The Apprenticeship Paradox: If AI Now Does the Junior Work, Where Will Tomorrow’s Experts Come From?

CULTURE & CODE

For generations, becoming good at something usually required spending years doing things you eventually became too senior to do.

Young lawyers reviewed contracts and case files. Analysts built spreadsheets and reconciled numbers. Recruiters screened hundreds of résumés. Accountants worked through transactions. Junior consultants researched markets and assembled presentations. First-time managers prepared reports, coordinated schedules, and solved relatively small operational problems.

We called much of this grunt work. And much of it is exactly the kind of work Artificial Intelligence is becoming remarkably good at doing.

From a productivity standpoint, this sounds like progress. Why ask talented young professionals to spend hours summarizing documents, searching databases, preparing first drafts, or crunching numbers when AI can increasingly do the same work in minutes?

But there is something hidden inside that logic.

Those junior professionals weren’t merely producing outputs. They were building own-most capabilities through the work itself — accumulating the experience, pattern recognition, and judgment that would eventually turn them into experts.

The spreadsheet wasn’t only a spreadsheet; the analyst was learning which numbers mattered. The contract review wasn’t merely document processing; the lawyer was developing pattern recognition. The résumé screening wasn’t simply administrative work; the recruiter was gradually learning the difference between an impressive CV and the right candidate.

What looked like low-level work was often doing two jobs at once: producing value for the organization while producing capability in the person.

And that creates one of the least discussed paradoxes of the AI transformation:

AI may remove much of the work people once had to do before they became experts. But that work was also part of how they became experts.

Welcome to the Apprenticeship Paradox.

The Career Ladder Was Also a Learning Machine

We usually think about the career ladder as an organizational hierarchy:

Analyst → Senior Analyst → Manager → Director → Executive

But beneath those titles sat something much more important: a learning architecture.

Each rung exposed people to progressively greater complexity, ambiguity, consequence, and responsibility. People did relatively simple work before difficult work. They observed experienced colleagues. They made smaller mistakes before being entrusted with larger decisions. Over time, thousands of experiences became patterns, and those patterns gradually became judgment.

That learning architecture is now beginning to change.

In its August 2026 article, Escaping the Pilot Trap: Building HR for the Agentic Era, McKinsey identifies exactly this tension. As agents absorb more entry-level work, it argues, the next generation’s judgment will increasingly have to be developed “on purpose rather than absorbed on the job.”

That is a much bigger problem than the disappearance of a few junior tasks.

AI can increasingly research, summarize, draft, reconcile, screen, analyze, schedule, model, and execute portions of workflows. As that capability grows, people may progressively move away from performing the underlying work toward supervising, challenging, improving, and governing the agents that perform it.

There is enormous productivity potential in that shift. But there is also a developmental risk –

We may be removing the bottom rungs of the career ladder without redesigning how people reach the top.

The Expertise Illusion

This creates another fascinating possibility: AI may allow people to produce expert-looking work before they have developed expert judgment.

A junior analyst equipped with powerful AI can produce sophisticated analysis. A young recruiter can access candidate intelligence that once required years of accumulated experience. A junior lawyer can generate an impressive legal synthesis in minutes.
The output gap between novice and expert may shrink dramatically.

But the judgment gap may not.

Access to expertise is not the same as possessing expertise.

This becomes particularly important as humans move from doing work toward overseeing it. In McKinsey’s April 2026 article, AI Is Everywhere. The Agentic Organization Isn’t—Yet, Alexis Krivkovich describes the evolution from having humans “in the loop” toward humans “above the loop.” When teams of agents perform most of a process, the human increasingly provides what he describes as the judgment layer on top.

The Rewired Enterprise

That sounds compelling. But it also creates an uncomfortable question:

How do you supervise work you never learned deeply enough to do yourself?

We could eventually find ourselves asking relatively inexperienced professionals to evaluate sophisticated AI outputs, identify subtle errors, challenge flawed assumptions, and exercise judgment over work they have never experienced firsthand.

That is the Apprenticeship Paradox at its most consequential.

The Judgment Economy Has a Talent Problem

This connects directly to something I explored in my previous Culture & Code article, The Human Bottleneck.
As AI makes information, analysis, and recommendations increasingly abundant, I argued that value begins migrating toward what remains scarce: human judgment.

But that raises the logical next question.

Where does judgment come from?

Judgment is not something we download in a leadership program. It develops through exposure: seeing situations repeatedly, making decisions, getting some of them wrong, understanding why, receiving feedback, observing consequences, and gradually recognizing patterns that aren’t always obvious in a textbook or a dataset.

McKinsey raises essentially this question: If AI eliminates much of the early-career work through which previous generations accumulated experience, how will younger employees develop the pattern recognition required to oversee increasingly capable AI systems?
This creates an intriguing contradiction.

The AI economy could simultaneously produce more intelligence while reducing some of the traditional opportunities through which humans develop judgment.

And if judgment is becoming more valuable, organizations can no longer afford to leave its development to chance.

