AI isn’t replacing you, it’s dismantling you

And that’s not necessarily bad news.

For months now, the same line has been everywhere. “AI is going to replace your job.” And for just as long, I’ve been making the same distinction. AI doesn’t replace jobs, it replaces tasks. A job is rarely a single, uniform block. It’s a mix of different tasks, shaped by judgment calls and trade offs. That complexity is exactly what makes it resistant to full automation.

I thought I had said everything there was to say on the subject. Then reality caught up with me. It’s the usual shift from theory to practice. On the ground, things turn out to be far more complex…

What brought me back to it was a conversation about audiovisual subtitling. A translator recently told me that automation tools are now widely used because they are considered “good enough.” Of course, she defended her craft. She pointed out that these tools still struggle with rhythm, with implied tone, with the subtle work of shifting, trimming, or even bending a sentence so it feels right on screen. Sometimes that even means adapting to words or usages that did not exist before. She wasn’t going to argue otherwise. But if you push the idea further, you also have to ask whether human work itself had already drifted toward a kind of “good enough.”

Either way, that phrase matters. What happens when “good enough” becomes the standard? That is where path dependency creeps in. You move in one direction, and it becomes harder and harder to turn back, even when you know it is not the best path. The shift is almost invisible. Tools, habits, and expectations all start to settle around this new baseline. Over time, even our sense of quality changes. Standards erode simply because we stop exercising them.

This shift rarely feels like a loss. Quite the opposite. It comes wrapped in a familiar promise. AI will free us from tedious work so we can focus on higher value tasks. That sounds right, but it is not that simple. Some of those time consuming tasks actually act as anchors. They give structure to the day. They steady the mind. They create pockets of autopilot that reduce cognitive load. Which leads to a more uncomfortable idea.

Just because something can be automated does not mean it should be.

Take a simple example. In 2018, Silicon Valley promoted Amazon Go as the future. No checkout, no friction, just seamless transactions. At the same time, moving in the opposite direction, the Dutch chain Jumbo introduced “chat checkouts,” slower lanes designed to encourage conversation. One model chased efficiency. The other turned a routine transaction into a social moment. One looked inevitable, the other almost irrational. By 2026, the outcome is clear. Amazon Go has shut down. Meanwhile, more than 200 chat checkouts are now operating across the Netherlands.

You see my point. I’ll say it again. Just because something can be automated does not mean it should be.

Now, with the rise of agent based systems in companies, we are crossing a new threshold. To automate a process, you first need to reproduce it. To reproduce it, you have to break it down. Step by step, you dismantle the process, the role, and eventually the job itself. You isolate each action. You spell out the rules. You put words on instincts that used to remain implicit. In other words, before you can automate a job, you have to break it down.

That process is revealing. It shows where the real value lies, but it is also uncomfortable, even a little brutal. It strips away the convenient ambiguity we tend to maintain about our own work. The question is whether we are ready for that kind of clarity. As Cardinal de Retz put it, “One only leaves ambiguity at one's own expense.”

Maybe we have been asking the wrong question all along. Instead of wondering whether AI will replace our jobs, we should look at how it forces us to take them apart, piece by piece. And in the end, what may surprise us is not what disappears, but what remains. Once everything is laid out, the jobs that have defined us for the past century may start to look very different, and perhaps less obvious than we once believed. Or not.

MD