When human work gradually ceases to be essential

From copilot to economic infrastructure

At first, AI is introduced as a productivity tool.

It writes a document, analyses data, helps with programming, prepares a presentation or proposes a decision.

Then its role can gradually change:

assist → recommend → decide by default → act autonomously

As systems become more reliable, human intervention can appear as an additional step: slower, more expensive and sometimes less efficient.

A company that automates more can produce faster and at lower cost. Its competitors are then encouraged to do the same. What was initially a technological choice can gradually become an economic necessity.

The same mechanism can then affect broader functions: recruitment, management, finance, research, logistics, customer relations, strategy or administration.

Ultimately, an increasing share of economic activity could even take place directly between systems: an agent who buys from another agent, negotiates a contract, allocates resources or organizes a production without direct human intervention.

The important change is therefore not only what AI can do, but the place that we are gradually giving it in the functioning of the economy.

An economy where work matters less

Today, for the majority of the population, work constitutes both a source of income and a form of economic power.

A company must recruit competent people, train them, pay them and often improve their working conditions to retain them.

A State also has an interest in having an educated, productive and healthy population.

The Intelligence Curse proposes to examine what would happen if a much larger share of value creation tomorrow came from capital, computation, energy, infrastructure and AI systems rather than from human labour.

In this scenario, several consequences become possible.

Less bargaining power for workers

The rarer and more necessary a human skill is, the more bargaining power the possessor has.

If this skill becomes readily available in the form of a reproducible AI system, this relationship may be reversed.

The question then is no longer just that of salary.

It is that of the economic value attached to being human in the production process.

The disappearance of entry-level career steps

The first positions likely to be automated may be precisely those which today allow one to learn a trade.

Junior analysts, assistants, beginning developers, young lawyers or consultants do not just produce: they acquire the experience which will then allow them to become experts.

If these tasks are automated, an organisation can gain efficiency in the short term while gradually eliminating its own skills transfer system.

This scenario precisely identifies the disappearance of junior roles and the progressive loss of know-how as early signs of dependence.

A growing concentration of economic power

If the best systems require a lot of capital, data centers, energy and infrastructure, those who control these resources can gain an increasing advantage.

The Intelligence Curse takes this reasoning to the hypothesis of a very rich society but much more fixed, in which owning capital would make it possible to directly purchase exceptional intellectual abilities, while talent or human labour would make it much less easy to change one's social position.

This scenario remains prospective.

But it raises an important question today:

What becomes of a society when the main economic contribution of the majority of its members becomes progressively less necessary?

“Intelligence curse”: when incentives change

The authors of The Intelligence Curse propose an analogy with the curse of resources.

In some economies very dependent on a natural resource, a large part of state revenue may come from oil or minerals rather than from the economic activity of the population. The literature on rentier states shows that this structure can profoundly alter political and institutional incentives.

The authors ask whether artificial intelligence capable of replacing a large share of labour could produce a comparable mechanism.

If companies gradually derive more value from automated systems than from their employees, and if governments derive more of their resources from highly automated companies, the economic incentive to invest in human capital could also evolve.

This is not a demonstrated consequence of current AI.

It is a systemic risk scenario.

But it broadens the question:

the risk may not just be that people lose their jobs. It is that the economy may gradually need them less.

And before unemployment?

This is where health at work becomes particularly important.

An economic upheaval of this nature would probably not first appear in national statistics.

It would appear in the organisation of work.

Well before a possible complete automation of a profession, we can observe:

• positions not replaced after a departure;
• the gradual disappearance of junior recruitment;
• smaller teams from which more production is required thanks to AI;
• certain tasks transferred to models;
• employees moving from producers to system supervisors;
• an increase in the work of controlling and correcting AI productions;
• a progressive loss of autonomy;
• objectives revised upwards after the introduction of a tool ;
• a decrease in human interactions;
• skills less and less practiced;
• increasing concern about the future of the profession;
• previously human decisions becoming algorithmic recommendations then automatic decisions.

Some of these transformations are already known in work on algorithmic management. The OECD notes in particular the risks of intensification of work, reduction of autonomy and invasion of private life; EU-OSHA also associates AI management systems with psychosocial risks, stress and fatigue when they are poorly designed or poorly deployed.

Tomorrow's systemic risk can therefore start with very ordinary organisational changes today.

Occupational health as an early warning system

Occupational health occupies a special position here.

It does not predict the future capabilities of the models.

It can, however, observe what their deployment actually does at work.

In France, this function corresponds directly to the missions of prevention and occupational health services: prevention of occupational risks, advice to companies and workers, health surveillance, health monitoring and, since July 2026, explicit support for the analysis of the consequences of significant organisational changes on health and safety.

This could make occupational health actors one of the first levels of detection of a much broader economic transformation.

Observe weak signals

Rather than just asking:

“How many jobs has AI eliminated?”

we can start to follow:

Is human work still necessary?

Do employees maintain their autonomy?

Are skills continuing to be developed?

Do young workers still have entry routes into the professions?

Are productivity gains really reducing workload or increasing expectations?

Can employees still understand and challenge decisions that affect them?

Would an organisation still be able to function if its AI system disappeared tomorrow?

This last question is particularly important.

A company can remain officially run by humans while gradually becoming incapable of operating without its automated systems.

This scenario precisely locates the tipping point when humans remain theoretically in charge but truly regaining control would require deactivating systems that the organisation no longer knows how to operate.

From individual signal to collective signal

The role of occupational health would obviously not be to monitor employees.

On the contrary, it would be to raise the consequences of technological deployment to the collective level.

An isolated consultation for work overload says little.

But if, after the deployment of a new tool, several teams simultaneously begin to report:

increase in workload + loss of autonomy + professional anxiety + disappearance of certain tasks + loss of meaning + reduction of workforce,

we no longer observe just an individual difficulty.

We potentially observe a transformation of the work system.

Occupational health data, field observations, organisational analyses, feedback from employees, the CSE, managers and prevention officers can thus build an early detection network.

Prevent before dependence sets in

Prevention should therefore evolve with the capabilities of the models.

It is not enough to evaluate a tool at the time of purchase.

We must regularly reassess what it has become capable of doing and the place it has progressively taken in the organisation.

We could follow four steps:

1. Assistance

AI helps the employee.

Prevention question: does it really improve work?

2. Delegation

Part of the work is entrusted to AI.

Prevention question: what skills, room for maneuver or human interactions are disappearing?

3. Dependency

The organisation begins to no longer be able to function effectively without the system.

Prevention question: is there still a real human capacity for recovery?

4. Substitution

Human work becomes less economically attractive than its automated alternative.

Prevention question: what are the consequences for employment, career paths, health, negotiating power and social cohesion?

The risk may be the transition itself

The most worrying scenario is not necessarily that of AI suddenly taking over the economy.

It may be much more mundane.

Every decision may seem rational:

automate a task;

do not replace a departure;

reduce a team;

increase an objective;

let the model make an additional decision;

removing a human procedure that has become rarely used.

None of these decisions alone constitutes a catastrophe.

But their accumulation can create a dependency in which humans remain officially in control while an increasing part of their ability to act has been transferred to the systems they use.

This is precisely what makes this risk difficult to detect.

The company can continue to operate.

Companies can become more productive.

Services can even improve.

And yet, in this scenario, the end result may be an economy that still functions very well but in which humans gradually cease to be the actors who actually make it work and who have the associated economic power.

Occupational health alone will not be able to prevent such a shift.

But it can help answer a much more immediate question:

Are we still using AI to improve human work, or are we starting to reorganize human work around the needs of AI?

And this distinction could become one of the most important signals to watch over the next few years.