We Shouldn’t Save the Grunt Work

The answer, however, cannot be nostalgia.

We shouldn’t preserve inefficient work simply because previous generations suffered through it. There is little value in forcing a junior employee to spend six hours producing something AI can generate in six minutes merely because “that’s how I learned.”

Technology has always changed how professions are learned. Calculators did not destroy mathematics. Spreadsheets did not eliminate financial expertise. Search engines did not eliminate research. AI should similarly allow us to remove repetitive, low-value work.
But there is a difference between eliminating a task and eliminating the learning embedded within that task.
Before automating junior work, organizations may therefore need to ask two questions instead of one.

The first is familiar:

What work should AI perform?

The second may prove more important:

What experiences must humans still have in order to develop the judgment we will eventually need them to exercise?

That changes the logic of job redesign. We are no longer optimizing solely for today’s productivity. We are simultaneously designing for tomorrow’s capability.

From Experience to Compressed Experience

Perhaps the answer is not to preserve traditional apprenticeship, but to reinvent it.

Historically, developing expertise depended partly on encountering enough situations over enough years. AI may allow organizations to create what I would call Compressed Experience — deliberately accelerating exposure to the situations through which judgment develops.

Instead of waiting years for certain problems to arise naturally, young professionals could work through simulations, historical cases, decision exercises, alternative scenarios, and AI-generated challenges. They could see not only what happened, but explore what might have happened had different decisions been made.

More importantly, instead of training employees simply to accept AI recommendations, organizations could teach them to interrogate them.

What assumptions did the agent make? What information might be missing? What alternative explanation exists? What would cause you to reject the recommendation? What are the second-order consequences? What would you decide differently — and why?

The developmental emphasis begins shifting from simply producing the work toward learning to question, interpret, and exercise judgment over the work.

And then, there arises an even more interesting irony here:

AI could create the Apprenticeship Paradox — and simultaneously become one of the tools for solving it.

The New Apprenticeship

Traditional apprenticeship roughly followed a familiar progression:

Watch → Help → Do → Master → Teach

The AI-era apprenticeship may need a different progression:

Observe → Simulate → Challenge → Decide → Reflect → Take Accountability

Senior professionals remain essential, but their developmental role changes. Instead of primarily teaching younger colleagues how to perform tasks, they increasingly need to teach them how to think about the outputs those tasks produce.

• Why does this recommendation feel wrong even though the numbers look right?
• What context is missing?
• What trade-off are we making?
• Where could this decision fail?
• When should you trust the model — and when should you challenge it?

These are difficult things to encode into a prompt because much of expertise is tacit. It lives in context, intuition, relationships, ethics, consequences, and thousands of patterns accumulated through experience.

AI may therefore reduce the need for some traditional forms of apprenticeship without reducing the need for apprenticeship itself.

The apprenticeship doesn’t disappear. It becomes more deliberate.

HR Must Redesign More Than Jobs

This is why the Apprenticeship Paradox is not simply a Learning and Development problem.

If AI changes how expertise develops, HR may need to rethink the entire architecture through which organizations create talent:

Job Architecture → Career Architecture → Learning → Performance → Talent Management → Succession → Leadership Development

Entry-level roles may need to be designed not simply around the productive value of work, but also around its developmental value.

Organizations will need to identify which experiences genuinely build judgment, expose people to them deliberately, create stronger mentoring and simulation, rotate talent through complex situations, and progressively increase decision accountability.

This becomes particularly important in the agentic organization. McKinsey’s Escaping the Pilot Trap anticipates a future in which people who once ran processes increasingly become the people who “build, tune, and govern the agents that run” those processes.

But people cannot intelligently govern what they do not sufficiently understand.

That may become one of HR’s most important responsibilities in the AI era: not merely deciding what work machines should inherit, but protecting — and redesigning — the experiences humans still need in order to grow.

Every Expert Was Once a Beginner

AI will almost certainly remove a great deal of junior work. Much of that will be good for organizations and good for young professionals.
But automation has a time-horizon problem.

A task that appears inefficient today may have been doing two jobs simultaneously: producing an output for the organization and producing capability in the employee.

AI can replace the first.

Organizations must deliberately replace the second.

Because today’s junior analyst may become tomorrow’s CFO. Today’s recruiter may become tomorrow’s CHRO. Today’s young engineer may eventually be responsible for decisions affecting millions of customers.

And today’s first-time manager may one day lead an organization whose workforce includes thousands of humans — and perhaps thousands more AI agents.

We cannot automate the experiences through which those people once learned and simply assume expertise will somehow appear ten years later.

If the Judgment Economy makes human judgment increasingly valuable, organizations will need something they previously received almost organically: a deliberate system for producing it.

The question is therefore not whether AI should do the junior work. Much of it probably will.

The real question is what replaces that work as the apprenticeship through which tomorrow’s experts learn to think, decide, and eventually lead.

Because every generation of experts was once a generation of beginners.

AI may change what beginners do. It cannot eliminate our need to turn them into experts.