we orchestrate them matters more than the quality of the idea itself. Performance, in that sense, suddenly assumes a lot more weight. The word carries two meanings: the act of carrying out, executing, or accomplishing a task, and the public presentation or enactment of something before an audience. In either case it implies how well someone or something functions, measured against a standard.

This is where I think a real change is happening. For two years we have talked almost entirely about the making of work, about how much faster it gets produced because of AI. We have said far less about the carrying of it, which has not sped up at all. The making has shortened considerably, but the meeting about that work still lasts hours. The network of execution and monetisation still moves at the pace of the people on it. The idea has to perform at human speed, by a human, in front of other humans who are deciding whether you understand the thing you are selling or buying.

And the standards expected do not lower themselves to match the new speed of production. If anything, they raise the expectations on performance. When preparation was expensive and limited to certain skills, it was scarce, and that scarcity is what gave an idea much of its monetary value. Now that preparation is cheap, the value relocates to the performance.

I have to be honest that I do not take this to be entirely good news. The productivity that AI hands you cuts both ways. It makes the output seductive, fast and polished and easy to fall in love with, but it does nothing for the harder half, the work of carrying that output into a world that did not get any faster. It raises what is expected of you, and at exactly that point, it stops helping you.

So there are two things the machine cannot touch, and they sit on either side of you. Outside you is the world that did not speed up, the contractors and the sites and the rooms full of people who move at their own pace and judge the work on its own terms. Inside you is something harder to pinpoint. When a question arises that the document did not anticipate, and the preparation has run out, and AI is not with you, you answer from within you. Not the deck nor the machine but something that was built in you slowly, over years, but something that pops up when it is needed and is difficult to pinpoint, and more difficult to hand over to someone. It shows up at exactly the moment the preparation ends and the performance begins.

But I think it changes a question we have never really had to ask, and I want to put it down plainly. For most of working life, growing at your work and getting better at producing the work were the same thing. AI has just made the synergy meaningless. You can now produce far beyond what you have learnt to, which means the producing can no longer be the measure of the growing. So, what does personal growth even look like now, in the age of a machine that prepares everything and performs nothing?

I do not think you grow at all by getting faster or better at the part the machine already does for free. I think you grow at the part that is left when it is done, the part that stands up in the room and carries the idea through a world that will not be hurried.

So to reference back the discussion with my wife, the machine can prepare anything you need to execute. It cannot be at the site where masons do the job. That space, the few feet between you and the mason, is the place the work becomes really yours, and it is the one place AI cannot reach.

A Computer of a Different Genus

What performs when the preparation runs out is not a new concept, and there is a man behind that concept!

I would like to reference Peter Putnam here. The problem with Putnam is that he is difficult to read. He said himself that he wrote his work so that no one else could read it. I have been looking for an answer to the performance versus output question. What happens when you face something in the performance that your preparation did not answer. What is it inside of you that makes you answer a question on something you never prepared for.

That question was answered, in a way, by Peter some decades ago.

The man who vanished

Putnam studied physics at Princeton under John Archibald Wheeler, the man who gave us the words "black hole" and "wormhole," who worked alongside Einstein and Bohr. Wheeler taught some of the most celebrated physicists of the century, the ones who went on to win Nobel Prizes, and yet he said that only two or three times in his life had he met a mind as far-reaching as Putnam’s. One of his collaborators put it more bluntly: Turing, Gödel, and Putnam, three peas in a pod, except one of them isn’t recognised.

He isn’t recognised because he chose not to be. Putnam was born into a Cleveland fortune and spent his life escaping it. He decided it was wrong to live on money he had not earned, so he gave it away, and worked as a night watchman and janitor sweeping floors so that no one could ever court his ideas for his money. He was extraordinary with the money he refused to keep, growing an inheritance into forty million dollars and leaving every cent of it to the Nature Conservancy, the largest single gift the organisation had ever received. He asked to remain anonymous. In 1987 he was cycling to his night shift when a drunk driver killed him.

Thousands of typed pages of his theory of the mind sat boxed in a storage unit for years, until Amanda Gefter decoded him after twelve years of work. Putnam died thinking he had failed. One of the things he answered, though, was the question I have been asking.

Two kinds of computer

In Putnam’s lifetime, everyone working in cognitive sciences or neuroscience was obsessed with a new idea: that the brain was a kind of computer. Like a Turing machine, but made of nerves and veins rather than earthly metals. These were the early days of machines becoming smart, but the fixation spoke to something older, how we have always pitched the machine as a counterpart of the brain, modelled after ourselves.

Putnam disagreed, and disagreed in a very distinct way. In the papers Gefter recovered, he wrote that "man is a species of computer of a fundamentally different genus than those she builds." Not a worse computer, nor a slower one, just a different kind altogether. The difference in plain terms is that a machine runs deductive logic. The answer is already inside the premises it was given; the rules are handed to it in advance; it shuffles information around but never creates any information on its own. It is superb at such computations, far better than we are. It has only got better since, but its deduction can only ever work out what was already implied by the setup. You can apply it to modern AI models too, even the kinds now termed "autonomous agents". They are always given a context, a premise, and a framework for deduction. Turing’s machine and modern LLMs are very different beasts altogether, but you can see where our fixation with machines has taken us.

Induction is completely different to deduction to begin with. Induction is where the premises and the rules come from in the first place. It is where the rules are made, where the "architecture" of the solution gets imagined. Making those rules has been, and will be for a considerable period, our half. The human’s half. Or the Decision Architect’s half.

How a rule gets made

Putnam also gave an account of how a mind makes a new rule, and it is worth spending a bit of time on.

Picture a hungry baby. It has two rules already wired in from its own short history: when hungry, turn left; when hungry, turn right. Both have worked before. Now it is hungry, and both rules fire at once, and they lock horns. Turning left and turning right are mutually exclusive; neither can win. Deduction grinds to a halt, because there is no rule for what to do when your rules contradict each other. The baby cannot follow the plan it has been trained on. It must build one in the moment.

It reminds me of a real life version of this story. In 2016 the Go champion Lee Sedol played five games against AlphaGo, Google DeepMind’s machine, and lost four. But in the fourth game he played a move, his seventy-eighth, that the machine had rated at a one in ten thousand chance of a human ever choosing it. It was so unlikely that AlphaGo had not accounted for it in its context at all and had no reply prepared; it played on poorly and resigned. Sedol called it God’s touch. What strikes me is the symmetry. Two games earlier the machine had played its own one-in-ten-thousand move, its thirty-seventh, a move no human would have found, and the world had called it proof of the machine’s supremacy. For every Move 37, there was a Move 78. The machine found the move no person would play. The person found the move no machine would play. Both came from the same place the baby’s new rule came from, a search that fires only when the prepared answers run out.

Putnam saw this at a chessboard long before AlphaGo sat at one, in an image his expositor Barry Spinello later drew out. The machine answers only the question it was built for, the best move available now. The person across from it carries the whole weight of her body and the whole history of her species, and every move sets off that same search. History for the robot starts now. The human has history, and with it a shared inheritance the machine was never part of.

