The Mirror or the Magic Wand? An age-old story, explosively relevant in the age of AI.

The last piece ended on the courage to face the elephant in the room: are we still relevant, and to whom? This one is about what happens next — our reflex, the moment the hard question appears, to skip the reflection and reach straight for the magic.

Here is a story I am sure you know. A young apprentice, gifted and impatient, is left alone in his master's workshop with a dull chore — haul the water — and just enough of the master's craft to be dangerous. So he enchants a broom to do the hauling for him. And it works. Gloriously. The broom hauls bucket after bucket, and the apprentice sits back, delighted with himself. Then the tub is full, and the broom does not stop. It will not stop. He does not know the word that ends the spell, so in a panic he grabs an axe and splits the broom in two — and now there are two brooms hauling, then four, then an army of them, and the water rises past his knees, and the workshop floods. The magic was entirely real. That was never the problem. The problem was a power that arrived faster than the judgment to wield it.

In Goethe's telling the master returns at the last moment, breaks the spell with a word, and is furious — the apprentice is bailed out, humiliated but dry. Here is where our version parts from Goethe's, and it is the part worth sitting with. Reach for the AI wand without reading the manual first, and no master walks through the door. No one breaks the spell. No one is even angry with you. You simply stand in the rising water, having made no progress at all — which, as we'll see, is its own kind of disaster, just a quieter one than Johann Wolfgang staged.

So before we talk about the wand everyone is selling, it is worth borrowing the one thing every one of these stories quietly insists on: look first.

Here is the thing about mirrors: they only ever show you what is already there.

AI hands you a power you did not have. It could also hand you an invaluable reflection — but probably one you would rather not see.

AI is a genuine wand. It hands people capabilities they did not have the week before — I have lived it; I created things with it I could never have built alone, and so have you. This piece is not a case against the wand. It is a case for the other half of the instrument nobody is selling.

Because the same tool is also a mirror, and the mirror is the half that decides whether the wand serves you or floods the workshop. A wand changes what is possible. A mirror changes what is visible. AI is both at once — and we are so dazzled by the first that we almost never look into the second. Every pitch you will hear is a wand pitch: wave it, and what you could not do becomes what you now can, the Magician's promise from the archetypes piece, the wonder weapon that will surely work in your hand too. What the pitch never mentions is the mirror standing right beside the wand, turned away.

Here is what that mirror does, quietly, beneath all the new capability. It does not add to you and it does not improve you. It reflects you — returns what is already there, faithfully, from a vantage your own eyes cannot reach. Stand something clear in front of it and it gives the clarity back, plainly. Stand something confused in front of it and it gives the confusion back, just as plainly, with no flattery and no edit. The tool's power is identical in both cases. The only thing that differs is what you set before the glass — and that reflection, not the spell, is the most valuable thing an AI engagement could produce. It is also the thing nobody puts on the invoice.

So the reflection is only ever as useful as the thing you put in front of the glass. Hold up the wrong thing and it faithfully reflects the wrong thing back. Consider what it returns in each case:

The productivity that never arrived is not the wand failing. It is the mirror, reflecting exactly how ready we were.

If AI were only a wand, three years and several fortunes in, we would see it in the numbers. We do not. A 2026 NBER survey of nearly 6,000 executives across four countries found roughly 90% reporting no measurable change in productivity over three years, with production-grade adoption still in single digits. The capability is genuinely there. The result, at the level of the whole organisation, mostly is not.

The comfortable reading is "the technology is overhyped" — the flattering reflection, the one we reach for first. The honest one is harder: productivity is an output of readiness, not of tooling. The gap between what AI cost and what it changed is not the measure of a weak wand. It is the mirror doing its only job — reflecting, without mercy, organisations that grabbed the wand and never once looked in the glass. They reached for the spell and skipped the reflection. The broom hauled. Nobody had decided where the water should go.

You have heard a worn version of this — "AI doesn't fix broken processes, it exposes them" — by now the keynote line everyone nods along to. But "exposes a broken process" is still the comfortable reflection, because a process is something you can fix without changing who you are. The reflections that actually cost something sit lower: the model, the relevance, the reason a customer still chooses you. The mirror shows them mercilessly, without emotion, true to the facts. We are just too well trained at not looking.

And the mirror is getting clearer, and sharper too. Ethan Mollick's "jagged frontier" — AI is brilliant at what looks hard, hopeless at what looks easy — was for two years the favourite reason to wait: look at all it still cannot do. Mollick now notes the frontier is narrowing; last year's reliable examples of AI incompetence have mostly closed. A clearer mirror does not make the reflection optional. It makes an honest reflection sharper and a comfortable self-deception more expensive, faster. The glass improving is not your reprieve. It is your deadline.

A mirror is not dumb. It is honest — dull by design, and that disinterest is exactly why you can trust it.

Now the part that decides everything, and it is not a feature of the tool. We should retire the idea that the mirror is "dumb" — and resist the urge to fix it by reaching for a "smart" mirror that interprets and improves the picture for us. That instinct is exactly backwards. A mirror's whole worth is that it does not interpret. It reflects what stands before it without flattery, without agenda, without an opinion about what you would prefer to see. The mirror that tells you what you want to hear — fairest of them all — is the one in the fairy tale, and you remember how that story ends. The honest mirror is dull by design, and the dullness is the point. It is the one thing in the building with no incentive to lie to you.

Which means the entire value of the reflection rests on the one thing the mirror cannot supply: the quality of what you put in front of it. And that is a question, not a technology. Aim it at "how do we use AI to write our emails faster?" and it will faithfully show you a company that writes emails faster. Aim it at "what do our customers actually hire us for, and is that still us?" — the elephant from the last piece — and it shows you something far more useful and far less comfortable. Same mirror, same technology. The only variable that changed was the question.

It stood in the forecourt as a warning: any answer you carried out of that temple, if you had not first understood who was asking and why, would be misread, misapplied, or simply wasted. They grasped two and a half thousand years ago that the oracle was useless to a visitor who had not done the harder work first. Swap the oracle for the model and not a word needs changing. Consult the most powerful Generative Pre-trained Transformer ever built without first knowing yourself, and whatever it hands back will be misread, misapplied, or wasted. The apprentice never read the book of sorcery, nor the inscription. He went straight to the spell.

This is where the advisor earns the room, and where most do not. Left alone with the mirror, an organisation frames a more comfortable question every single time — because the comfortable question comes with a deck and a deadline, and the real one may not even have a name yet. The work is not to operate the mirror; anyone can hold up glass. The work is to put the right thing in front of it — to aim it at what you would rather not see, and to stay in the room when the reflection arrives. The Jester from the first piece is the one who can: not selling you the wand, and therefore free to point at what the glass is showing. The consultant who sells the wand alone is selling the shortcut — and the shortcut is the apprentice's road.

And lest this sound like a metaphor stretched too far — it has a physical seat. My son, a psychologist who crossed over and went deep into data science, is the one who gave me this part of the thinking. He points out that recognising yourself is not the same act as seeing the world; the brain treats them differently. Reflecting on who you are runs through the mPFC, the medial prefrontal cortex, and it does not merely store what you are — it weights it by what matters to you. Self-recognition is effortful, distinct, and, offered the choice, the very thing we are most inclined to skip in favour of the easier view out the window. Which is the whole problem in a sentence: the wand points the gaze outward, where looking is easy. The mirror asks you to turn it around — and turning it around is work the mind would honestly rather avoid. That outside-in view of yourself, the one your own eyes cannot give you, is precisely what the honest glass is for.