My relationship with AI: between working tool and silent normalisation

Do you remember Clippy? That pixelated assistant that appeared in Word 97, represented by an animated paperclip with eyes, which would pop up unannounced the moment it detected (or believed it detected) that you needed help. I remember Clippy. “It looks like you’re writing a letter.” “Would you like some help?” The answer was almost always no. Or rather: you closed the window, ignored it, and carried on with your letter in Comic Sans MS and your rainbow WordArt. But you did not disable it straight away. Just in case.

That “just in case” always struck me as the right posture towards a tool whose value you could not yet quite measure. Clippy was clumsy, often wide of the mark, sometimes genuinely useless. It did, however, already raise an interesting question: can a machine anticipate what I need before I have put it into words myself? The answer was no, back then. Or so rarely that it did not count.

Twenty-five years later, I am asked a question I hear more and more often, in meetings, in messages, sometimes with a note of anxiety in the voice: “How do you work with AI?” The implication being: do you use it, have you got to grips with it, is it changing your work, is it going to replace you? It is that question I want to answer here, honestly, without posturing, and without pretending I have it all figured out.

Execution is a done deal. So what?

Let us be direct: on certain tasks in my work (web design and web development) I simply cannot compete. When I have read the same block of code for the fiftieth time and my brain slides right past the error without seeing it, AI spots it in three seconds. When I need to write content in a language that is not my native one, it produces a solid draft I could not have written as quickly, nor as cleanly. When I tested Claude Design, it took six seconds to generate something so polished that I turned to my sister and said “we’re doomed” Something that would have taken me several hours, or days.

So yes, in terms of execution, we are no longer playing in the same league. It would be dishonest to claim otherwise, and rather pointless to argue against it. It has simply become reality.

That honesty deserves to be taken further, though. These tasks have always existed, and so have the tools to speed them up. In my field, Wix promised a website in thirty minutes. Shopify promised a shop without a developer. Divi promised design without a designer. These tools exist, they are used, they have their place, and they did not kill the professions they were supposed to make obsolete. They shifted the value. What mattered before in the assembling now matters in the judgement, the choices, the direction.

AI does the same thing, at a superior speed and scale. It is one more tool in a long line of tools that promised to change everything, but mostly changed what we need to focus our attention on.

What interests me is not whether it executes better than I do on those tasks. It does, full stop. The more interesting question is: what does that free up, and what does it not replace?

What accessibility does not say

Since execution has become commonplace, the prevailing discourse has turned it into a promise for everyone. For months now, what floods every feed are tools, pre-configured stacks, ready-made system access, crash courses to “boost your business with AI.” The underlying message is always the same: it is simple, it is accessible, anyone can do it. Personally, that promise is, honestly, more anxiety-inducing than anything else.
Not because it is entirely false. But because it omits what matters most.
The numbers tell a slightly different story. According to a recent Gallup study, 69% of daily AI users describe themselves as curious about the technology, compared to only 28% of those who never use it. Curiosity does not come from use, it precedes and conditions it. In other words, someone who already understands a little about how things work, who is used to testing, questioning, failing and starting again, will get infinitely more out of these tools than someone who does not have the time, the context, and who has been sold the idea that it is simple.

That is not a value judgement. It is a reality of technological adoption that has repeated itself with every wave for thirty years.

That is where the trap of the “accessible to all” promise lies: it shifts the blame. If the tool is simple and it does not work for you, the problem is you. It is a well-worn commercial mechanism, and it works all the better because the product is real and the promise is not entirely dishonest.

What is rarely sold in those Instagram posts, LinkedIn updates, or newsletters is the actual learning curve, the number of failed prompts before you learn to formulate a good one, the need to be able to evaluate what the tool produces rather than taking it at face value. Researchers who use AI daily have learned this the hard way: last year they believed AI outperformed humans in more than half of potential use cases. This year, after experience, that figure has fallen to less than a third. The tool has not regressed. Their expectations have been recalibrated by reality.

