AI is becoming very good at the kind of work traditionally handed to junior employees. For employers, the logic is compelling. An experienced developer assisted by AI can produce far more than the same developer could only a few years ago. A lawyer can review material faster. A security tester can automate increasing portions of investigation and validation. The productivity gain is real.
And once that happens, another question follows naturally: Why hire as many junior people? If one experienced professional working with AI can deliver what previously required a small team, maintaining the old staffing structure starts to look inefficient. Lower payroll, faster output, less supervision. More work accomplished by fewer people.
For quite some time, this could look like an unusually successful technological transition. But there is something hidden inside that equation.
What Exactly Creates a Senior?
Consider how an experienced software developer becomes experienced. It does not happen simply because ten years pass. A junior developer writes ordinary code, misunderstands requirements, then gets code rejected in review. After correction and the deploy, code breaks in an unusual way. Debugging gets confusing. Someone more experienced steps in to help diagnose the problem. A strange production failure gets identified. Junior learns why the code failed.
It is through this sort of failures that junior developers learn to recognize real life complex failure patterns, and that is what we usually call experience. The same mechanism exists across different professions.
Gaining experience through mundane, manual work is a meandering and inefficient process, but it does two things at once; It produces both today’s output and tomorrow’s experienced professionals.
The First Rungs Disappear
AI now seems to attack the Achilles heel of the junior-senior arrangement. It does not replace the senior professional. Instead, it replaces much of the work normally performed by juniors on the way to becoming one. Organizations shift toward something like Senior professional + AI instead of Senior professional + several junior professionals.
Today, that can be a very attractive trade, but now another question appears: if those junior tasks were also the environment through which people accumulated experience, what happens when we remove enough of them? The problem is no longer simply that fewer junior jobs exist. It is that fewer people may travel through the process that creates senior professionals. This changes the nature of the issue. We are no longer talking only about employment, but replenishment.
Problems That Arrive With A Delay
Replenishment failures have an inconvenient characteristic. They often look harmless at first.
Consider demographic decline. When birth rates fall, a country does not immediately run out of workers. At the beginning, the situation may even look favorable. There are fewer dependent children while a large working-age population remains productive. The problem starts to appear decades later, when that large cohort eventually ages and and starts to retire. If too few people were born behind them, the shortage becomes visible only after the decision that created it is far in the past. And by then, it cannot be fixed quickly. A country that suddenly discovers it needs more thirty-year-old workers cannot create them by encouraging more births today.
What raises the alarm bells for me is that we are not particularly good at responding to problems whose consequences are delayed. We tend to notice them only when they become acute, because no matter how well they add-up on paper, invisible risks receive little attention while things are going well. Demographic aging itself is a good example: even many developed, well-governed countries failed to respond effectively before the consequences became difficult to ignore.
Professional expertise may likely have a similar lag. Suppose companies increasingly reduce junior hiring while experienced employees remain in place. For ten or fifteen years, nothing necessarily would look wrong. In fact, the opposite may happen; AI makes the existing senior workforce extraordinarily productive and the organization gets more output from fewer people. Costs fall, margins improve, projects move faster, and the strategy appears to be working beautifully. Meanwhile, the number of people progressing behind that senior cohort quietly shrinks. The shortage does not become obvious until later. By then, the people who would have accumulated fifteen or twenty years of experience simply do not exist in sufficient numbers. And like demographics, expertise has a maturation period. You cannot manufacture twenty years of professional judgment in six months.
The Expertise Dividend
This means we could enter a period that looks remarkably prosperous, in which AI dramatically increases capacity. Organizations combine decades of accumulated human experience with machines capable of handling enormous amounts of execution, which may create an expertise dividend.
However, human expertise is a form of accumulated capital. A workforce containing experienced engineers, doctors, managers, auditors, researchers or lawyers represents decades of mistakes, mentoring, experimentation and exposure to reality. AI can increase the return on that capital tremendously, but if organizations simultaneously reduce the mechanisms through which new human capital is created, then part of the apparent productivity gain may not be sustainable.
In financial terms, the system could look as though it is generating exceptional income while partly running down its asset base. Or, more simply:
We may be borrowing against future expertise to increase today’s gains.
The short-term return is very visible, but the long-term liability is not.
Now Move Twenty Years Forward
Imagine that this continues. During the 2030s, AI handles an increasing proportion of routine professional work. Junior hiring remains lower than it once was. Existing senior professionals become more productive because AI amplifies what they can do. The system works. Then time passes.
The generation whose expertise was primarily formed before AI became deeply embedded in professional work begins retiring. Highly experienced people become scarcer, and their time becomes more expensive. Organizations respond in the most economically rational way available; by automating even more.
AI becomes responsible for increasingly complex work because there are fewer experienced humans available to perform or supervise it. Another decade passes. Now ask a different question from the one we started with. Not: How many junior jobs has AI replaced? But: How many humans remain capable of independently evaluating what AI is doing?
At this point, there are at least two very different futures.
Future A: AI Becomes the Expertise
Perhaps none of this turns out to be a serious problem. AI may not only become better at execution, but may also become better at the deeper forms of professional judgment that currently distinguish senior people from junior ones.
