Will artificial intelligence eliminate jobs?
A study by Coface and the Observatory of Threatened and Emerging Jobs estimates that, with the use of artificial intelligence (AI), nearly 5 million jobs in France are at risk by the end of the decade1 . Employment, productivity, skills, automation: the real impact will not only be social, but also economic, in that it will force every role, every profession and every organisation to prove its true value in order to avoid being replaced by AI. Contrary to alarmist theories predicting the end of work, we argue that artificial intelligence is instead putting work to the test to reveal its true value. Indeed, this new technology will force organisations to scrutinise their value chain: areas requiring skill will be enhanced; those characterised by friction, slowness or complexity-driven rent-seeking will be exposed and replaced.
Which jobs are most at risk from AI?
Major technological waves do not always replace entire jobs. They break down activities, transform tasks, reshape skills and shift value. AI accelerates this trend because it affects language, synthesis, documentary research, writing, coding, analysis, translation, customer support and even the production of materials. It is therefore moving into the very heart of cognitive work. It does not merely affect factories or administrative offices; it is encroaching on roles that were thought to be protected by qualifications, expertise, experience or a mastery of procedures. AI is, first and foremost, a technology for restructuring tasks. It writes, summarises, classifies, translates, compares, codes, explains, simulates, rephrases and assists. It does not replace lawyers, consultants, developers, communications specialists, research officers or administrative managers. It takes over fragments of their work that involve basic, repetitive tasks.
Will AI replace humans in the workplace?
The question raised by this new reality is that of the place of human value. AI challenges economic models based on billing for time spent, which previously allowed for the inclusion of human time (technology monitoring, training, informal discussions, etc.). This point is crucial because if we think solely in terms of job roles, we fail to see the real transformation. A job may remain officially stable whilst its content changes profoundly. A role may retain its title whilst losing the tasks that required judgement. An organisation may maintain its workforce whilst shifting the intelligence of the process to models, platforms, service providers or proprietary systems. Empirical studies are already showing significant productivity gains in certain contexts. In customer support, access to an AI assistant increases agents’ average productivity and is particularly beneficial to less experienced workers2. In professional writing tasks, the use of ChatGPT reduces completion time and improves the average quality of output3. Research by Microsoft Research, based on real-world use of Copilot, indicates that information-related activities, writing, teaching, consultancy and communication are among the areas most directly affected4. However, these results do not prove that AI will replace everything. They do prove that whole sections of average output are becoming less costly.
The AI labour market will not be uniform. Recent reports from the IMF, the ILO, the WEF and PwC all agree on one point: exposure will be high, but the effects will be polarised5. Some roles will see gains in productivity, pay and autonomy. Others will be standardised, monitored or made more replaceable. AI is creating a new hierarchy. At the top are those who design, own or control the models, data, infrastructure and interfaces. Next come those who use AI to start businesses, sell, make decisions, produce more quickly and become more autonomous. Then there are those who work under AI: employees who are assessed, directed, scheduled or monitored by systems they do not control. Finally, there are those whose value rested primarily on procedures, formats or access privileges.
The issue of junior staff will be particularly challenging. If the entry-level tasks in a profession are automated, how does one still learn? If AI produces the first research, the first report, the first summary, the first line of code, the first draft, what scope for training remains? This risk of a break in the training chain for new employees directly threatens the intergenerational transfer of know-how. By outsourcing to language models the routine tasks that often served as a means of learning the trade, organisations risk preventing newcomers from gaining an understanding of their profession and from developing their operational intuition6. Without a proactive commitment to implementing new forms of guided learning, organisations could face significant recruitment challenges and may ultimately prove unable to replenish their pool of experts.
What is a complexity rent?
A complexity rent arises when an actor derives their value less from what they produce than from their ability to navigate a complexity that they sometimes help to maintain. This rent exists in large organisations, public administrations, regulated professions, consultancy, compliance, certain coordination roles and some support functions. It is fuelled by procedures, interfaces, standards, formats, specialised language and information asymmetries.
AI directly targets these areas. It reduces the cost of understanding a document. It speeds up access to a summary. It makes it easier to compare tenders, contracts, standards or public policies. It enables an individual to produce a monitoring report, a memo, an analysis or a support document that previously only larger organisations could produce at a reasonable cost. It does not eliminate high-level expertise. It undermines mid-level expertise, which thrived on access, billable hours, jargon and artificial scarcity.
This is where the debate takes an economic turn. AI does not merely destroy jobs; it destroys productive excuses. It asks every activity: are you a skill or a hindrance? Are you a rare capability or a costly interface? Do you produce a decision, a relationship, a risk-taking move, an invention, or just yet another document? This line of questioning is certainly brutal, but it is useful. Many organisations have become cumbersome because they have piled up layers of coordination to manage their own complexity. They have created roles to fix problems that they themselves had created. AI can serve to expose these hidden costs. It reveals which tasks existed to create value and which existed solely because the system was too difficult to navigate.