That is what informed use looks like. Fortunately, it cannot be bought with a subscription.

Intellectual laziness, the real issue

One of the aspects that troubles me most about the widespread use of AI is not the speed, nor the power, nor even the questions around employment. It is something quieter, more insidious: the way we take what it produces for granted, and the intellectual laziness that follows.

Before, when I came across a bug in my code, I would search online for related topics. I would land on forums that had the answer, or did not. I would dig until I could interpret the solutions other developers shared or commented on. Now, I catch myself abandoning that habit and turning almost by default to AI. Out of “convenience.” The irony is that I only return to my forums out of something closer to resignation.

That convenience is not a gain. It is an enormous loss, because it makes us lazy. AI is the hyper-available friend: you ask it a question, it answers within the second. All that remains is to copy, paste, publish.

Sometimes the content is read back, more often barely skimmed. The response looks well structured, the tone is confident, the words are in the right place. So we trust it. What nobody says clearly enough is that AI does not know. It predicts. It assembles probable sequences from what it has ingested. That is technically remarkable, useful in many contexts, but it is not the same thing as understanding, verifying, or reasoning.

The problem, to my mind, is not the tool. The problem is what we delegate to it without realising: perspective, reformulation, doubt. Reflexes that took years to develop through experience, habit, and setbacks. In the age of AI, they are either exercised or they atrophy, depending on whether we use them or not.

We already live in a context saturated with information, where the shortcut has become the norm. AI amplifies and normalises that temptation on a new scale. Why search, filter, and synthesise yourself when the synthesis arrives in two seconds? Why learn, when everything is available, accessible in a click? The question is legitimate. The answer should be too: because the quality of a synthesis, or of any learning, depends on the quality of the questions asked, on the ability to recognise what is missing, on the experience that allows you to sense when something rings false.
That perspective cannot be generated. It is built slowly, through experience and through the practice of doubt.

What I observe around me, in client deliverables, in published content, in professional exchanges, is a kind of writing that resembles AI without bearing its name. Smooth, structured, empty of any genuine point of view. Nobody admits to it, everybody recognises it. It is perhaps the most concrete sign of that drift: not that AI writes in our place, but that we are beginning to think the way it has taught us to formulate.

AI: a tool, not an instruction?

Let us return to our friend Clippy, the animated paperclip.
That irritating paperclip had one fundamental flaw: it imposed itself. It did not ask whether you needed help, it presumed you did, and appeared in your workspace uninvited. That is why we closed it. Not because it was useless in absolute terms, but because nobody likes having someone else decide what they need.

AI in 2026 is infinitely more capable than Clippy. Sorry, Clippy. It is also, in many contexts, infinitely more present without our consent. In the tools we use, in the interfaces we no longer truly choose, in the suggestions we accept by default without having asked for them. The technical progress is real. So is the imposition.

My relationship with AI is straightforward to summarise: it is a working tool, not a way of life. I use it when it is useful to me, in specific contexts aligned with what I want to develop, with a critical eye on what it produces. I do not have it on my phone to answer everyday questions. I know I would not install it of my own accord. I do not delegate to it what I am capable of thinking through myself, not out of romantic principle, but because I believe it is through thinking that we remain capable of thinking, of growing.

What troubles me deeply is not the tool. It is the speed at which its use is being normalised without being genuinely questioned. Adopting something because you have understood its value and its limits is not the same as adopting it because it is “simpler” or “faster.” The second posture is precisely what the stack vendors are counting on. It is also the one that produces the poorest and least considered uses.

I use AI knowingly, and I refuse to have it imposed on me. The distinction matters: this is not scepticism, it is not resistance to change. It is simply believing that a tool, however powerful, however useful, deserves to have its entry into one’s practice decided by oneself.

Clippy disappeared in 2007, without ceremony. Too intrusive, too certain of knowing what you needed before you knew yourself. We closed it because we saw it coming. What nobody saw coming is that we would stop closing.

---

Photo credit : Photo de Cash Macanayasur Unsplash