Having observed millions of real-world cases, maintaining them in its persistent memory, it could learn continuously from outcomes. Run simulations. Develop increasingly sophisticated causal models. It could even compare its experience across more organizations, patients, systems and failures than any individual human could encounter in a lifetime. It may turn out that what we currently call tacit knowledge is actually reproducible. Maybe AI could become better at preserving and transferring professional expertise than human apprenticeship ever was.
If that happens, the disappearance of traditional junior career paths may eventually matter much less than we assume. If AI genuinely becomes the superior repository and executor of expertise, then maintaining old training structures purely because they once produced human experts could become unnecessary.
Under this future, AI is not consuming human expertise without replacement. AI itself becomes the replacement.
That possibility should also be taken seriously. Otherwise, concern about preserving human capability can easily turn into nostalgia.
But there is another possibility.
Future B: Independent Human Expertise Still Matters
AI may still become indispensable without making independent human expertise obsolete. There may remain situations in which humans are valuable precisely because they are not merely reproducing the same reasoning process.
An experienced person may recognize that the objective itself is wrong, or notice an unprecedented situation, or question assumptions that everyone else has accepted, or simply look at an answer that appears perfectly reasonable and say: No. Something is wrong here.
If that still remains important in our future with AI, then the replenishment problem becomes serious. Society may discover that it still needs experienced humans after spending decades weakening the path through which those humans were created.
The important point is that AI could continue improving, while the human capability decline at the same time. This creates a strange possibility: The machine becomes more capable while the society using it becomes less capable of independently judging the machine’s output.
When the Better Option Becomes the Only Option
This is where the replenishment problem can turn into a dependency ratchet. At first, AI is simply the better choice; cheaper, faster, more scalable, so using it is only rational. But as people receive less independent practice and fewer new experts are produced, so avoiding AI becomes progressively harder.
Initially: Human performs task versus Human + AI performs task better
Later: Human + AI versus a shrinking number of humans capable of doing it without AI
Eventually, perhaps: AI-assisted expertise versus no practical independent alternative
At that point, dependence has changed character: AI is no longer merely the economically superior option. It has become an integral part of the capability itself. And the better AI becomes, the easier it is to justify allowing the independent human alternative to weaken further.
The cycle will very likely reinforce itself:
- AI removes junior work
- fewer people accumulate deep experience
- fewer independent experts emerge
- human expertise becomes scarcer and more expensive
- AI becomes even more economically attractive
- even fewer opportunities remain to develop independent expertise
The movement is easy in one direction. Reversing it may require a generation. This is the dependency ratchet.
We Do Not Know Which Future We Are Entering
The temptation here is to make a prediction between whether AI will deskill society, or it will become capable enough that human expertise no longer matters. However, neither conclusion is justified yet. The difficulty is that organizations must make decisions long before we can confidently know.
If Future A is correct, maintaining large traditional junior workforce simply to preserve an old expertise pipeline may eventually prove unnecessary. If Future B is correct, dismantling that pipeline too aggressively could leave us with a shortage that takes decades to repair. By the time we know which path we are on, reversing the decision will likely be difficult.
Perhaps Apprenticeship Has to Change
The answer cannot simply be to preserve inefficient work forever. If AI can perform a task faster, cheaper and better, forcing humans to perform it only because previous generations learned that way is not a sustainable strategy. Therefore the more useful question becomes:
If productive work no longer automatically creates expertise, how do we create expertise deliberately?
That may require a different approach to professional development. Junior developers might attempt problems before seeing AI-generated solutions. Medical trainees may need environments where diagnosis comes before AI guidance. Engineers could spend more time in simulation and controlled failure scenarios. Security professionals may need to investigate manually before automation reveals the result.
It makes sense to make AI an important part of this process, but its role should not always be to provide the answer as quickly as possible. Sometimes the better teacher may be an AI that asks questions, challenges assumptions, forces the learner to commit to a hypothesis and only then provides feedback.
For most of modern professional life, expertise formation happened largely as a side effect of productive work. If AI increasingly separates those two functions, we may have to treat them separately. The future model may look something like:
AI produces much of the output.
Organizations deliberately produce the expertise.
That may sound inefficient, but so does raising children when viewed only through the lens of this year’s economic output. Some forms of replenishment are costly, precisely because their returns arrive much later.
Where Will the Seniors Come From?
The current debate about AI and employment tends to focus on how many jobs will disappear. That is not the most important long-term question. A more consequential question is what happens to the mechanisms through which humans acquire deep, independent capability.
AI may ultimately become good enough that preserving large amounts of that capability turns out to be unnecessary. That is Future A.
Or independent human expertise may remain important precisely because AI becomes so deeply embedded in consequential work. That is Future B.
While we do not yet know which future time will bring. Expertise has one characteristic that makes the uncertainty difficult to ignore: It takes time. Judgment compounds slowly through exposure, mistakes, responsibility and experience. If we stop replenishing it, the consequences may remain invisible precisely while the decision appears most successful. By the time we notice what is missing, the people who would have filled the gap may never have been trained. And that leaves a question worth asking before the first rungs of the ladder disappear completely:
If tomorrow’s AI still needs expert supervision, where will tomorrow’s experts come from?