If AI performs the beginner work, how will tomorrow's experts learn?
The apprenticeship problem is no longer theoretical.
As AI takes on more execution-heavy junior work, organizations face a practical question: how will people acquire the judgment that used to develop through drafting, reconciling, researching, coding, revising, and observing experts at close range?
The answer is not to preserve inefficient work for its own sake. It is to preserve deliberate practice, feedback, increasing responsibility, and contact with expertise even when production no longer requires the novice to perform every step manually.
If organizations automate the work through which novices once learned, they inherit a new obligation: create another route to expertise.
Many professions teach judgment through tasks that are easy to dismiss as drudgery. Junior employees draft, reconcile, research, code, interview, prepare, and revise. The work is inefficient. It is also how people learn what experts notice.
AI can remove much of that friction. Organizations should welcome the productivity and fear the learning loss at the same time.
The answer is not to preserve pointless manual work. It is to redesign apprenticeship deliberately. Let novices try the task before seeing the AI answer. Ask them to explain why the answer is right or wrong. Rotate between assisted and unassisted work. Expose them to edge cases. Have experts explain their judgment out loud. Measure what learners can still do without AI.
A large 2026 randomized field experiment in science adds an optimistic counterpoint. Customized LLM-generated feedback delivered across tens of thousands of preprints increased the likelihood that authors revised their manuscripts by roughly 12.5 percent relative to baseline, with larger effects in feedback-scarce contexts. AI can democratize access to critique.
The design question is whether critique becomes a scaffold for expertise or a substitute for it.
The apprenticeship challenge is therefore a design problem, not a nostalgia project. The goal is not to preserve every tedious junior task. It is to preserve the experiences that teach people how to notice errors, explain choices, recover from failure, and eventually carry responsibility without the machine beside them.
Efficiency can remove the very experiences through which judgment used to be learned.
Every profession contains work that senior people are delighted to stop doing and junior people once had to do. First drafts. Research sweeps. Reconciliation. Basic coding. Routine analysis. Meeting notes. Document review. These tasks are obvious candidates for automation.
They are also frequently apprenticeship infrastructure.
A junior analyst learns not only by receiving the correct answer but by wrestling with messy inputs. A young grant professional learns funder behavior by reading hundreds of opportunities, not merely by receiving a ranked list. A beginning producer develops taste by making choices that fail. A new developer learns architecture partly by discovering why a seemingly reasonable implementation breaks.
If AI absorbs the task, organizations must preserve the learning function by design.
The solution is not nostalgia for drudgery. It is separating productive necessity from developmental necessity. Some work no longer needs to be done manually for the organization, but may still need to be practiced by the learner.
This is where AI can become a tutor rather than a substitute. Ask the learner to attempt the problem first. Have the model critique the attempt rather than replace it. Require explanation of why an answer is correct. Introduce edge cases. Remove assistance periodically. Measure unaided retention. Have experts narrate the signals they notice.
Research on tacit knowledge integration reinforces the stakes. If organizations can use AI to distribute patterns learned from experts, they may improve access to expertise, but they also need mechanisms that preserve how new experts emerge.
The employment question is therefore larger than whether junior jobs disappear. It is whether the pathway from novice to trusted professional remains intact when the economic justification for many novice tasks weakens.
Organizations that solve that problem will possess a long-term advantage. They will capture AI efficiency without consuming their own leadership pipeline.
The problem is not nostalgia for tedious work. Much junior work is tedious because it was designed around old constraints. The challenge is to identify which repetitions were merely inefficient and which repetitions were developmental.
An analyst who manually builds a model may learn where assumptions hide. A young attorney who reviews documents may learn patterns that later support judgment. A fundraiser who researches prospects may develop intuition about fit. A producer who listens through weak ideas develops taste partly through rejection. If AI removes every imperfect first attempt, it may also remove the feedback loops that create expertise.
The answer is not to force novices to work inefficiently forever. It is to redesign apprenticeship so that AI handles low-value friction while humans still perform enough diagnostic, interpretive, and decision work to develop competence. That means graduated assistance rather than instant substitution.
Apprenticeship without nostalgia
There is a reasonable objection to the apprenticeship argument: much entry-level work was never good pedagogy. It existed because organizations needed inexpensive labor. Asking a young professional to spend months formatting slides, searching databases, or reconciling routine records is not automatically character building.
That objection is correct. The goal is not to preserve drudgery. The goal is to identify the learning hidden inside the drudgery and redesign it deliberately. If reading hundreds of grant opportunities taught a fundraiser to recognize fit, the organization can create a shorter, structured exercise in which the novice ranks opportunities, explains the reasoning, compares it with AI, and receives expert feedback. The learning function survives even if the production burden disappears.
This can make apprenticeship better than it was. AI can generate practice cases, vary difficulty, provide immediate critique, expose learners to rare scenarios, and let experts spend more time coaching judgment instead of correcting formatting. The danger is not automation itself. It is automation without a theory of development.
Every automated junior workflow should therefore trigger a second design question: what capability did people used to acquire here, and where will they acquire it now? If leaders cannot answer, the productivity gain may be borrowing against the organization's future expertise.
The invisible curriculum of junior work
Entry-level work has always contained an invisible curriculum. The junior lawyer reviewing documents learns what senior lawyers notice. The development officer researching prospects learns which signals actually matter. The analyst cleaning data learns how messy institutional information really is. The assistant preparing a briefing learns the difference between what is technically true and what is strategically relevant. The task may look routine from above, but repetition builds pattern recognition.
Automation can remove the task while also removing the curriculum. This is the apprenticeship problem in its strongest form. An organization can rationally automate low-level work this quarter and irrationally discover three years later that fewer employees have developed the judgment required for senior work.
The counterargument: apprenticeship has always changed We should resist nostalgia. Many traditional apprenticeships were inefficient, exclusionary, poorly supervised, or built around work that