AI and work: what if the real threat were the loss of judgment, not of employment?
AI produces instantly, abundantly, often adequately. But producing does not mean being valuable. In a world of cognitive abundance, the key competency becomes that of the person who knows how to evaluate, direct and take ownership — the operator of abundance.
Producing is not the same as being valuable
AI generates a draft of legal submissions, a risk analysis, or a course outline within seconds. But it does not answer the essential question: is this content good, in this context, for this person, given these stakes? That judgment rests on tacit knowledge, built from experience, mistakes, and a reading of the balance of power — which Michael Polanyi summed up as follows: we can know more than we can tell. That kind of knowledge cannot be captured in a prompt.
An amplifying mirror, not a freely accessible resource
Anthropic’s economists analysed millions of interactions with Claude. The result: 52% fall into augmentative modes, where the human iterates, adjusts and redirects. Above all, the quality of the response is directly correlated with the sophistication of the question asked.
The abundance generated by AI is not a freely accessible resource. It is conditioned by the competence of the person requesting it. This is not a technical law, but a cognitive one.
The operator of abundance
Neither a model programmer nor a passive user: the operator of abundance knows how to frame a problem that a model can act on, assess the relevance of an output, inject the context that the machine cannot have, and take responsibility for a decision based in part on suggestions they did not generate themselves. This competency is profoundly intellectual: it presupposes having been exposed to the difficulty of the tasks being delegated, before delegating them.
Delegating without having learned: cognitive de-skilling
The most worrying warning sign is not unemployment. It is the slowdown in the recruitment of young graduates in the most exposed professions. Companies are no longer hiring as many juniors for tasks that AI performs faster and more cheaply — depriving a generation of the learning by doing described by economist Kenneth Arrow.
Drafting is being delegated without having learned how to draft. Outputs are being validated without having developed the critical competency needed to detect what AI does poorly. With agentic AI — which acts, executes and orchestrates without human validation at each step — this risk is amplified: oversight becomes most necessary precisely where it is hardest to exercise.
Key takeaway
What AI cannot do in our place is decide that a given output is worthwhile, in this context, for this person, given these stakes. Training the next generations in this judgment — not merely in the use of tools — is an urgent priority. This is, in particular, the task of higher education institutions.
