Work and the economy

17.1Automation: a different wave

In context

The "centaur," or the human augmented by AI

Between "AI replaces" and "AI changes nothing," a third path has a name, borrowed from chess. After Kasparov's defeat by Deep Blue (1997), so-called "centaur" games emerged: a human supported by software, a pairing that beat both the best humans alone and the best machines alone. The lesson became general: for many tasks, the best performer is neither the human nor the AI, but the well-designed duo, in which the machine handles computation, search and the first draft, and the human provides judgment, context and final responsibility. This is the model many professions aim for (a physician assisted in diagnosis, a lawyer, a developer, an analyst). Its success, however, is in no way automatic: it presupposes a good division of roles, an interface that keeps the AI controllable, and vigilance against automation bias (Chapter 21), that temptation to follow the machine without questioning it further. The centaur performs well only if the human truly stays at the controls.

17.2Jobs threatened, transformed, created

Diagram17.1. Three fates, not just one. The dominant scenario is not mass disappearance, but recomposition: most jobs will see part of their tasks automated, which demands an adaptation of skills.

The most exposed: data entry, basic translation, first-level customer support, certain administrative and accounting functions, repetitive development.
The new jobs: machine learning engineer, data scientist, prompt engineer, AI ethicist, model "trainer," AI-specialized lawyer, human-machine coordinator. In France, more than 160,000 AI-related job postings were published in 2026, placing the country at the top in Europe.

17.3The effect on entry-level workers

17.4The hidden labor behind AI

17.5Productivity, growth and the great economic debate

On one side, the optimists recall that every technological revolution (the steam engine, electricity, computing) destroyed jobs but created more, by increasing productivity and creating new wealth and new needs. In their view, AI will free humans from thankless tasks, boost growth and give rise to jobs unimaginable today. To believe that the amount of work is fixed (and therefore that an automated job is a lost job) would be a classic fallacy (the belief in a "fixed amount of work," the lump of labor fallacy). This is the logic of creative destruction described by the economist Joseph Schumpeter: technical progress constantly destroys old activities, but gives rise to new ones, often more productive.

On the other side, the cautious stress that "this time may be different": the wave is faster, broader (it affects almost all sectors at once) and goes after the cognitive, without anyone knowing toward which tasks humans will redeploy. They point to the risks of worsening inequalities (between those who own AI and those subjected to it, between the skilled and the unskilled, between countries) and to the brutality of the transition, even if the long-term outcome were favorable. It is within this framework that fundamental debates over distribution resurface: a universal basic income (paying everyone a basic income, financed by the gains of automation), but also competing avenues, such as training and transition financed by the community, the sharing of working time, a rethought taxation of capital (or even of robots), or AI "dividends." None commands consensus, but all start from the same observation: if AI creates a great deal of value by displacing labor, the question of who benefits will not settle itself.

17.6The return of the body: robotics and manual work

In context

Replace or support

As with cognitive work, the right question is not "will the robot do everything?" but "which tasks, and to replace or to augment?" The most exposed sectors are those where the gestures are repetitive and the environment controlled: logistics and warehouses (the most mature ground, already populated by mobile robots), handling, factory assembly, and, in the longer term, cleaning, catering, agriculture or construction, more disorderly settings and therefore more difficult. But the same movement opens the way to augmentation: collaborative robots ("cobots") that assist an operator, exoskeletons that relieve a warehouse worker's back or help a caregiver, teleoperated arms that multiply a precision gesture (Chapter 13). In many cases, the most likely scenario in the short term is not the factory without humans, but the human whose physical strain decreases and productivity increases, on the model of the centaur (Section 17.1) applied to physical work.

Debate

This extension of automation to manual work raises its own questions. First, it could affect jobs that had until now served as a refuge: people are often advised to retrain toward "hands-on" trades reputed to be safe from AI, advice that loses its force if robots advance quickly. Second, retraining out of manual work is socially harder (training, mobility, age), and the effect is often concentrated on regions already weakened by past waves of automation. Finally, the argument of the labor shortage (in logistics, elder care, certain hard-to-fill trades) is real and tempers the picture: in these sectors, the robot sometimes fills a gap rather than eliminating a position. The conclusion echoes that of the rest of the chapter: neither the general disappearance of work nor the status quo, but a recomposition whose scale and timing will depend, for physical work, on the still-uncertain pace of progress in robotics, and whose social consequences will, here too, come down to political choices about training and distribution (Chapter 25).

Key takeaways (Chapter 17)

The current wave of automation is different: it affects cognitive and creative tasks, not just manual ones. Think in tasks, not jobs: does AI replace or augment?
In an unprecedented turn, this wave hit the "white collar" before the "blue collar" (Moravec's paradox); but embodied AI (robots, cobots, exoskeletons, Chapter 13) now extends automation, and augmentation, to manual work.
The dominant scenario is recomposition (most jobs transformed, a minority eliminated, a minority augmented); new jobs are emerging.
Early-career jobs are the most affected (around 16% decline for 22-to-25-year-olds in exposed jobs in the United States), mainly via a slowdown in hiring.
AI rests on hidden human labor (annotation, human feedback, moderation), often precarious and located in the Global South: a question of ethics and justice.
The great economic debate (destruction versus abundance) remains open; the central stakes are training and the distribution of the gains, a political choice.

From the labor market to markets in the broader sense: Chapter 18 surveys AI across finance, transport and the other major sectors.

17.1Automation: a different wave#

17.2Jobs threatened, transformed, created#

17.3The effect on entry-level workers#

17.4The hidden labor behind AI#

17.5Productivity, growth and the great economic debate#

17.6The return of the body: robotics and manual work#