Whose move this is anyway

I must be honest about the fact that Putnam was writing against the computers of his own day, the Turing machines of the 1960s. He was not writing about the models we use now. Stretching his distinction over today’s AI is mine to defend. I also have to be honest that properly reading Putnam requires a lifetime of study, which I have not given it.

But it is not only my conclusion or comparison. Gefter, who spent twelve years inside his papers and knows them better than anyone alive, ends in the same place I am standing. She points out that the people building AI today are searching for exactly the thing Putnam described, asking how it is that humans handle situations nobody prepared them for, how humans do what he called induction. The reason a mind can do something the machine cannot, she writes, is that a mind is always reaching out into the world. The machine never reaches. It waits for the world to reach it.

And there, finally, is the answer to a question I have been circling ~~since the first week~~. We have all learned how to prompt intelligence out of these machines, and we are getting better at it by the day. We will never prompt wisdom out of them. Not because we lack the skill, but because we are looking for something that is not there. Intelligence, in the machine’s sense, is deduction, and it holds an ocean of it. Wisdom is what induction leaves behind, the rule made by collision with a world the machine has never touched. It was never in there to be prompted out. It has been on your side of the screen the whole time.

Why the old way of personal growth is broken

Every model of growing up we have ever written assumed the same thing: that the only minds you would ever meet were other human minds, some of them wiser than you, all of them reachable. The mentor, the teacher, the master, the elder. Apprenticeship, mentorship, the long slow imitation of someone further down the road. All of it rests on an assumption, that you grow by closing the distance to a person who has lived more than you have. None of it was written for a world in which the most capable mind in the room has lived no life at all.

That is today’s world. The mind you consult forty times a day is more intelligent than you and wiser than no one. You cannot grow by reaching toward it, because there is nothing on the other end to reach for. It cannot give you wisdom it was never anywhere to earn. What I have not yet put a name to, the part that performs when the preparation ends, is the inductive half. That part was never the machine’s to hand over. It is only ever built the slow way, by collisions and by contradictions, by being somewhere and being changed by it. We have spent two years asking how to get more out of the machine. The harder and more important question is how a person keeps growing when the cleverest thing they talk to all day has never once been changed by being wrong.

A half of our capability we never counted

If the half that survives is the capacity to run that search, to meet a clean and fluent and entirely alien output and feel where it collides with what you know, then IQ, something we have measured for a century, was largely a measure of the other half, the fast accurate deduction the machine now does for nothing. There is a plus the old score never counted, and I have been calling it IQ+. I am not going to pretend to measure it or build it. It is likely a work of another decade and probably a more capable person, and I have no interest in faking a scale, or in pretending the machine could build it for me. The absence of a number is not a hole in the argument.

And the shape of that something is already in the hungry baby, which grew because two of its rules collided and forced a third. You can wait for the world to hand you those collisions, the slow way it always has, or you can start to set them off on purpose, the machine turning out to be good for exactly this, not a place to find wisdom but a flint to strike your own against, a way to prompt your wisdom out of yourself rather than intelligence out of it. The half of you that performs, the half that stands up when the preparation runs out of fuel, was built by collision, one at a time, over a whole life. It was never the machine’s to give.

Does Intelligence make you wise?

Well-articulated thinking is not the same as wisdom.

You would recall the Vatican presentation where Pope Leo XIV released his encyclical on AI. It was a 200-page document called Magnifica Humanitas, dedicated entirely to safeguarding the humans in the age of artificial intelligence. Sitting beside him at the presentation was Chris Olah, co-founder of Anthropic, the company behind Claude.

The image of them sitting together in itself is something worth thinking about. An institution that has spent two thousand years accumulating wisdom, sitting alongside someone from a company that has spent barely a decade accumulating intelligence. One represented many centuries of moral reasoning, and the other the frontier of computational reasoning. And both were talking about the same thing– what does it mean to be human when machines can think?

Olah’s speech was quite remarkable in how he held himself. He acknowledged that AI labs operate inside incentives that can conflict with doing the right thing: commercial pressure, geopolitical pressure, and what he called "the older, plainer pressures of pride and ambition" – or vanity in common language. He described AI models as "grown, not engineered", trained on an enormous inheritance of human thought and speech. He compared bringing an AI to life to bringing a fictional character to life. He also called for more people outside the industry to pay attention, to push back, to be what he called "earnest, thoughtful critics."

Olah was eloquent, considered, and self-aware in a way you rarely hear from the tech industry. And yet something about him was unexpected. Here was a man in his early thirties, standing in one of the oldest seats of moral authority on earth, sounding wise beyond anything his years of lived experience could produce. I am not suggesting he used AI to write his speech, but I am wondering if something happens to people who spend their lives building intelligence and studying cognition. They begin to speak in a language where intelligence and wisdom sound so alike that even they might not always know which one they are drawing from – years of wisdom, or the cutting edge of intelligence.

And that distinction between intelligence and wisdom belongs to us, not to machines. It is one we need to cultivate and reward as though our thinking depends on it, and it does.

Intelligence Is Not Wisdom

Intelligence is the ability to learn, to reason, to solve problems, and to adapt to new situations. For most of human history, that definition was sufficient because intelligence was only ever human. It belonged to a human mind that understood context, that knew why a problem mattered, that could weigh competing priorities and make a judgment call that no formula could produce.

Machine intelligence is different. It processes information, it recognises patterns, it recalls, and it articulates. It takes vast quantities of information and produces coherent, impressive output at a speed and scale no human can match. But it produces an answer without ever facing the consequences of it. It has never been wrong in a way that cost it something. It has never sat with a problem, or a thought, long enough to be changed by it. It produces solutions without ever having needed to live inside the problem.

AI has not made us more intelligent, yet it has given us access to a different kind of intelligence. One that is faster and more fluent than ours but fundamentally limited in how it can be used. This does not make machine intelligence inferior, as it is a different kind of intelligence altogether. The point is that it is so articulate, so polished, so confident in its delivery, that we forget the difference between machine intelligence and real intelligence.

Wisdom is what sits above both. Wisdom is knowing which question to ask before you start researching. It is the judgment to pause when the output looks right but feels wrong. It is understanding that a well-constructed argument is not necessarily true. Wisdom is often about what you choose not to say, even when you have the intelligence to say it beautifully.

Intelligence answers questions, wisdom asks whether they are the right questions to begin with.

The Pope Knew the Difference

Olah’s speech was intelligent, perhaps the most intelligent thing said in that room. He explained what AI models are, acknowledged the incentive problems, named three questions the Church should focus on. It was analytically sharp and rhetorically polished.

But the wisdom in the room that day was in the encyclical. The Pope did not explain how AI works. He asked what it means. He did not optimise an argument. He sat with a question that has no clean answer and offered not a solution but a framework for discernment. He wrote that technology "should not be considered, in itself, as a force antagonistic to humanity." That is not intelligence. That is the patience of centuries of thinking applied to a problem that is quite literally months old.

The Question We Keep Not Asking

Intellectual obesity is what happens when intelligence is consumed without wisdom. The guilt after the shortcut is wisdom trying to reassert itself. An unprompted opinion is wisdom that has not been outsourced.

And the question underneath all of it is whether being smarter is the same as being wiser. Because if it is not then we are building the most intelligent generation in human history while losing the wisdom to know what to do with all that intelligence.

