If AI Does the Junior Work, Where Do Tomorrow's Senior People Come From?
The Vanishing Apprenticeship
Companies are celebrating the fact that AI can now do work once given to junior employees.
Research. Drafting. Modeling. Coding. Analysis. Presentation prep.
It looks like efficiency.
But junior work was never only about getting junior work done.
Junior work was one of the primary ways senior judgment was built.
We may be automating the task while accidentally automating away the experience that created expertise.
And the entry point is already narrowing. Stanford researchers found that employment among 22-to-25-year-olds in highly AI-exposed occupations is now roughly 19% below where it would have been had it kept pace with less-exposed work. The change is showing up primarily because companies are hiring fewer young workers.
The Work Nobody Wanted to Do Was Doing Something Important
Consider the "low-value" work: first drafts, basic models, market research, code review, client prep, meeting notes, document checking, spreadsheet cleanup, repetitive analysis, junior presentations.
These tasks look inefficient.
But hidden inside them was: repetition, pattern recognition, error correction, exposure to consequences, feedback, judgment formation.
The inefficiency was partly the training.
Senior Judgment Is Built, Not Downloaded
A senior professional does not become senior because they spent ten years employed.
They become senior because they encountered: hundreds of bad assumptions, failed deals, broken models, difficult clients, wrong forecasts, revisions, and consequences.
Experience is not time served. Experience is cognition exercised under consequence.
The Corporate Contradiction
Companies want: fewer juniors, more automation, and senior-level judgment earlier.
But if you remove the experiences that create judgment, where exactly is that judgment supposed to come from?
The contradiction is already visible on Wall Street. UBS now wants junior bankers who are proficient with AI, even as AI increasingly performs the research, analysis, and presentation work through which junior bankers traditionally learned the profession. The future senior banker is expected to supervise the machine earlier. But who teaches them what good work looks like before they become the supervisor?
Companies may be demanding senior judgment from people while eliminating the path through which senior judgment was historically acquired.
The Illusion of Accelerated Expertise
A junior employee with AI may produce work that looks dramatically more sophisticated than their actual capability.
The work may be: cleaner, faster, more complete, more polished, apparently strategic.
BairesDev surveyed 1,569 developers across 77 countries. 85% of junior developers said AI improved their understanding. Yet only 16% of senior developers said juniors fully understood the AI-generated code they submitted. Twenty-four percent of juniors said writing code from scratch was the task they felt least confident performing without AI.
AI can accelerate the appearance of expertise faster than the acquisition of expertise.
The Verification Trap
People say: "The junior doesn't need to do the work anymore. They just need to supervise the AI."
Excellent.
Then ask: How do you supervise work you never learned to perform?
You cannot reliably audit: code you do not understand, financial models you never built, legal reasoning you never practiced, research methods you never learned, or strategic assumptions you never had to test.
Supervision is itself a senior skill. You cannot skip straight to it.
Apprenticeship Is Becoming Invisible
Historically, apprenticeship happened naturally. Junior does task. Senior reviews. Junior gets corrected. Junior does it again. Patterns accumulate. Judgment forms.
AI changes that loop.
Potential new loop: Junior asks AI. AI produces work. Senior sees acceptable output. Task moves forward.
The missing part? Correction of the junior's thinking.
If the AI makes the first draft, who is forming the junior's first judgment?
This Is Bigger Than Entry-Level Jobs
This is not an employment-protection article.
The deeper argument: Organizations need a mechanism for producing expertise.
If the old mechanism disappears, you need a new one. Otherwise: succession weakens, institutional knowledge thins, senior talent becomes scarce, companies become dependent on external intelligence, and fewer humans can independently audit the machines.
This is a CEO issue.
The 2036 Problem
It is 2036.
Your best senior analyst retires.
The person replacing her spent the previous decade with AI doing most of the modeling, research, synthesis, and first-pass reasoning.
They have ten years of employment history.
Do they have ten years of judgment?
Or do they have ten years of access to judgment?
The Real Question Is Not Whether AI Replaces Juniors
The wrong question: "How many junior roles can AI eliminate?"
The better question: "How will we manufacture senior capability once AI performs the work that used to create it?"
This should be the article's central corporate question.
Maybe Apprenticeship Doesn't Disappear. Maybe It Changes.
There is another possibility. AI may eliminate low-value repetition and allow juniors to tackle harder problems earlier. Stanford research also finds that AI used for augmentation rather than automation does not show the same negative employment pattern, and other research suggests AI can help lower-skilled workers perform more complex tasks.
That would be progress.
But only if organizations deliberately replace the learning that automation removes.
The mistake is not automating junior work.
The mistake is assuming apprenticeship will automatically survive the automation.
The Recursum Connection
Most AI optimizes for: Task completed.
Recursum treats task completion as only one outcome. The deeper question is whether capability accumulates around the task.
That means the system does not merely preserve the answer. It preserves: why the decision was made, what assumptions changed, what failed, what consequences followed, where contradictions appeared, and what the human learned.
The purpose of intelligent infrastructure should not be to remove the work that creates judgment.
It should accelerate the process through which judgment is acquired.
The intelligence should compound in the human, not only in the machine.
The CEO Test
Executives need a brutally simple diagnostic:
Look at every task AI has removed from your junior workforce.
Then ask: What capability used to be learned by doing that task? Where is that capability being learned now?
If there is no answer:
You did not automate a task. You deleted part of your apprenticeship system.
Ending
AI will do more junior work.
That is probably irreversible.
The answer is not to preserve inefficient tasks simply because humans used to do them.
The answer is to preserve the formation that those tasks quietly produced.
Because companies do not merely need work completed today.
They need people capable of making harder decisions tomorrow.
If AI does all the junior work, we need a new answer to a very old question:
How do you make a senior?
The companies that solve that problem will not merely have better AI.
They will still have people worth putting in charge.
TL;DR: AI doing junior work risks breaking a company's talent-production system. Junior work was a primary way senior judgment was built through repetition, error, and feedback. Stanford research shows employment for 22-25 year olds in AI-exposed roles is down 19% due to reduced hiring. Senior judgment is "cognition exercised under consequence," not downloaded.
Companies demand senior judgment while eliminating its traditional acquisition path. UBS's new AI proficiency requirement for junior bankers, even as AI handles their traditional tasks, highlights this contradiction. AI creates an "illusion of accelerated expertise," as juniors produce sophisticated work without underlying capability (BairesDev: 85% of juniors felt AI improved understanding, but only 16% of seniors thought juniors fully understood AI-generated code).
This leads to a "verification trap": you can't supervise work you never learned to perform. Apprenticeship becomes invisible if AI produces the first draft, leaving juniors' judgment unformed. This is a CEO issue, as it threatens a company's mechanism for producing expertise. The "2036 Problem" asks if future leaders will have judgment or just access to it. While AI could improve apprenticeship by compressing it, the mistake is assuming it will survive automation automatically.
Recursum optimizes for "capability accumulated," ensuring intelligence compounds in the human. CEOs must ask what capabilities are no longer being learned and where that learning is being replaced. If AI does all the junior work, companies need a new answer to how they make a senior.
By Ernesto Verdugo, AI Architect, Founder of Verdugo Labs, and Creator of Recursum. He builds systems for what happens when human and artificial intelligence stop working separately.
Links: