The professions most exposed to automation are not the least qualified.
The study released by Anthropic this month on the impact of AI on the job market contains a paradox that few executives have noticed. The professions most exposed to automation are not the least qualified. These are financial analysts, customer relations specialists, lawyers, marketing managers. Well-paid, qualified profiles at the heart of our service organizations. For every ten point increase in the rate of exposure to AI, job growth projections decline by 0.6 points according to the US Bureau of Labor Statistics. And since 2023, hiring of young people aged 22 to 25 in the most exposed professions has fallen by 14%.
The signal is weak. But it doesn’t measure what will actually happen in your teams.
I’m convinced we’re asking the wrong question. Debating how many jobs AI is eliminating is looking at the trees and missing the forest. What is happening is more structural, deeper, and paradoxically less visible: we are experiencing a cognitive consolidation. Like the agricultural consolidation of the post-war years, generative AI is redrawing the “plots” of intellectual work — not always to destroy, but by radically transforming the ecosystem in which skills are born, transmitted and organized. What we choose to preserve or sacrifice in this movement will determine the health of our organizations ten years from now.
From rural bocage to cognitive bocage
In the French countryside, post-war land consolidation had a relentless logic: group plots together, remove hedges and embankments to allow tractors to pass, and optimize yields. What looked like “disorder”—groves, sunken paths, wet meadows—was not an inefficiency to be corrected. It was an invisible infrastructure: water regulation, soil protection, biodiversity corridors, structuring of rural sociabilities. By destroying it, we gained productivity and lost resilience.
In organizations, there is a “cognitive bocage” that is just as little visible. The micro-roles that don’t appear on any organizational chart: the one who knows who to address, the one who simplifies a slide that’s too complex, the one who gets everyone back on the same page at the end of the meeting. Apparent “wastes of time”: cross-reading, going back and forth on a memo, double validation of an analysis. Informal sociability: hallway conversations, spontaneous mentoring, learning by observation. AI is extraordinarily effective at removing these hedges. It accelerates, simplifies, automates. But if we just “put AI everywhere” to go faster, we risk rediscovering what agronomists understood too late: what looks like productive inefficiency is often balancing infrastructure.
The low step of the staircase: where is judgment formed?
The number that concerns me most in the Anthropic study is not the one on overall employment. It’s the one about young workers. This 14% slowdown in hiring among 22-25 year olds in the most exposed professions is not the destruction of jobs — it is silent exclusion at entry. And this exclusion touches on something essential in intellectual organizations: the ladder of skills.
The “junior” positions – analyst assistant, monitoring manager, memo writer, reporting manager – were the bottom step of this staircase. We learned to write, to synthesize, to reformulate, to adapt to an interlocutor. Judgment through experience was developed, the ability to detect an error in a figure, to feel that an analysis holds up or that it falters. If these functions are replaced by AI tools before organizations have redesigned an alternative learning path, we are producing senior experts without having manufactured future seniors. AI does the work. Nobody learns. And in five years, who validates AI outputs with enough experience to spot what it misses?
The decision tree that managers avoid formalizing
Most COMEXes approach AI strategy from the angle of tools and productivity gains. Rarely under that of organizational arbitrations. Yet there is an implicit decision tree that every leader goes through, most often without naming it.
First node: are we looking to reduce costs in the short term, or to transform the way of working? The answer conditions everything: architecture, governance, HR policy. Who adopts AI to undo jobs answers the first question. Whoever adopts it to open up new spaces for value creation responds to the second. The two do not boil down to the same strategy.
Second node: which perimeters does AI enter first? Automating transactions seems obvious – but eliminating associated positions without rebuilding associated career paths means drying up the entry points into the profession. Touching on analysis and content modifies the very heart of intellectual training. Letting AI enter into management without safeguards means reorganizing the context of the decision itself: who decides, on what data, with what safety nets?
Third node — the most neglected: what do we do with the tasks that AI eliminates? Either they disappear altogether, or they are reconverted into supervision, teaching and quality control activities. In the first case, we destroy without rebuilding. In the second, we transform the hedges: they do not disappear, they change their function. This is the difference between an organization that will go through the transition and emerge stronger, and an organization that will discover, a few years later, that it has gained in efficiency and lost in capacity.
Towards ecological governance of AI
I am not advocating caution or slowing down. I plead for lucidity about what we choose to build. Generative AI is a new infrastructure — as structuring as the electricity grid or large-scale logistics. It redraws the territory of intellectual work. The question is not “adopt or resist.” It is: what landscape are we creating?
The leaders who get ahead will not be those who deploy the fastest. They will be those who have mapped their cognitive hedgerow, protected learning paths, and designed an organizational “green framework” – these preserved areas where judgment is formed, where corporate culture is transmitted, where the next leaders learn to decide. Agricultural consolidation has taught us that you cannot rebuild a bocage in ten years. It would be a shame to have to learn the same lesson in our organizations.
