Lire en français

AI is meant to free us from work. In practice, it settles us into the role of permanent judges.

We execute less.

We arbitrate… far more.

As one output follows another, there comes the moment when the tool answers before the thought has formed. I catch myself reading every sentence the way you inspect a part fresh from the mould: a smooth object, no seam, nothing to orient by, and the sense that the flaw could sit anywhere.

Outside work, I use several AIs: same questions, answers compared, divergences hunted down.

At work I have one tool. That is the rule where I am, and it is a good rule. So I cross-check differently: I go back to the sources, I change the angle, I build my prompts on facts, I ask non-leading questions so that I do not manufacture the answer I am expecting.

And the result is the same.

Production takes me almost no time.

But verification consumes the exact attention the work itself demanded.

I have not saved time — I have moved it.

This load is not the effect of overuse. It exists with a single tool, and I have found no way to remove it, only ways to make it bearable.

Intellectual fatigue has changed in nature. It has become harder to see, because it leaves no trace in a schedule.

This is what the last control point costs the person who holds it. Judgment at high frequency, on very short cycles, stripped of the bearings that gradual production used to provide.

Producing faster is a gain, provided the outputs are designed to make human review sustainable: standardised formats, so the eye knows where to settle. The reasoning behind the choices, not only the result.

I now treat this as a specification requirement, not a comfort. A tool that produces quickly but whose output is expensive to verify does not save time, it moves the cost onto someone.

And I do not claim to have solved this for myself. I can measure what I gain in production. I have no instrument for measuring what it costs me in attention.

Does the benefit of AI outweigh the fatigue of judgment, for you?