The conversation had started exactly where most conversations about AI start today: productivity.
I was meeting the HR Director of a large pan-African industrial group, and we were discussing what AI could change in HR, which activities could be automated and how organisations could capture real value from the technology. These are familiar questions for me. I have spent much of my career working at the intersection of HR, technology and transformation.
Then the discussion moved somewhere else.
If AI progressively takes over the analysis, preparation, research and first-level problem solving traditionally performed by younger professionals, what happens to the way they become experienced professionals?
The question stayed with me because I had been approaching the same issue from another direction while writing my book, The Death of Skills. In the book, I question our tendency to look at talent mainly through the skills people already possess, and argue for a broader view based on potential, learning speed and trajectory.
But there is something we rarely say about trajectory.
A trajectory needs a path.
And I am beginning to wonder whether, in our legitimate effort to make work more efficient, we may be removing some of the steps that once helped people grow.
1. In 2001, I thought I was implementing an HR system
I started my career as an engineer working on HR information systems at Royal Air Maroc. We were deploying HR Access across a group of more than 6,000 employees and ten subsidiaries, with developers and HR experts working together on payroll, time management, recruitment and other HR processes.
At the time, I naturally focused on systems, interfaces, specifications and data. Yet much of what shaped me happened around the system rather than inside it.
I remember the many discussions where something that looked perfectly clear in a specification became much less obvious once payroll experts, HR teams and developers sat around the same table. Sometimes we discovered a technical problem. At other times, the technology was simply revealing an unclear process, a business rule nobody had fully challenged, or two departments that believed they agreed because they were using the same words.
I did not call this learning. I was simply trying to get the job done.
Years later, after moving into Talent and Recruitment and eventually becoming Head of HR, I understood how valuable those situations had been. They had taught me to move between technical logic and organisational reality, and to listen for the problem behind the problem.
AI would have saved me many hours in 2001, and I would gladly have used it.
But there is something I would not want it to have saved me from: the exposure.
Because I can see today that the system was not the only thing we were building.
The work was also building us.
2. Junior work has always produced two things
We usually evaluate work through its visible output. A report has to be produced, candidates need to be screened, data must be analysed, a process needs to be tested or a presentation has to be prepared.
But junior work has always produced something else at the same time.
A recruiter reviewing hundreds of applications gradually develops pattern recognition. A consultant building an analysis learns what information matters and what can be ignored. An HRIS consultant testing processes begins to understand why apparently small exceptions can become major implementation risks.
The company receives the deliverable.
The individual accumulates experience.
For decades, these two outputs were almost impossible to separate, so organisations never really needed to think about them separately. Learning was embedded inside work.
AI changes this.
We can now obtain the first output without necessarily creating the second.
That is why I increasingly think of some junior activities as having an apprenticeship layer. The task itself may not deserve to survive, but the learning experience hidden inside it may still be valuable.
This changes the automation question.
Instead of asking only, “Can AI do this task?”, I believe leaders should also ask:
“What was this task teaching the person who used to do it?”
The answer will sometimes be “nothing important.” In that case, automate it without hesitation.
But sometimes the answer will be much more uncomfortable.
3. Then a junior consultant made the problem very real
The same question came back to me during an exchange with a junior consultant.
He had worked with AI on an analysis, and the result was very good: well structured, clear, professional and much more mature than what a young consultant would typically have been able to produce so quickly only a few years ago.
My reaction was not to question the use of AI. Quite the opposite. This is exactly the kind of productivity improvement we should encourage.
But our conversation made me think about something else.
The most interesting question was no longer whether the analysis was good. It was what the consultant had learned while producing it.
Could he explain why one assumption mattered more than another? What would he change if the client’s context were different? Which recommendation would he remove first if implementation capacity became limited? How would he recognise that a perfectly logical recommendation would simply not work in that particular organisation?
The issue was not that he could not answer those questions. The exchange simply made visible a new reality.
AI can make the output mature much faster than the professional.
For most of my career, the sophistication of someone’s work gave at least some indication of the experience behind it. That relationship is becoming weaker.
A junior can now produce a senior-looking deliverable.
But senior-looking output is not senior-level judgment.
And this creates a new challenge for managers: when the visible signals of expertise can be generated by AI, how do we know when expertise is actually there?
4. Knowledge is becoming cheap. Judgment is not.
Over the years, my career moved from technology to recruitment, HR leadership, consulting, entrepreneurship and finally the leadership of complex HR transformation programmes across different countries.
Each transition changed the way I looked at the same problem.
As an engineer, I looked for logical consistency. As an HR leader, I learned that a technically correct decision could still create very human consequences. As a consultant, I learned that understanding the client’s problem usually matters more than arriving quickly with an answer. As a business leader, I learned that transformation eventually has to produce measurable value.
Today, these perspectives often come together in the same executive conversation.
That is what experience does. It connects things that, earlier in a career, we tend to see separately.
I see professional development as a progression through four levels:
Knowledge → Practice → Judgment → Accountability
Knowledge tells me what normally works. Practice allows me to apply it repeatedly. Judgment tells me when the normal answer does not fit the situation. Accountability changes the way I think because I know that someone will have to live with the consequences of my decision.
AI can compress the first level dramatically and support the second. It may even become an extraordinary tool for challenging judgment.
What it cannot do for us is magically create twenty years of exposure.
Knowing the ten major risks of a multi-country HR transformation is useful. Recognising, during a difficult steering committee, which apparently minor issue is actually threatening the programme is something different.
