Learning to say stop: the illusion of thinking in the age of AI
Why, as an engineering student, I decided to deliberately put friction back into the way I work.
A few months ago, I was given a project assignment. A classic: code a Snake game. My first reflex was not to read the brief. My first reflex was to copy-paste it into an AI. The code came out in a few seconds: clean, commented, working. And as I reread it, I realized something uncomfortable: I had understood neither the assignment nor the solution. I hadn't solved anything. I had simply outsourced my own thinking.
That moment, tiny as it was, served as an alarm. Not against AI: I use it every day and I believe in it. But against my use of AI. Against that habit, now automatic, of delegating the effort before even trying to think.
That is what I want to talk about. Not to moralize, but because I believe many of us, particularly in technical studies, are slowly sliding into a dependence we no longer even see.
I. AI, a tool that became a reflex#
Let's be clear from the start: AI is not the problem. It is probably one of the most powerful tools a student or an engineer has ever had in their hands. The problem is using it without thinking.
In two or three years, something shifted. We went from learning to prompting. Before, when we got stuck, we searched: we reread the course, dug into the docs, tested, got it wrong, started over. Today, getting stuck triggers an almost Pavlovian reflex: we open a chat window.
Code generated on the fly. An automatic summary of an article we haven't read. An instant answer to an exercise we haven't looked at. Every micro-task now has its shortcut. And that shortcut gives a heady feeling of productivity.
But it is an illusion. What we have set up, without realizing it, is a short circuit of thought:
II. The blind spot: silent regression#
The paradox of our time is cruel: everything is more accessible, and yet we learn less deeply.
What atrophies doesn't show right away. These are abilities we lose in small doses, with no alarm, until the day we realize they no longer respond.
Active reading. Reading a brief, analyzing it, identifying what is being asked, forming a hypothesis. We now skip this step and go straight to the generated result. We skim the assignment, or don't read it at all.
Active listening. And this may be the most insidious point. We attend conferences, lectures, presentations… to “summarize them later with AI”. We record, we transcribe, we dump it all into a model, we wait for the key ideas. But by outsourcing our mental presence, we are no longer really there. I have myself listened to entire conferences without retaining the essentials, simply because I was counting on the tool to summarize them afterwards. Except that a summary does not replace the gradual mental construction that attentive listening produces: thinking while you listen, anticipating an idea, making connections in real time. That experience cannot be delegated.
Reasoning. We move from “I understand, therefore I create” to “I generate, therefore I execute”. The ability to follow a line of logic over several steps, to hold a thread, erodes when we never have to use it anymore.
And this is where a clear distinction must be made, because it changes everything.
Automating, or cheating yourself?#
AI is excellent at automating simple, repetitive technical tasks. Generating a PowerShell snippet to configure a workstation, quickly analyzing a CSV, structuring a config file: there, it saves me real time without costing me anything cognitively. That is pure augmentation.
The problem starts when we use it to avoid effort where the effort is precisely the point of the exercise.
Sending a project brief (a Snake, as it happens) into a chat without even reading it is not automation. It is outsourcing your thinking. The student isn't coding. They don't understand. They learn nothing. They have just produced a deliverable. And a deliverable produced without understanding is a degree being hollowed out from the inside.
III. The systemic risk: depending on an AI controlled by others#
So far, I have talked about an individual risk. But there is another one, structural, and far more rarely put into words: we are becoming dependent on tools we do not control.
The models we use every day belong to a handful of players. Their business model is young, unstable, and nothing guarantees it will stay as accessible as it is today.
Remember the original pitch: “AI for everyone”. Anyone with a basic PC gets access to unprecedented reasoning power. Wonderful. Except that free is an illusion, and things are already sliding toward premium offers. A consumer plan costs more than 50 or $80 tomorrow? When you control both the tool and the way an entire generation has come to use it, you set whatever price you want.
This is the intellectual blackout scenario: the day the taps are tightened (through price, through regulation, through a simple business decision), what are we left with? If we have spent two years delegating our thinking, we wake up with an atrophied skill and a rising bill.
Yes, AI speeds up research, engineering and innovation. But we have to ask ourselves honestly: at what cost, and what lies in store for us if we have forgotten how to think without it?
Conclusion: taking back control#
Learning to say stop does not mean rejecting AI. That would be absurd, and I would be the biggest hypocrite of all. It means putting it back in its place.
AI should come after the effort, not instead of the effort. It is the shortcut you take once you have understood, not the highway that spares you from understanding. It augments thinking that already exists; it does not replace it.
In practice, here is what I have started imposing on myself:
- A friction rule: try to solve the problem on my own for 30 minutes before opening a chat. Most of the time, I find the answer. And even when I don't, I come to the AI with a real question, not a copy-paste.
- Relearning the basics by hand: rereading a course, understanding a concept by myself, taking handwritten notes, drawing diagrams.
- Spotting cognitive laziness: that precise moment when I reach for AI not because it is useful, but because it is comfortable.
You don't build muscle by watching someone else lift weights. It is exactly the same for the brain.
The brain is a muscle. And in two years, we have almost forgotten how to make it work. Learning to say stop means rediscovering the pleasure of understanding, not just of producing.
What we master sets us free. What we submit to locks us in.