By cutting your junior roles, you are deciding on your experts for 2035

The calculation works for twelve months. It does not work for ten years.

There is a trade-off happening in many companies right now, and it is rarely written down in black and white.

The sociologist Yann Ferguson puts it this way:

Are we looking at a junior + AI combo that demonetises experience? It’s not better, but it’s so much cheaper, so we make an efficiency calculation.

Here is why this calculation is short-sighted. Knowledge transfer has never happened through mentorship programmes, despite everything we write in training plans. It happened through delegation. We would entrust a first-level task to someone less experienced. They would fail. We would correct them. They would learn. The junior did the boring stuff, and that was both the entry price and the training ground.

AI is taking over exactly those tasks.

The data that follows comes from a study published in May 2026 by Projet Sens, a collective of seventeen major French groups representing 1.5 million employees. Nothing equivalent exists in Mauritius, and I will not pretend otherwise. But the trade-off described is one I see happening here.

The diamond

The Capgemini report cited in the study describes the structural effect. The organisation is evolving towards a diamond shape. The share of managers is expected to rise from 44% to 53% of teams. That of entry-level employees is expected to drop from 44% to 32%.

It is not middle-management roles that are becoming scarce. It is the entry points.

A company that freezes its junior recruitment today is not just saving on its payroll. It is making a decision about its pool of experts in ten years' time, without anyone around the table framing it as such.

In Mauritius, the issue has an extra layer. Our young graduates are already leaving. A company that shuts its front door is not only sending a signal to a candidate, it is fuelling an exodus that we then spend years compensating for with recruitment from abroad, which is more expensive and slower.

The junior who will never be able to supervise the tool

The problem has a second, more discreet layer.

A manager describes their intern in the study:

His whole day-to-day life was exclusively handled by AI. And what struck me was the total lack of critical distance regarding the answer.

A junior who only learns with AI will never develop the expertise that would allow them to supervise that same AI. We are creating a dependency.

And since the validation of AI outputs is precisely becoming the core of the work of the future, as more than six out of ten executives and 71% of employees anticipate according to Capgemini, we are creating this dependency on the skill that will count the most.

The two fears

There is an interesting symmetry in what managers are saying.

On one hand, the fear that young people will not develop any critical distance. On the other, exactly the opposite, expressed by a customer service manager:

I am actually afraid of being outclassed by the youngster who, since the age of 8, has managed to train themselves with a much higher learning capacity than me.

Anne-Sophie Martenot, HR Director at the Banque de France, sees the common thread: with AI, companies will have to rethink the valuation of expertise, as the gaps may narrow between the very good employees who progress a little and those who will catch up with them more easily.

Both fears are saying the same thing. Expertise is no longer visible.

Discernment leaves no trace

This is the core of the problem, and it is more vicious than a budget issue.

Discernment leaves no trace when everything goes well. The output of AI, however, is spectacular.

A senior who avoids an error produces nothing visible. There is no deliverable. No slide. Nothing to show in a meeting. An AI that generates fifteen pages in thirty seconds produces something very visible.

Guess what the organisation values in its annual reviews.

Laetitia Villedieu, HR Director at ADP, draws a parallel with school:

When students have access to all knowledge, what forms the legitimacy of the person who used to be the 'knower'? We must shift what we expect from a manager. It is not necessarily to be an expert, it is to provide tools, to develop the collective.

Learning tasks

There is a concrete counter-measure, and it fits into one sentence: identify learning tasks that cannot be delegated to AI.

These are not necessarily the difficult tasks. They are those that expose one to reality. The variability of situations. The instructive error. Decision-making under uncertainty. The dissatisfied customer in front of you. These things cannot be automated and become all the more precious as everything else becomes fluid.

Paradoxically, AI can accelerate junior training. By automating tedious tasks, it can give them access to complex situations earlier. On one condition, and it is absolute: that an experienced professional is there to support them.

Elisabeth Terrail, HR Director at the EDF group, extends the question to the education system: knowing how to use AI will be a prerequisite, but knowing how to keep control of it and have a critical perspective will be the real skill, and initial training must incorporate it as such.

For Mauritius, this sentence deserves to be read twice. It concerns our universities as much as our companies.

Three questions before your next budget review

How many entry-level positions have we opened this year, compared to the previous three years? If the curve is going down without it being a conscious decision, it is a decision nonetheless.

What are the three tasks in which our juniors really learn the trade, and which ones have we automated this year?

Who among us knows how to do what AI cannot do, and is this person recognised for that in their performance review?

What used to happen by itself will now have to be decided. It is more work. It is also the only option.

Claudine

Sources

Jean-Baptiste Barféty, "Working with AI, deciding between humans", Projet Sens, 19 May 2026. Study conducted with 63 managers, 25 HR directors, and 26 experts, sponsored by Laura Chaubard and Jean-Dominique Senard. projet-sens.fr

Capgemini Research Institute, data on the evolution of organisational structures, cited in the Projet Sens study (2026).

Yann Ferguson, Scientific Director of LaborIA, Inria, associate researcher at CERTOP, interview cited in the Projet Sens study.

Quotes from Laetitia Villedieu (ADP), Anne-Sophie Martenot (Banque de France), and Elisabeth Terrail (EDF) from the hearings of the Projet Sens study.

Manager quotes come from the study's anonymised focus groups.