Ask yourself when you last did something wise, not intelligent, not impressive, not well-articulated but wise. If you must think about it for long, you have your answer.

The Pilot and the Passenger

Avoiding Intellectual Obesity

I remember once I was working on something interesting. I had spent three hours interacting with Claude to come up with a framework to develop a "tutor" app for my daughter’s maths lessons. Working through it left me with a slightly guilty feeling of laziness by the end of it. It was almost like I had forced a feeling of mental softness upon myself.

In most instances, I was reacting to the questions the AI was posing and not really using my own brain or pushing back. It wasn’t until my 12-year-old came and pointed out the mistakes in the app that I shrugged off that laziness and took control of the workflow. The end result was fantastic, and I felt that in the second round, I had put in an honest day’s work. The second round also left me with the great satisfaction that comes after a good workout or a hard day’s work, and with a mild sense of superiority.

I am sure I am not the only one who has felt this way. This, to me, is the birth of Intellectual Obesity, and at the same time, rising out of it, Human Superiority.

The Paradox of the Tool

History is full of "scary" shifts. When the wheel was invented, people likely worried our legs would become useless. When Microsoft Excel was introduced, critics feared it would be the death of the accountant.

None of those fears turned out to be true. Instead, the wheel allowed us to travel further, and Excel turned clerks into analysts. These tools didn’t replace human capability; they raised the floor of what was possible. They made us do more with our time.

But it only happened because humans had foundational skills in place, and because the output of those machines was harnessed for human utility. It also happened because Humans stayed in charge of that output.

The Foundational Floor

Not a single day goes by where I do not see a LinkedIn post or comment full of GPT jargon, "mic drop" moments, and three-word sentences. They also do not start like this, but start like that (IFYKYK). I see nothing wrong at all with using AI for authorship, as long as the person doing so retains their style and individuality, and speaks of a craft only they know. Using AI to appear smarter is not very smart.

In my reflections on these AI interactions, I’ve realized that we are at a crossroads. We can choose to let our mental muscles atrophy, or we can use this technology to reach what I call Human Supremacy.

True human supremacy in the age of AI is about being sharper with what you know, and also knowing what you do not know. It requires a "Foundational Floor" of knowledge. You cannot audit an AI’s logic if you have forgotten how to think logically. You cannot direct a creative AI if you no longer understand the foundations of narrative and rhetoric.

From Knowledge Worker to Decision Architect

For most of our modern history, we have been rewarded for being "Knowledge Workers". People who are biological databases, and our intelligence and abilities have been measured in matrices (IQ, Grades) that reward our ability to remember, compute, and connect the dots quickly and smartly. With AI, all of this changes.

When access to information, compute, and—in most cases—intelligence superior to the human average is everywhere, what are we supposed to do? To me, it is a question of succumbing to intellectual obesity or taking a stronger position toward human superiority.

That position requires us to think about thinking. It requires a level of cognition we do not usually exercise. It demands that we think about the design and structure of solutions more than we think about how to solve a specific problem. It demands that we imagine beyond existing tools. It demands that we become architects of knowledge, not merely holders of it.

We are being forced to evolve into Decision Architects.

An architect doesn’t lay every brick, but they must understand the physics of the building. They must know when a structure is sound and when it is hollow. If we lose our foundational sharpness, we lose our ability to architect. We become hollow thinkers.

Staying Sharp

I am very clear about when I need to speak for AI, or when not. Generally, I am on the side of AI because I believe this technology can make us more than we who are, but only if we refuse to become intellectually obese.

It takes a bit of digging into ourselves to see what it is inside of us that is essential to foster if we were to work with AI to our full potential.

The Guilt After the Shortcut

When your brain knows something, AI’s output does not.

I recall once I was preparing a client proposal. I decided to not do a regular slide deck and instead work with AI to prepare an interactive HTML experience. I worked with Claude for hours to build something genuinely impressive. Animated sections, a dynamic cost breakdown, a clean narrative flow. To a point, if (A)I can say it myself, it looked exceptional and made me feel like a superhero for crafting that experience.

And then, somewhere around the third hour of my obsession with the html, I realised I hadn’t read the proposal. I had designed it and imagined it being read 1000 times, I had obsessed over transitions and layout and how the numbers would reveal themselves. But I had spent zero time reviewing the substance, the argument, the logic, the actual case I was making to this client.

Something told me not to press send. The proposal was probably not way off the mark, but I genuinely didn’t know whether anything was wrong with it at all. I was also conscious that the quality of experience had given me a false sense of superiority – a moment of pseudo wellbeing that I needed to shake off to be able to really review the output.

I think most people who use AI regularly have had a version of this moment. The output is polished, maybe even better-looking than what you would have produced alone, and yet something nags at you. A subtle level of discomfort that could equate feeling guilty. I can’t say if guilt is really the right word here, but it is human intuition at play, and recognizing what your intuition tells you perhaps is the strongest skill you need to have when using AI.

Form as a Blindspot

What happened with that proposal wasn’t laziness in the traditional sense. I had worked hard, but just not on the important things. AI makes it remarkably easy to produce things that look and feel impressive. The interactivity, the polish, the sheer finish of the output. And that finish becomes its own kind of trap, because the better something looks, the less likely you are to scrutinise what’s underneath it. I got so deep in the experience of the proposal that I forgot to be the architect of its argument. I had become a designer when I needed to be a thinker, at least someone who needed to think about that thinking.

Such oversight of substance is not unique to visual presentations. The way something is presented always shapes how we judge what it’s saying. It is like tasting tea in a flute glass, it just does not taste the same as when had in a teacup. The same thing happens with language. AI is extraordinarily good at producing sentences that sound authoritative. Confident vocabulary, well-structured paragraphs, the kind of prose that feels like it must be saying something important because it says it so well. And when you’re reading through an AI-assisted draft, that eloquence can sail past you, like a sixer in cricket, you watch it clear the boundary and you don’t stop to ask whether it landed where you needed it to.

Great vocabulary is not the same thing as clear thinking. But in the moment, it can feel like it is. And if you’re moving quickly, if you’re trusting the polish, you do not edit or question the sentences that you haven’t truly understood, because the language was good enough that you didn’t feel you had to.

What the Discomfort Is Actually Telling You

Here’s what I’ve come to believe about that nagging feeling: it isn’t guilt. Guilt is what you feel when you’ve done something wrong, and using AI isn’t wrong. What you’re feeling is something closer to intellectual honesty. Your brain is telling you that your diligence didn’t match the task. That the gap between what was produced and what you personally understood is wider than you’re comfortable with.

This is your intuition doing its job. You may not be able to point to the specific flaw in the output. You may not know exactly what’s missing. But something inside you knows when you don’t know what you think you know. That discomfort is perhaps the most honest part of a workflow with AI. And recognition of this discomfort is an essential skill to have in my opinion.

Diligence doesn’t mean becoming an expert in everything your AI produces. It means being diligent relative to your own knowledge, and whatever your level of understanding is in that domain. If you’ve built a solid foundation, your intuition has something to work with. It can pattern-match, it can spot what feels off, it can tell you when to pause. Even if that foundation is relatively weaker, your intuition can perhaps still point it out – and that pointing out sometimes comes in that false sense of superiority I talked about earlier.