How can we prepare for the transformation of work brought about by AI?
For businesses, the pitfall lies in confusing automation with organisational intelligence. Replacing a task with an AI output does not necessarily create value. It may save time, but it can also lead to noise, errors, dependence on a tool, a loss of business knowledge or an invisible decline in quality.
The right strategy is to assess tasks according to their actual value. Low-value tasks must be eliminated or heavily automated. Repetitive but necessary tasks must be standardised. Tasks that foster learning must be rethought, not eliminated. Tasks requiring judgement must remain in the hands of people capable of understanding the data, the model’s limitations, the context and ultimate responsibility.
The company that succeeds with AI will not be the one that has adopted the most tools. It will be the one that has understood which tasks are sources of friction, which are skills and which are opportunities for learning. Sustainable gains will not come from blind automation. They will come from the ability to eliminate unnecessary costs whilst strengthening the skills that truly create value. Making a strategic and visionary choice, by thinking outside the box (for example, through ‘edge innovation’7), will indeed remain a wholly human skill.
AI can lower barriers to entry, but it can also create new ones. Advanced models require computing power, data, talent, cloud infrastructure, global distribution and costly legal compliance. These conditions favour the major players. A technology capable of disintermediating organisations can also reinforce a few dominant platforms.
The risk is simple: replacing the old rents of complexity with infrastructure rents. If a handful of players control the models, APIs, clouds, standards, datasets, professional interfaces and marketplaces, then AI does not truly liberate labour. It merely shifts dependency. It takes power away from traditional organisations and transfers it to more powerful platforms.
Regulation should therefore avoid two mistakes. The first would be to hold back AI in the name of a blanket defence of existing jobs. This would protect old rent-seeking arrangements and slow down productivity gains. The second would be to impose compliance requirements so onerous that only large corporations could cope with them. This would lock the market in the name of security.
The right approach is more demanding: clear accountability, auditability, data portability, interoperability, a diversity of suppliers, limits on proprietary lock-ins, and transparency regarding critical dependencies. The aim is not to freeze the way work was done before AI. It is to ensure that the removal of old frictions does not create new forms of captivity.
The note is available below
- De Calignon, G. (2026). AI: the major upheaval to come in the labour market. Les Echos. https://www.lesechos.fr/monde/europe/ia-le-grand-bouleversement-a-venir-du-marche-du-travail-2221760 ↩︎
- Brynjolfsson, E., Li, D., Raymond, L., ‘Generative AI at Work’, Quarterly Journal of Economics, 2025. https://academic.oup.com/qje/article/140/2/889/7990658 ↩︎
- Noy, S., Zhang, W., ‘Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence’, Science, 2023. https://www.science.org/doi/10.1126/science.adh2586 ↩︎
- Tomlinson, K., Jaffe, S., Wang, W., Counts, S., Suri, S., ‘Working with AI: Measuring the Applicability of Generative AI to Occupations’, Microsoft Research, 2025. https://arxiv.org/abs/2507.07935 ↩︎
- IMF, Gen-AI: Artificial Intelligence and the Future of Work, Staff Discussion Note, 2024. https://www.imf.org/en/publications/staff-discussion-notes/issues/2024/01/14/gen-ai-artificial-intelligence-and-the-future-of-work-542379
ILO, Generative AI and Jobs: A Refined Global Index of Occupational Exposure, Working Paper, 2025. https://www.ilo.org/publications/generative-ai-and-jobs-refined-global-index-occupational-exposure
World Economic Forum, The Future of Jobs Report 2025. https://www.weforum.org/publications/the-future-of-jobs-report-2025/
PwC, 2026 Global AI Jobs Barometer, 2026. https://www.pwc.com/gx/en/services/ai/ai-jobs-barometer.html ↩︎ - Morandini, S., Fraboni, F., De Angelis, M., Puzzo, G., Giusino, D., & Pietrantoni, L. (2023). The impact of artificial intelligence on workers’ skills: Upskilling and reskilling in organisations. Informing Science: The International Journal of an Emerging Transdiscipline, 26, 39–68. https://doi.org/10.28945/5078 ↩︎
- Bazalgette, D., Langlois-Berthelot, J., Gaie, C. (30 October 2024). ‘Edge innovation’: a creative approach to revealing the unexpected? Polytechnique Insights. https://www.polytechnique-insights.com/tribunes/science/edge-innovation-une-approche-creative-pour-reveler-linattendu/ ↩︎