That difference is where experience begins.
5. The business case contains a hidden line
This is where the problem becomes more uncomfortable because I also look at AI as someone who has managed a business.
If AI enables six consultants to deliver what once required ten, there is a real economic opportunity. Productivity improves, delivery capacity increases and margins can improve. Ignoring that would not be responsible management.
But imagine that we repeat the logic year after year.
We recruit fewer junior professionals because less junior work is required. The juniors we do recruit encounter fewer first-level problems because AI resolves more of them. Senior experts become increasingly productive because AI enables them to handle more work directly.
The model may perform extremely well for several years.
Then, one day, the organisation discovers that everybody wants senior people and not enough companies have been producing them.
This is what I call capability debt.
We already understand technical debt, data debt and process debt. Capability debt is created when a decision that improves today’s productivity reduces the organisation’s ability to generate tomorrow’s expertise.
It will not appear immediately in financial reporting. It will emerge later through weak succession pipelines, dependence on a few critical experts, higher external recruitment costs and professionals whose performance drops sharply when they face a situation that does not fit the pattern suggested by their tools.
The original productivity gain was real.
The business case was simply incomplete.
6. My students changed the way I saw the solution
For a while, this reasoning can lead to a rather defensive conclusion: perhaps we need to protect junior work.
I do not believe that.
My experience as a professor has actually convinced me of the opposite. The students I teach today have access to an intellectual resource that my generation could never have imagined. They can test an idea, ask for another explanation, simulate a case, challenge an argument and explore a new field within minutes.
That can create extraordinary professionals.
In The Death of Skills, I introduced Time-to-Skill, the time required to build a capability that an organisation needs. AI has the potential to reduce Time-to-Skill dramatically.
But I now believe we need to look at another measure beside it:
Time-to-Judgment
How long does it take someone to understand not only the standard answer, but when to use it, when to modify it and when to reject it completely?
AI can shorten Time-to-Skill much faster than Time-to-Judgment.
This creates a fascinating possibility: we could produce a generation that learns faster than any generation before it, while exposing that same generation to fewer of the situations through which professional judgment was traditionally built.
The solution, therefore, cannot be to preserve old work.
It has to be to design experience deliberately.
7. From career paths to experience architecture
This is where I think organisations need to go much further than AI training or new competency frameworks.
For critical professions, I would build what I call an Experience Architecture.
It would start with a simple question: What does someone need to have experienced, not only learned, before we trust them at the next level?
I see four practical mechanisms.
1. Build an Experience Map, not only a Skills Map
For every critical role, identify the experiences that create judgment.
A future transformation leader may need to have managed a failed assumption, handled a difficult client, defended an unpopular recommendation, dealt with a local exception to a global model and taken responsibility for a decision whose outcome was uncertain.
Those experiences are different from skills.
You cannot complete an online course and tick them off.
2. Use a dual-pass model for learning-critical work
Not every task should begin with AI.
When the learning value is important, ask the professional to form an initial view before seeing the AI recommendation. Then compare both, identify the differences and discuss why they occurred.
The goal is not to prove that the human is better.
The value lies in the gap between what I expected and what I discovered.
That gap is where judgment develops.
3. Create a Decision Portfolio
Today we track courses completed, certifications and years of experience. I would increasingly track decisions.
Which decisions has this person actually made? Under what level of uncertainty? What assumptions did they make? What happened afterwards? Where did they disagree with AI or with a senior colleague, and what did they learn from the outcome?
Over time, this creates something much richer than a traditional development record.
It becomes evidence of judgment.
4. Protect developmental friction
AI should remove administrative friction aggressively.
Nobody needs to become a better leader by spending hours reformatting a spreadsheet or copying information between systems.
But some friction has developmental value.
A difficult stakeholder conversation, a disagreement between HR and IT, a client challenging your recommendation, a local subsidiary refusing a global design, or discovering that a technically perfect answer fails when real people use it can all create expertise.
The challenge will be to distinguish wasteful friction, which should disappear, from developmental friction, which organisations need to preserve or recreate.
That is a much more sophisticated AI strategy than simply automating everything technically possible.
8. I finally understand what my first job was giving me
When I think back to that conversation with the HR Director of the pan-African industrial group, I now see why the question stayed with me.
It was not really a conversation about junior jobs.
It was a conversation about how an organisation reproduces its own expertise.
The junior consultant brought the same question to life from the opposite direction. AI was already giving him capabilities that my generation took much longer to acquire, and that is a positive development. The challenge is to make sure that faster access to capability also leads to deeper judgment.
This takes me back to my own first years at Royal Air Maroc.
There are many things I did then that I would automate immediately today. I have no nostalgia for inefficient work.
But I would not automate away the conversations with payroll experts, developers and HR managers who disagreed with me. I would not remove the moments when a perfectly logical design collided with reality, or the responsibility that came when the problem finally became mine to solve.
At 23, I thought these things were simply part of the job.
Twenty-five years later, I understand that they were how the job was building me.
AI gives us an extraordinary opportunity to remove unnecessary work and dramatically accelerate learning. We should take that opportunity.
But if work no longer develops people almost automatically, then organisations will have to become much more intentional about the experiences through which expertise is created.
Perhaps this is the real talent question of the AI era.
Not:
How much junior work can we automate?
But:
Once we have automated it, how will we build the people we will need ten years from now?
We are learning how to make juniors dramatically more productive.
Our next challenge is harder, and far more important:
learning how to make them experienced.