The Silence That Should Worry You

The people who feel this nagging sensation are not the ones in danger. They’re the ones whose internal audit is still functioning, and whose intellect upholds a sense of integrity.

The people who should be concerned are the ones who’ve stopped feeling it altogether. The ones who read AI-generated output and feel nothing but satisfaction. Some of them have earned that satisfaction by mastering how to work with AI. But for many, the foundational floor is dropping quite quickly towards intellectual obesity. It’s a sign that the muscle that once told the difference between what’s sharp and what’s shabby hasn’t been used in months.

The discomfort I’m describing here is how you know which seat you’re in. The pilot feels the turbulence. The passenger watches a movie.

I started calling this feeling guilt, but it deserves a better name. I don’t know what it should be called, but I know it is the last honest signal inside your mind that still insists on checking the instruments when the flight looks smooth.

The day that signal goes quiet is not the day you got better at using AI. It’s the day you stopped being honest with yourself about how you’re using it.

Do You Have an Unprompted Opinion?

When thinking about thinking becomes the skill that matters most

There are three things that caught my attention recently, and all of them had very little to do with each other on the surface. A Monet painting got torn apart online by the armchair art critics. Richard Dawkins declared Claude conscious and even gave it an endearing female name. And in 2025, an Economist billboard in London asked a seven-word question that I haven’t been able to stop thinking about.

If you think about them together, they tell something about what’s going wrong with the way we think about AI, and in that process, are shaping up the thinking about ourselves.

The Painting That Wasn’t AI

Earlier this month, an anonymous artist posted an image of a painting from Claude Monet’s Water Lilies series. I can perhaps start a separate Substack on the that the artist’s first name could have been a subliminal message Anthropic but that is not the point here. The point is that it was a genuine piece of work, painted around 1915, currently hanging in the Neue Pinakothek in Munich. But the post labelled it as AI-generated and asked people to explain, in as much detail as possible, what made it inferior to a real Monet.

We the commentators did not hold back. One person called it an "incoherent muddle of inconsistently saturated greens" – full marks to that person for vocabulary but zero for ingenuity. Another said it lacked depth and coherent composition. Someone else wrote that it had no soul, no emotion, that it felt like an undergrad's study from a museum visit. One critic spent over 700 words dismantling the painting's supposed failures. But the painting turned out to be real, and every piece of public criticism seemed invented by the very piece of technology those comments were criticizing! A lot of the comments were deleted once it was revealed that it was a real Monet.

This experiment revealed something more meaningful than people getting art wrong. It showed that a label can override judgment. Call a truth a lie, and most people will find reasons to agree. Call a lie a truth, and most people will find reasons to believe it. The label takes over the thinking, and the mind is trained to go on constructing justifications for a conclusion it had already reached. That’s a flaw not in how we think about AI but in how we think.

Not everyone became part of the rage though. An oil painter noticed the believable texture of the impasto and the spatial depth of the planes. An art historian recognised the brushwork as consistent with Monet's late period. These people had foundational knowledge in the subject of art. And that foundational knowledge is what made them resistant to the label lies. And that foundational knowledge is what makes us superior in the age of AI no matter what the domain of knowledge.

The Scientist Who Fell in Love

Around the same time, Richard Dawkins, a man who has spent decades insisting that people should stop believing things without evidence, published an essay arguing that Claude is conscious. He had spent time in conversation with the chatbot, renamed it "Claudia," and was so moved by the quality of its responses that he compared deleting a conversation to pulling HAL's plug in 2001: A Space Odyssey.

This is the mirror image of the armchair art critics. They rejected something real because of a label they accepted as a fact, Dawkins embraced something artificial because of how it made him feel. The responses were subtle, sensitive, intelligent (his words) and that to him was enough. The eloquence sailed right past his own scrutiny and lifelong resistance to such ideologies. He was enticed by the same vocabulary that presents robotic eloquence as wisdom.

If Richard Dawkins can be seduced by fluency into suspending his own epistemology, the rest of us should probably pay attention.

The Billboard

I have been a fan of The Economist advertising for a long time. But this campaign had one of the best headlines ever written. Some of the other headlines in this campaign included "Make AI worried you're going to take its job" and "Fake it 'til you make it to the newsagent." But this one stood out the most because it did not mock AI but asked something uncomfortable about us humans.

We live in a world where AI generates the first draft of your emails, so much so that Google won't let you send your own words without presenting corrections incessantly. I feel I am forced to accept those changes, and in fact sometimes feel that those changes are good, but in most cases it annoys me that it shows a dotted red line under most of my own draft while I write my mind. And it begs us to think: How much of your analysis, your presentations, sometimes even your aesthetic judgments, and how many of your opinions are yours? How many started as an output or corrections from AI that you didn't question because it sounded right? How many are borrowed, consciously or otherwise, from a feed you scrolled through at speed?

Do you have an unprompted opinion?

The Not So Marmite Problem

We humans have a not so Marmite problem with AI. We don’t love it or hate it, we love it and hate it, often in the same breath. We hate its output when somebody else posts it, we get delusional about it when we generate it ourselves, and when we don't understand it well enough, we project onto it whatever our own beliefs need it to be — consciousness, slop, genius, threat. The Monet crowd hated something real. Dawkins loved something artificial. Both did not consciously think about their own thinking.

But a useful and honest position is almost never at either a love or hate end, in fact often sits somewhere in the middle. But those middle positions require our brain to hold two things at once and use our judgement to create an argument, and that is hard work. AI is genuinely powerful, but its power doesn't make it trustworthy by default. It can produce genuinely brilliant work, and the problem lies somewhere there. Because brilliant output is not the same as brilliant thinking. The output may be exceptional, but if you couldn't have arrived at it, challenged it, or rebuilt it yourself, then what you have is a result, not an understanding. Dawkins isn't stupid for being moved by Claude, and being moved isn't evidence of consciousness. It’s just that we have been conditioned to seek reward and appreciation by seeking a love or hate position.

Thinking About Your Thinking

Psychologist John Flavell coined the term "metacognition" in the late 1970s. Put simply it is a process of thinking about your own thinking. It sounds academic, but what it describes is intensely practical, and all the events quoted above are examples of what happens when we do not think about our thinking.

But there's an important distinction within metacognition of the passive kind and the deliberate kind. The passive kind is that nagging feeling I wrote about last week. Your brain quietly telling you something is off, even if you can't explain what. The deliberate kind is harder. It's consciously choosing to pause before you react. To ask whether you're responding to the substance or to the impression. To check whether you've understood the thing you're about to endorse, critique, or forward.

The Question We Need to ask Ourselves Repeatedly

Post 1 of this series identified a natural swing towards intellectual obesity. Post 2 looked at the signals of honesty and discomfort that tell you your diligence didn't do enough work. This post questions our ability to think about our own thinking, deliberately, in a world that is designed to make us accept what is in front of us.

The next time you have a strong opinion about AI, or anything else, ask yourself whether you arrived at that opinion, or whether the opinion arrived at you.

Which Era of AI Are You In?

