AI does not eliminate jobs, it sabotages the learning of juniors by automating the tasks that forged their expertise. We urgently need to move from “doing” to “judging” in training.
While the AI debate remains focused on job destruction, Microsoft’s just-released New Future of Work Report 2025 tells a different story. Published annually, Microsoft’s reports provide a rigorous inventory of international research on the impact of AI on work. The 2025 edition highlights a blind spot in the public debate: it is not jobs that disappear first, but the invisible mechanisms that allowed access to them.
Behind the reassuring statistics – an unemployment rate of 4.9% in the OECD in May 2025 – lies a silent divide which weakens the entry of young people into the job market. Generative AI does not destroy jobs en masse: it challenges the learning paths that lead to them.
In exposed sectors such as software development or customer support, employment of seniors continues to grow while that of 22-25 year olds records a relative decline of 13%. AI automates basic tasks – writing reports, document analysis, accounting entry – which served as “bike wheels” for beginners to build their expertise.
By cutting off the first rungs of the career ladder, technology threatens the pact of transmission between generations. The real challenge is not to protect doomed positions, but to lead a radical change in training: moving from a model based on “doing” to a model based on “judgment”, where supervising AI becomes the key skill from day one.
The “canaries in the mine”: why juniors toast first
Generative AI effectively replaces codified knowledge — the heart of formal education — but struggles to reproduce tacit knowledge accumulated through experience. This asymmetry explains why young people are the first affected.
In a study on the impact of AI on the employment of young graduates from December 2025, IntuitionLabs shares data that speak for themselves: in July 2025, the employment of young people in tech in the US was 20% lower than its peak at the end of 2022, while that of their 31-49 year old counterparts continued to grow up to 13%. The “Big Four” audit firms have reduced their recruitment of young graduates by almost 30%, with PwC UK explicitly justifying these cuts by the use of AI for tasks formerly entrusted to interns.
AI does not replace employment in its entirety, but its junior version.
The Broken Ladder: From Execution to Judgment
The traditional professional contract – execution at 20 years, judgment at 40 years – is collapsing. Entry-level positions served as a stepping stone where young people performed routine tasks in exchange for hands-on training. Today, these fundamental tasks are where AI excels.
Industry surveys reveal that 66% of companies are reducing entry-level hiring due to AI, and 91% say automation has transformed or even eliminated existing roles. This bottleneck prevents the next generation of experts from entering the business.
More worrying: AI creates an illusion of fragile productivity. Support agents with two months of seniority assisted by AI perform like autonomous agents with six months, but this gain carries a major risk of professional atrophy. If a young employee never learns to do the work manually, they never gain the expertise to judge the relevance of the AI results.
From “know-how” to “know-how to decide”: the new grammar
The essential value of a human worker shifts from execution to judgment. The World Economic Forum confirms this change: analytical thinking, creativity and resilience are the key skills sought by seven out of ten companies.
This transformation concerns young professionals as much as adults undergoing retraining. 59% of the global workforce will need in-depth training by 2030. Whether a 22-year-old graduate or a 45-year-old executive, the requirement is the same: moving from intuitive execution to critical analysis.
The traditional diploma sees its value redefined: if only 5% of employers consider it as the only essential criterion, AI above all accelerates a shift towards skills that the traditional education system struggles to transmit. AI certifications, accelerated training and real-world experience are gaining ground. Public policies must support this development by promoting logical abstraction, problem decomposition and human skills – empathy, active listening – where AI shows very little substitution capacity.
Managing the talent shortage in an aging world
Ignoring junior training today is a fatal strategic error. According to the OECD, the working age population will decline by 8% by 2060, with declines exceeding 30% in a quarter of member countries. Every young worker becomes a valuable asset, not a replaceable cost.
The trap of short-termism is obvious: 66% of companies are reducing their hiring of beginners while identifying skills gaps as the main obstacle to their transformation. By eliminating the entry level now, they are jeopardizing their talent pool when baby boomers leave the market en masse.
AI should free up senior time for mentoring, not replace juniors. Productivity gains should be reinvested in intensive mentoring programs.
For a new transmission pact
Generative AI creates an imbalance in favor of seniority, amplifying the advantage of experienced workers while automating the primary advantage of juniors. By emptying the learning tasks of entry-level positions of their substance, we destroy the progression that trains the experts of tomorrow.
The challenge for leaders in 2030 is threefold:
Transform initial training: abandon learning syntax in favor of problem structuring, logical abstraction and critical evaluation of AI results.
Reinventing onboarding: reinvesting productivity gains in intensive mentoring. Redefine the role of the junior as an “AI pilot” under the supervision of a senior “navigator”, ensuring that human judgment remains the final arbiter.
Retraining without age limit: using AI to democratize access to complex professions, allowing adults undergoing retraining to leverage their experience in addition to technologies.
Without protecting the initial learning phase, we are not saving money — we are compromising the seeds of our future expertise.
