By Pascale Caron

The announcement is spectacular. In one sentence, it condenses contemporary anxiety about artificial intelligence: office jobs are doomed to disappear within eighteen months. The statement, attributed to Mustafa Suleyman, head of Microsoft AI, immediately circulated through media and digital ecosystems, fueling a familiar narrative: that of rapid and massive substitution of human labor by machines. Yet, when we trace back to the source and compare this statement with available empirical research, the picture becomes more complex. What’s at stake is not so much the disappearance of white-collar workers as the reconfiguration of value, tasks, and professional hierarchies.

A Prophecy Born from an Interview

The statement comes from an interview with the Financial Times. Mustafa Suleyman discusses the hypothesis of AI reaching a level of performance “equivalent to that of humans on most professional tasks” within a twelve to eighteen-month timeframe (Financial Times, 2026). The key phrase concerns the automation of tasks, not the elimination of jobs. The nuance is crucial: a job is not a single task, but an assembly of activities, responsibilities, social interactions, and situated decisions.

This semantic shift—automated tasks becoming eliminated jobs—constitutes the first bias in media reception. It transforms a technological projection into social destiny.

Labor Economics Doesn’t Speak of Disappearance

Major international institutions describe a very different landscape. The International Labour Organization’s report on generative AI emphasizes that the dominant effect will be the augmentation of work rather than its replacement, particularly in skilled professions (ILO, Generative AI and Jobs, 2023). The OECD uses the notion of “exposure” to AI technologies: certain jobs will be significantly transformed, but rarely automated in their entirety (OECD, AI and the Labour Market, 2024).

The IMF estimates that nearly 40% of jobs will be affected by AI, but to varying degrees, ranging from complementarity to partial substitution (IMF, 2024). As for McKinsey, its projections place around 2030—not eighteen months—the possibility of automating approximately 30% of hours worked in advanced economies (McKinsey Global Institute, The Future of Work in Europe, 2024).

Temporality is therefore the central issue: organizational adoption has historically been much slower than technical capability.

The Software Engineering Precedent

The software development sector is often presented as the laboratory for this transformation. Satya Nadella indicated that between 20% and 30% of code produced at Microsoft would now be generated by AI (TechCrunch, 2025). This data is real, but its interpretation requires caution. Producing code is not the same as designing a system, ensuring its security, guaranteeing its robustness, or assuming legal responsibility for it.

Empirical research on the use of programming assistants shows significant productivity gains for certain repetitive tasks, but also an increase in the need for code review, validation, and supervision. Automation shifts work toward activities with greater responsibility rather than eliminating it.

Microsoft CTO Kevin Scott’s projection that 95% of code could be generated by AI by 2030 represents a strategic scenario rather than a measured trajectory. It fits into a logic of investment orientation and accelerated adoption.

Layoffs: Causality or Correlation?

The coincidence between AI deployment and workforce reduction plans at certain large companies fuels the idea of direct replacement. Microsoft did indeed eliminate several thousand positions in 2025 (GeekWire, 2025). But analysis of restructuring shows that it stems from financial trade-offs, managerial reorganizations, and strategic reallocations toward activities deemed priority.

Attributing these job cuts to AI alone means ignoring the structural dynamics of technological cycles: each wave of innovation is accompanied by a redeployment of human capital.

Real Uses in Legal and Accounting Professions

Thomson Reuters’ 2026 report on AI in professional services offers a valuable indicator: in law firms, AI is primarily used for documentary research, analysis, and text synthesis. These tasks represent productivity gains but do not eliminate the need for expertise, interpretation, and accountability (Thomson Reuters, AI in Professional Services Report, 2026).

The history of office technologies shows that tools that promised to massively reduce administrative employment often produced the opposite effect: they increased organizational complexity and created new roles.

The Strategic Dimension of Tech Leaders’ Discourse

Prospective statements by Big Tech leaders are not neutral. They participate in a performative strategy: announcing a transformation helps make it credible, guides investments, and accelerates adoption. They also structure competition among players by establishing the idea of an inevitable race.

This phenomenon is well documented in the history of digital technologies: narratives of disruption often precede actual transformations.

Work Transformation: Toward Skills Polarization

  • While the rapid disappearance of white-collar workers remains a myth, the ongoing transformation is real. It is organized around three dynamics: The delegation of standardized tasks,
  • the increasing value of supervisory, arbitration, and relationship activities,
  • polarization between profiles capable of orchestrating AI and those executing residual tasks.

This reconfiguration recalls David Autor’s analyses on task automation and labor market polarization (Autor, Why Are There Still So Many Jobs?, Journal of Economic Perspectives, 2015).

The Question of Social Time

The determining factor is not technical capability, but the social time of adoption. Integrating an AI system into an organization involves legal, cultural, and managerial transformations. It requires redefining responsibilities, certifying processes, and managing risk.

Historically, these transformations span a decade or more.

Toward a New Definition of White-Collar Worker

The real issue may lie elsewhere. The white-collar worker is no longer one who executes routine cognitive tasks, but one who knows how to orchestrate hybrid systems, interpret probabilistic results, and make decisions under uncertainty.

In other words, the boundary between intellectual work and automated work shifts toward the ability to produce meaning and assume responsibility.

An Economy of Trust

AI introduces a new variable in skilled professions: trust. Who is responsible for an error? Who certifies a document? Who puts their signature on the line? These dimensions, absent from technocentric discourse, nevertheless constitute the core of value in legal, financial, or consulting professions. As long as these questions remain unresolved, total substitution remains improbable.

Conclusion: From Disappearance to Metamorphosis

The prophecy of white-collar workers’ disappearance functions as a revealer. It says less about what will happen than it reveals how society apprehends technology: between fascination and anxiety. Empirical data converge toward a scenario of profound but gradual transformation, where AI reconfigures tasks, skills, and value chains without mechanically eliminating professions. The question therefore is not: Will white-collar workers disappear?
It becomes: Which activities will retain human value in an automated cognitive environment? And, ultimately, what redefinition of skilled work are we willing to accept?