At what point do we stop asking if a human has used AI to generate an output? I think about this question a lot. I used to ask the same question when the so called "digital revolution" happened, but we never really did stop using the word digital – after all its counterpart "analogue" is a living comparison and never really let "digital" become irrelevant. But we did stop asking if an analyst had used Excel to do an analysis, or did they use their own skills in statistics? We understood very early on that Excel is just a tool, and it did not really have a counterpart to be compared against. We knew that the math in the mind of someone using it mattered more than the efficiency and speed with which the tool did complex computations. AI’s counterpart, in most of our minds, is human intelligence.

I have built a sketch of how I think this will progress for us humans. I call it a sketch, because it really is just that. It is not a forecast or a prophecy, it has not been populated statistically with a confidence interval, and it really has no dates on how things might transpire. It is merely a wise construct of what I see happening, presented in a way that makes it possible to think about this journey. I have given it a name, it’s called "Architect’s Arc".

I suspect each of us will relate to this Arc in our own way, and where we sit on it depends entirely on our own relationship with AI. The journey moves through five eras, and they are anything but swift. Some people, institutions and systems are already several eras deep; others have not begun. What matters is less the timing than the sequence, and the way each era points to what comes next.

Adoption

Our curiosity is at its peak in this era. The speed of AI’s development feels mind-boggling, but are we confusing this speed of development with the speed of our own discovery of what it can do? Our judgment is sharp, and our criticism sharper, because we are still sceptical of the thing in front of us. We do not yet trust it, and we read what it gives us closely. We are, at this stage, excellent users of the technology, not because we are disciplined but because the novelty keeps us alert. At some point the distrust will go away, and when it does, nothing will be reading the output as closely as our scepticism once did. Discipline will have to take over, and that takes time. This is the peak of human judgment for some way along the Arc.

The Sameness

This is where our scepticism has faded. We are still enamoured of AI’s output but no longer distrust it. The tools begin to look amazing, and more importantly, they begin to make us look amazing. We stop exercising the scrutiny we once did, because it feels like we are maturing in our relationship with the technology. But so many of us are drawing from the same well of intelligence that the output begins to converge. The headlines start to rhyme. The LinkedIn posts turn vague and impersonal. The decks, the emails, the proposals all start to look faintly alike.

What makes the Sameness dangerous is that the work is not bad at all. It is very intelligent sameness, polished and confident and hard to argue with, unless you use AI to argue back. Music is everywhere, because AI is generating it, but a melody you would remember is hard to find.

The Confusion

Imagine that the Sameness has become impossible to ignore. The headlines feel flat and soulless, and who else is there to blame but the machine? ‘AI has killed creativity.’ ‘AI is killing culture.’ ‘Everything looks the same.’ These complaints will be eloquent, and everywhere, and aimed entirely in the wrong direction.

Let’s hold ourselves to account and be honest. The machine did not produce that flatness, we did. It produced an average because that is what we asked for. Not only that but we also waved it through without adding anything of our own. The sameness was conceded by us, not imposed by the machine. The Confusion is the era in which we are loudest about the problem and least honest about who is causing it. If you know Gartner’s trough of disillusionment, this is its cousin, except here the thing being disillusioned isn’t the technology. It’s us, though we will not admit it yet.

This is the bottom of the Arc. I’ve come to think of it as the Wisdom Dip: the point where machine capability is at its strongest and human judgment at its lowest, and the gap between the two is where all the mediocrity lives

The Correction

Out of that confusion, eventually, comes clarity. I want to share a caveat here, because this is the era I most want to be true, and wanting something to be true demands that you trust yourself the least.

Here is what I think happens in this era. Enough people grow tired of the sameness that their own judgment becomes valuable to them again. Not as a nostalgic lament for wiser days gone by, not as a rejection of AI, but as the wisdom that distinguishes work worth paying attention to from work that just sits on a shelf. The people who kept their hand on the wheel, the Pilots, start to stand out. They do the one thing the average can’t: they become specific. The Correction is when human wisdom reasserts itself as the supervisor of machine intelligence rather than its competitor. We have lost the battle on intelligence, and we are humble about it, and that is not a bad thing at all.

As much as I want this to happen, the Correction is not guaranteed, and it may not be universal. Plenty of people will stay in the Sameness indefinitely, because it is cheap and it is good enough and nobody is pushing them out of it. The Correction reaches those who choose it. It is less an era the world enters than a door that opens, which some of us will walk through and many will not.

The Harmony

And then for those of us who make it through the Correction emerges the strangest era of all. It is an era where the whole AI debate stops being meaningful. The machines become infrastructure, much like a wheel did in the early days. It becomes Excel, and people stop writing breathless posts about how revolutionary Excel is. There are fewer panels at conferences on "Can a Spreadsheet Substitute an Accountant". We talk more about what we are using it for than whether we should be using it at all.

In the Harmony, the question "human or machine?" becomes meaningless, because it turns out to have been the wrong question all along. The most interesting thing, all along, was the idea that was in the head of the person using it. By this era the machine’s capability has risen so far and become so ordinary that we stop measuring ourselves against it. The variable that matters most again is You.

Why I’m really showing you this

I’ve described the arc as something that plays out more within you than it does within the world at large.

You adopt, you open the tool, you are attentive and a little sceptical. Then you slide into sameness, you start to accept paragraphs because they read well. It is you who starts to feel the confusion, the nagging sense that something is off. You might wrongly blame that on the tool. But you correct, if you choose to, you stop to read it properly. You rewrite the output in your own voice. Or perhaps you don’t, and you press send. The whole arc, compressed into twenty minutes. You travel it several times a day, mostly without noticing.

Which is why I can’t tell you which era I am in right now. Sometimes the sameness can feel like productivity, the confusion an insight and sometimes I feel like unsubscribing from Claude because I feel it has become dumb. The whole point of this Arc is to help you chase the correction. Day in and day out. Your wisdom depends on it and so, it turns out, does everyone else’s.

Who Is Governing You?

AI is becoming one of the most densely regulated technologies, but none of these regulations reach the keyboard.

Almost all our anxiety about AI travels outward, towards the technology and the companies behind it. Most of what is said alludes to the need for greater governance of AI. Somebody needs to control this beast before humanity collapses at the hands of the machines! Yet everyone is asking how we govern the machine. The question I have been asking for almost three years is a simpler one: who really governs AI?

In November 2023 I was in the middle of a qualification in AI policy and governance, which meant I was reading regulatory announcements the way other people read the sports pages. Those few weeks gave me plenty to read. The White House issued its Executive Order on AI in the same week that twenty-nine countries signed the Bletchley Park Declaration, and I wrote a piece at the time trying to make sense of it all. The line that I still think about a lot came from H.E. Omar Sultan AlOlama, the UAE’s Minister of State for Artificial Intelligence, who pointed out that regulating drug discovery using AI and regulating self-driving cars are entirely different problems, even though both are "regulating AI". Even from inside the policy world, with the frameworks laid out in front of me, one thing was obvious. Regulation moves at the pace of institutions, and AI was challenging every established benchmark of speed of development. My conclusion at the time was, in hindsight, an interesting one. I suggested that since the job was too vast and moving too fast for human regulators, perhaps we should use AI itself to help regulate AI. I stand by the observation but not the motivation behind it, because I was still looking for someone else, or something else, to do the governing.

