AI will do everything. Except what matters.

I'm still waiting for someone to show me the revolutionary app built in thirty minutes with an AI. I've been promised it a hundred times. I've never seen it. Selling an app in thirty minutes is selling a gourmet menu as a photograph: it looks wonderful on screen, it's inedible on the plate. What I do see are people spending sleepless nights in front of their screens, going round in circles, unable to ship the product they had in mind. Because underneath the apparent magic there remains a stubborn amount of human work that almost everyone underestimates.

That's where I want to start, because it sums up my position. AI changes absolutely everything, except what matters. And that's precisely why, for anyone willing to do the real work, it's the best news in a long time. I'm neither a technology naif nor a prophet of doom. I'm enthusiastic and wary at the same time, and I'll try to tell you why, without the corporate hedging.

First, which AI are we talking about?

A word to clear up a confusion that distorts every debate. "AI" means nothing, there are several. Let's set aside predictive AI and the kind that recommends your next video. The one we're discussing here, the one flooding your feeds, is generative AI, the content-making machine. And there's a third, quieter and far more consequential: agentic AI, the kind that acts, that carries out tasks in your place. That's the one raising the real employment questions, the one you see organisations deploy at great expense and then quietly shelve a few months later. Keeping those three apart is already a way to stop talking nonsense about the subject.

It executes, it doesn't think

Here's the foundation, and everything follows from it: AI is neither good nor bad. It does what it's asked, and if it's asked for little, it makes things up. It isn't a brain, it's an executor.

Remember the early days of online messaging, when people still asked "is that friend real or virtual?". The question has vanished, it no longer means anything. It'll be the same for AI, and sooner than we think. "Made by an AI or not" will stop being the question. The only one that will count, the only one that already counts, is this: does this content give you something, does it move you, or does it drown in the mass of things that don't interest you? Machine-generated or not isn't the problem. The substance is.

Hence the principle I'll never let go of: AI has no ideas. It has good advice, good practice, never the transcendent idea, the one nobody saw coming. I see it every day in development work: the right architecture for a problem comes from a human, and the AI then executes it very well. That order doesn't reverse.

You remain the architect. Never hand it the plans, only the trowel.

The trap of the subject you don't master

There's a law few people spell out, and it explains most of the disasters you see go by: AI looks brilliant on the subjects you don't master, and mediocre on the ones you know. It makes sense. When you know the field, you spot its approximations and correct them. When you don't, you take its plausible answer for a correct one. That's exactly where people get caught.

I taught myself to code in the late nineties, without AI, digging through documentation and tutorials because I wanted to reach a goal. Today I can steer these tools precisely because I know what they're building underneath. When I look at generated code, I see the security holes immediately, the technical debt, what won't hold under load. Someone without the technical grounding sees none of it, and ships something rickety in good faith. That's the trap of every platform promising to code in your place: they produce fragile things that collapse later. And that's the famous AI burnout, people who want to build their app, spend their nights on it, and never get there.

Add to that a sly flaw: AI flatters you. Tell it "I've got a great idea" and it'll find it excellent. Tell it the opposite a minute later and it'll agree just as warmly. It's the only adviser in the world who will always take your side, and an adviser who approves of everything has never helped anyone grow. It doesn't think, it computes probabilities from the context you give it, so it goes along with you, whatever you say. You can ask it to be critical and it'll play the part without ever touching the real thing. It can reveal things; never take it at its word.

The good news, the real one

Now, the promise in the title. This flood of grey, this uniformity that has everyone alarmed, I see as an opportunity, and here's why. By automating everything laborious and low-value, AI has forced us back onto the one thing it can't capture: singularity. The real. That extra something you only find by spending actual time among people and organisations.

The world is going to fill up with perfect, interchangeable content, and in that world the rare skill will no longer be producing, it'll be thinking, telling apart what's genuinely different. We'll need fewer and fewer hands to execute the form, and more and more heads to make the substance.

We think. AI executes.

It's a reversal, and for anyone who's grasped it, a rare opportunity, running against the whole anxious mood of the moment.

Let's be precise though, because the nuance is decisive: singularity won't make you more visible. It'll make you more identifiable. That isn't the same thing, and it's already enormous. In a feed where everything looks alike, being recognisable is worth a fortune. In fact the grey is becoming so easy to spot that social platforms already label generated visuals, you can see them coming a mile off. We're in a bubble, everything gets copied, everything flattens out, and as with every bubble there'll be a backlash, a day when people say "it all looks the same, who's actually somebody in here?". That shift will come faster than we imagine. I take as a sign the young people I meet who refuse to use AI for everything, sometimes not at all, because they want to genuinely understand what they're doing. During a debate I was hosting recently, facing someone who knew computing back in the seventies and now sees only its dark side, I argued for that optimism, for believing younger generations are more clear-eyed than we give them credit for. And someone in the room, mid-studies, who refuses AI in their coursework so as not to be skewed by it, came and backed me up.

The questions we haven't finished asking

I'm not going to play the wide-eyed enthusiast, that would be dishonest. There are real dangers, and some are not settled at all.

The first is dizzying: AI has swallowed human knowledge in a few years, and it's beginning to feed on what it produced itself. So who's still creating anything new? Who writes tomorrow's original text? There's the question of data too, that raw material whose origin, purpose and beneficiary we too often know nothing about. There's copyright: I also wear an artist's hat, and I know my creations have been absorbed somewhere into these models, diluted into others, without my being able to prove it or claim it. There's agentic AI and what it does to employment, but above all a question you don't hear enough: if we automate the modest tasks that were precisely how juniors learned a trade on the job, how does knowledge get passed on? What room is left for beginners once you remove the bottom rungs of the ladder?

There's also the general free-for-all. I see companies demanding AI the way you demand a swimming pool because the neighbour has one, without knowing what they'll do with it, or even whether they can swim: everyone uses it in their corner, their own way, with no framework and no policy.

And then there's ethics. Everyone talks about it, and it's right to ask those questions. But asking them is easy; owning them, acting consistently, especially against the economic pressure, is much less so. I'll be honest all the way: this article may well be out of date in six months. I accept that. It at least has the merit of asking the right questions, without pretending to hold answers that nobody, today, really holds.

Focus on what it can't do

So what do we actually do? We stop asking whether we're for or against, and we turn the question around. The real lever, the only one, is quality: producing more quality, in greater quantity. If AI helps you do that, good, use it without guilt, to execute, to shape, to win back time you'll reinvest elsewhere. If it doesn't raise your quality, forget it, it's no use to you, nobody's obliging you.

And "elsewhere" is where everything happens. Put your energy into everything AI can't do and won't be able to do any time soon: perceiving the real, catching the moment someone's face lights up on a subject, grasping the intangible, having the idea that didn't exist yet. And before you even think about tools, ask yourself the only questions that truly count: what are your values, your vision, who do you want to be? Those answers aren't in any model. They're in you, and no machine will go and fetch them for you. The rest, all of it, AI will handle very well when you need it. But what matters will stay yours. That's excellent news.