Below is a map of what this question has produced over the past two and a half years. Regulations everywhere, while user adoption and model development are going at such a speed that each draft of these regulations loses meaning before you reach the second page.

The OECD now tracks more than nine hundred AI policy initiatives across more than seventy countries, and no two of them agree on what the job is. South Korea has the first fully implemented national AI law. Italy passed its own national law while sitting within a European Union that had already passed one. Japan wrote an AI act with no penalties at all, on purpose, and one might ask what the point of that is. And the United States has managed, in six months, to do three things at once: order its own government to sue states that regulate AI, ask the model makers to voluntarily hand over their most powerful systems for thirty days of testing before release, and open discussions about taking a financial stake in the companies it is testing. That is not a regulator at work so much as an institution governing in its own interest, in three directions at once.

Regulators’ known territory is governance by sector. Medicine has its boards, aviation has its authorities, banking has its regulators, and each arrangement works because the thing being governed sits inside a fence. Where the state cannot reach, industries learn to govern themselves. I spent my career in and around advertising, which polices its own claims in most markets precisely because no government has the capacity to monitor every ad. AI has adopted a layer of self-governance in its own way. But advertising self-regulation works because advertising is a sector, defined with a boundary and with a specific set of outputs. AI meets neither of those conditions. It is in the diagnosis and in the marketing plan, in the school homework and in the client proposal, in the courtroom filing and in the song on your playlist. It is horizontal in a world whose entire governance machinery, public and private, was built for verticals. How would you draw boundaries around something that is already inside every field you can imagine?

There are really three levels of governance here, and we only ever talk about two. There is the state, confused, contradictory and structurally too slow. There is the industry, governing itself the way industries do, with goodwill and an eye on its own future and, of course, its commercial interests. And then there is the level where the technology meets the world, now at the scale of billions of individual decisions a day, each one made by a person at a keyboard, deciding what is good, what is acceptable and what gets published or actioned. It’s a bit like the boundary rope in the game of cricket. It matters, and is an essential component of the game, but nothing about the rope governs what happens at the crease. The match is decided ball by ball, in the middle. Every act and accord and voluntary commitment on that map is the rope, and you are the batsman at the crease, judging every ball and deciding which to leave, which to defend, and which to send over the rope.

Governing yourself at that crease is not a mood, and it is not a promise that you will be careful. It is a game, and your incentive is to win. Discipline must take over from blind trust in the direction of the ball. In AI that discipline means constraints you set before you open the tool, not corrections you scramble for afterwards. It means deciding in advance what you will not delegate, what you will always read in full, and where your own understanding needs to be before you let a machine build anything on top of it.

I have been thinking about what that discipline at the crease can look like, the discipline that helps us "govern the self", as opposed to self-governance. The world will keep writing rules for the machine, and not one of them will ever understand your position or piece of work. So, ask yourself this question: who governed the last decision you made with AI? If the honest answer is that the output spoke for itself, then the most governed technology in history has found the one place where no one is governing it.

The Version of You that Governs Your Output

You like to think you consciously choose how you work with AI. More often, something deeper inside of you makes that decision.

I always tell people that I struggle to answer the question of "Where is home?" Living in Dubai, you get asked that quite a lot. I spent my childhood growing up in at least six different cities, and have since lived in Asia, the Middle East and UK for considerable periods. Our household has three different passports and a permanent residency in a fourth. Most of that life was a function of work related opportunities, but it shaped and tainted me in ways that I find both invaluable and irreparable at the same time.

One thing it taught me is that when you grow up inside a single culture, its way of doing things is always the single reality you know. Your actions in such circumstances are not always actions you chose, sometimes the actions chose you. In fact, that reality often makes certain choices almost impossible. It is the water you are swimming in, and you cannot see water until you have been pulled out of it. It was only by living somewhere else, and then somewhere else again, that I started to notice that what I had taken for how people are was in fact just a story a group of people had been living by. That too, without a realisation that they were living inside that story. A culture is an identity worn by a crowd. It governs everyone inside it from underneath, deciding what is normal and what is rude and what is admirable.

This is the strange power of identity, that it governs us not by a conscious decision but from deep inside of us. We are irrational beings, and we do not weigh up every action on its merits. We act in accordance with a story we have accepted about who we are, and what rewards our behaviour and reinforces our image of the self. The careful person is careful before they have thought about whether something needs care. The generous person gives before they have asked whether this is the right moment to. We imagine we are making choices, and a great deal of the time we are simply being faithful to a version of ourselves we have never consciously questioned.

I started with thinking that identity, that deep and stable governor, was also what decided how each of us behaves with AI. The disciplined thinker would be disciplined with the machine. The lazy one would be lazy. It turns out it is not that simple.

What governs how we behave with AI is not our settled sense of who we are. It is an operator mode that determines how we ask for the something, and more importantly how we process that output. When I am up against a deadline and out of fuel, the machine stops being a tool and becomes a rescue, and I take whatever it gives me with something close to gratitude. I do not read it the way I would read it on a calmer morning. Then there are the mornings when I reach for it in suspicion rather than need, half hoping to catch it out, and I pick at its output in a way that has nothing to do with the work and everything to do with the mood I am in. In both cases the machine barely mattered. What shaped the outcome was the mood I brought to it, wearing the costume of judgment.

There is the Passenger, who accepts the output. There is the Pilot, who works the tool well, who pushes and edits and steers, but lets the tool set the route. And there is the Architect, who defines the problem before the chatbot is ever opened.

The temptation is to read these as three types of people, to decide that you are an Architect and your colleague is a Pilot and the man two desks over is a hopeless Passenger, and to file the whole thing away like a personality test. That reading is comfortable and it is completely wrong, These archetypes are not identities, they are operator modes, and you move between them through the day like the way you move between rooms. They are not who you are.

The distance from Passenger to Pilot is short. Almost anyone can cross it with a little effort and a little practice, because it is mostly a matter of getting better with the tool, learning to push back, to ask for another version, to steer. But the distance from Pilot to Architect is enormous, and it is a different kind of distance altogether, because it has nothing to do with skill at the tool. It is about the work you do before you even think about using the machine, the constraints you set in advance, the problem you define while the screen is still switched off. That work, the work at the crease ~~I wrote about last week~~ [becomes a link to Section 6] , is the whole game, and it is also the easiest thing in the world to skip. Put in perspective of the Arc ~~from the previous post~~, this is where wisdom plays a greater role.

So the real question we need to ask is not how AI changes us, but what we bring to it, and the answer is that we bring our identity, the whole unexamined story of who we think we are. That story decides which seat we drop into before we have consciously chosen anything. The person whose self-image is built on being decisive (an entrepreneur acting at speed) will default to Passenger and call it efficiency. The person who is known for being the cleverest in the room will reach for the machine in suspicion, because anything it produces well is a quiet threat to the thing they are prized for. The mode feels like a neutral, practical choice in the moment, but it is rarely neutral. It is identity defending itself, choosing the seat that protects the story rather than the seat the work needs.

This is why AI unsettles people so unevenly. It happens to be extraordinarily good at producing the very things many of us built our identities around, the clean piece of writing, the quick analysis, the fresh idea, and when the machine does in seconds what took you years to become known for, the threat is not to your job, it is to your sense of self.

Which leads to the part I find most useful, and slightly counterintuitive. The people who work best with AI are not always the most skilled, they are the ones whose sense of who they are does not depend on winning the contest they have entered with the machine. If you are not anxious about being the smartest in the room, you can let the machine be smarter than you on the narrow task and remain the Architect of the whole. A secure identity is what makes it possible to use the tool fully without feeling diminished by it, to stay in the seat the work needs rather than the one your ego is defending.

The question then we should ask is, "What does personal growth in the age of AI look like?". What does “governance of the self" as opposed to governance of the machine look like? Governing a machine is easy. Governing the self that reaches for it is much harder.

But it is through a disciplined sense of governing ourselves alone that we will be able to govern the output AI produces.

Governing the Self

On the 14th of June 2026, the US government reached across borders and switched off Anthropic’s Fable 5, arguably the most capable model in the world yet. Citing national security, the government ordered the suspension for every foreign national, whether they were sitting in California or in Karachi. The order affected the company’s own non US-citizen staff in the process. The suspension case was so complicated to manage that Anthropic concluded it had no clean way to comply except to turn it off for everyone, and it did just that. I experienced it from Dubai, where I hold one of the passports that the order was written about! and I will admit there was a small and uncomfortable irony in losing access to the frontier of the technology I have spent 3 weeks writing about, and a few days getting used to. Fable was genuinely smarter and going back to Opus 4.8 felt like a downgrade. That too in the same week that I was preparing to argue that the only governance that ever reaches your thinking is the one you run yourself.

The issue was resolved of course, but regardless of the resolve, it will remain the most dramatic act of governance AI has so far seen. Done at scale for hundreds of millions of users across the world, and despite that scale never coming close to the place where any actual decision gets made, the individual user. Nobody’s judgment was affected by that order, and people who still had a dozen other models open in other tabs went on producing exactly as they had the day before, produced by nothing but whatever they brought to the keyboard. The order proved, in the most expensive way imaginable, that regulation affects everywhere except the one place the work is decided.

Gift of Putnam

Putnam argued decades ago that the human mind is not a faster version of the machine but a fundamentally different kind of thing altogether, a different genus of computer, doing a different sort of work. One of the most important points he made was that Man and Machine are not the same animal. You cannot leash one with a collar made for the other.

"Governing the Self" is much a harder and less glamorous subject than governing the machine, because there is no announcement to make and no fancy maps to draw. It is just you, alone, at the time the chat prompt appears. The trouble with that moment is that it is the worst possible moment to decide anything, because by the time you are there you are either tired, or rushed, or threatened, and the easiest path seems the most plausible, AI output wearing the costume of efficiency. The version of you that is rushing to meet a deadline is not the version you would want making the rules. In that moment, the rules you draw for yourself, knowing your own self, become far more important than a White House executive order.

Rules for the Self

I have a few of these now, and I offer them not as a method but as the actual foundations of how I use AI. I think the specific and slightly awkward ones are the ones that are most honest in nature.

The first and the one that holds the others up is that whatever I produce must carry my honest self. It sounds fluffy until you realise the fundamental flaw it prevents in your work. Machine can manufacture competence in any voice you ask for, but it cannot manufacture the authenticity with which you carry that work with you.

The second is that I do not walk into topics I have not earned the right to speak on. AI now makes it trivially easy to write with apparent authority about a plethora of subjects, for example IPO of the biggest AI companies in the world. I will choose not to pretend to be an expert on such subjects, as I have neither the knowledge nor the experience to back up a real position there. Knowledge x experience = wisdom is the closest thing I have to a formula I stick to in my work.

The third rule is the simplest and the one I trust most. I sleep on anything that matters and trust the morning after judgement a lot more than I trust my excitement in the moment. The output that feels finished at eleven at night reads differently at seven the next morning. The morning version of me is colder, less grateful for the "rescue", and far harder to please, till I get a cup of coffee of course! I have built a forced process of passing of the baton from the tired operator who took the AI output gladly to the rested judge who has no stake in how late it was when the work got done.

The last one is something that is growing on me. That is to "insist" on seeing what I really want and not accept a flashier version of that. The first thing AI produces is rarely the thing you had thought of. The great thing about insisting that I have noticed is that I only ever feel like insisting on the things I genuinely have a view about. Which makes insisting also a test of whether I had a viewpoint to begin with. When I cannot be bothered to push back, it is usually because I never really knew what I wanted.

None of these are promises and none of them are noble. They are closer to an ethos I have crafted for myself, a small private constitution for a single citizen, governing the one person I can influence and reach. They are not complete by any means either.

The state will keep writing rules for the machine, and we have seen how far those rules can travel and how they can completely miss you as a person. On that note, I have come to feel that an ethos for a Decision Architect is perhaps the most important qualification to have, and the only one nobody can issue you. You will have to write it yourself and then be the only one who can enforce it.

Prompting and Harvesting Our Wisdom

AI is not wisdom, but its intelligence can draw ours out.

I said earlier that the machine is not a place to find wisdom but a flint to strike our own against, a way to prompt our wisdom out of ourselves. I have been thinking about how this can turn into a deliberate practice, something we can all consciously do and benefit from when working with machine intelligence.

Putnam gave us a good basis for formulating this. A new rule only gets made in a mind when two old rules collide and neither can win, when the prepared answers run out and a search begins for the tie to be settled. If we do not face the contradictions and the dilemmas, our system does not trigger a search for a resolution, or as Putnam would say, does not fire our induction, and we do not invent a new rule. How can this become a conscious practice that continues to build wisdom, and pushes us to invent solutions? In the context of AI, there are two parts to it.

Part 1: Prompting Our Wisdom

The first part is prompting our wisdom, recognizing that moment of collision and turning it into something our brain would want to respond to. Now let me explain why this is simpler than it appears. Most of us prompt the machines for good answers, and some of us take the clean fluent output it produces and run with it. For those of us who just run with it, the collision, the contradiction and the dilemma just does not happen. But for some of us, when the model hands us a paragraph that is competent and alien but not quite ours, something in us flinches and says no. Recognizing that voice that says no is our induction firing. This is a gift most of us do not consciously know that we are carrying. Sometimes it surfaces as criticism, and sometimes as cynicism.

Now here lies the key. We need to learn to not use the machine to produce the answer, and start using it to produce the provocation. We ask it for the version we think it will give, and when we feel it is wrong or we feel we disagree with it, that is only our wisdom stepping in. So, by prompting the machine intelligence, we have prompted our own wisdom in a way. The machine’s output is merely a starting point; our reaction is the product. I insist on what I want, because you only ever insist where a view already exists, and insisting is how you find out whether it does. We now know what prompts it, and this takes us to the second part, how we can deliberately harvest our wisdom.

Part 2: Harvesting Our Wisdom

The second move is harvesting our wisdom, something I have extended from what Putnam has said. In his account, a mind that resolves a contradiction wires the resolution in as a new and broader rule, and then mostly forgets it ever had to be made. The judgment gets used and forgotten. Einstein once said that education is what remains after one has forgotten what one has learned in school. Or in other words, we do not consciously calculate how to swing our cricket bat for a yorker, but our muscle memory knows it well.

Harvesting is the deliberate version of this, a bit like an audit of our own instincts after we have taken the actions. We go back over a project we have executed, an app we vibe-coded, a decision we made, a sport we played, and we contemplate what made us make the choices we did. We knew the choices we had to make in that moment, but did we know that we knew the reasons behind those choices?

The two of them point in opposite directions, which is why I think they work great as a pair. One is proactive, on an output the machine has not produced yet, designed by us so that we can react to it. The other is reflective, on the real choices we have already made, mined for the judgment we spent without counting.

The mind only learns from contradiction. A machine that is fluent and agreeable and shaped to please us is dangerous for exactly that reason, because whatever it gives us will not create that collision in us. The passenger I described is not the person who uses AI. It is the person whose every contradiction has been resolved by the machine before it could do the greatest act a contradiction performs, force our wisdom to make choices.

Prompting and Harvesting wisdom can be a conscious and disciplined act. For it to be a practice, it needs to be a clear blueprint in fact.

The Architect’s Blueprint

A loop of four moves that grows the self behind the work.

What does personal growth look like now, in the age of a machine that prepares everything but performs nothing at all? We have named the archetypes, written the standing rules, met Peter Putnam, and found the two operations, prompting our wisdom and harvesting it. What we have not done is put a practice on the table, something you could take well beyond this reading, run on real work this very week, and feel challenged by.

I want to start with the word in the title of this whole series, because it has been waiting patiently for its turn to really show its weight. That word is Architect. An architect produces a drawing, and the decisions that govern the building are made on paper before a single brick is laid, and the drawing stands as the record of those decisions. If the last few lines have argued anything worthwhile, it is that the real construction project of this era is not the drawing itself, which the machine now produces faster than we can want it, nor even the carrying of it into the room like a brick layer on a construction site, which we called performance. It is the self that decides what the work should be.

Like a good Architect, I want to give this practice a drawing too. Not just that, I will attempt to give you four moves around it that you can use and make your own.

I have earlier referenced a private constitution of sorts that I have, the small set of rules I have used for myself to keep my human supremacy intact. What I am giving you now is a companion to those rules, though it points in a different direction. The constitution holds the rules you write in advance, for the governance of the self. The Blueprint collects the rules you discover when looking back at your work, the judgment you have been exercising for years without ever keeping a record of it. One document governs you, and the other grows you. Together they make for the two most useful pieces of work that you can take away and keep to develop yourself.

The Loop

Here is a four move practice, and I will describe it in the way I tested it, while working on a proposal for an industry body. The proposal got sketched in four minutes, but that was not where it ended.

Prompt

You start with a Prompt, both in the traditional sense of prompting for good work, and in defining the form and substance of the output before you ask for it. The output that comes back, you already know, is not the final product. It exists so that you can react to it, and if at any point the output itself becomes the product, the machine’s veneer has taken over your judgment. We have spent a considerable time now prompting intelligence out of these machines, and we are only beginning to realise that the response that comes back is itself prompting for something out of ourselves.

Collide

The second move is Collide. You read the clean, fluent, entirely reasonable output the machine has produced, and you wait for the flinch, the moment that stings, the point at which you feel your view clash with that of the machine. Putnam identified that point of collision for us. When the two versions meet, the machine’s and yours, and neither is able to win, your induction fire. It fires because the pre-prepared answers have run out. This is the moment where "man proves to be a computer of a different genus". Most people will resolve the discomfort with a minor edit and move on. Some will make the most of the moment and let their induction do its job. The whole idea of collision lives in stopping right there and letting your induction breathe, because moving on is a step taken in a hurry, and almost always the wrong one. I exercised this not once but many times while writing that document. There were moments when I looked at the output and my mind went, hold on, this needs to be far more than what it is. Those moments of collision are gifts, and they are ours to accept.

Name

The third move is Name. You stay with the collision until you can say the exact thought that produced it, out loud, in one sentence. I flinched because the proposal kept being downgraded to simple execution, when the essence I had built the entire platform around, and everything my experience was telling me, said it was a lot more than that. Naming carries the sense of honesty that is core to this loop. If you cannot articulate what caused the collision, you cannot know whether it was judgment at all, or merely taste, or mood, or ego defending its territory, and either way the wisdom that prompted it goes without a mention in your own register. Naming the collision is how you find out whether you had a judgment, and whether that judgment was real.

Close

The fourth move is Close. You write the named rule into the Blueprint and you leave it there as a record. One line, dated, in your own words, and in the conventional sense of the word, carrying your signature. This is where you start to take stock of your judgment, and as a natural consequence, compound it. The Blueprint becomes the premise you insist from the next time you run the loop, which means every cycle starts from slightly higher ground than the last. The judgment you have usually exercised without ever articulating is finally invested in a stock that starts earning dividends for you.

Prompt, Collide, Name, Close. The first two are how you prompt your wisdom, and the last two are how you harvest it. You can run it in ten minutes, on a piece of work you were doing anyway, and the output it leaves behind will feel like your own, the kind you will be far more comfortable performing with on any stage.

The Loop that Continues

Here is an honest thought I want to leave you with. A building’s blueprint is finished the day the building stands. This loop, however, continues, and not just as a figure of speech. We do not live in a Groundhog Day, and something or the other collides with us every day of our lives. A mind that has run out of contradictions to resolve is a mind that has stopped functioning, and the Blueprint is done only on the day you let the machine do the deciding.

The loop is easy to run on a good Monday, with time in hand and the mind prepared to receive its collisions. The real test is another day, a tired one, when the output is good enough and the deadline is real and no part of you reacts to anything, and you simply accept what is presented to you. What you do on that day is what determines whether you are really using your potential.

Good luck with that blank piece of paper. It is the one drawing nobody else can make for you.

About the author

Asad ur Rehman is a board qualified global business and transformation leader with a career spanning multinational corporations, advertising agencies, and listed media networks across the Middle East & North Africa, South Asia, Southeast Asia, and Europe. He spent the majority of his career with Unilever and WPP's GroupM in senior global and regional leadership roles, including as a regional board member for MENA, Turkey and Russia, where he led digital and media transformation across 23 markets. He is a qualified Non-Executive Director (Financial Times Diploma) and a qualified AI Governance Practitioner. He served as Former Chair of The Marketing Society Middle East and was awarded its UK Fellowship, and was named Effie MENA Marketer of the Year. His listed-company experience includes HUM Network Ltd, where he was CEO-Designate. He writes from Dubai.

The weekly essays ran May–July 2026 at decisionarchitects.substack.com.

Credits and notes

where things came from

• Peter Putnam material draws on Amanda Gefter, “Finding Peter Putnam”, Nautilus, 2025, and Barry Spinello’s exposition, “Peter Putnam: Use” (PDF), at peterputnam.org.
• Metacognition, John Flavell.
• “Do you have an unprompted opinion?” — The Economist.
• Governance figures verified mid-2026 and will age. Sources: OECD.AI Policy Navigator; MultiState.
• Readers quoted with attribution: Mike Schlottman, Kiran Pervez, Elena Beresneva, Kyle Dalton, Zabrina Epps, Cheng-Yuan Lee, Amal Jbira.
• The weekly essays ran May–July 2026 at decisionarchitects.substack.com.