Everyone Wants Experience. Who Will Provide It?

A company can make a sound case for hiring fewer junior employees when artificial intelligence reduces the work available to them. It needs a separate case for how it will develop the people it still expects to need.

Consider a proposal to reduce next year’s analyst intake. Salary savings are relatively easy to estimate. The harder questions concern the work that remains: how much review will shift to senior staff, what the remaining analysts will learn, and who will be ready to take on greater responsibility when experienced employees leave.

The proposal might still be right. It should have to survive those questions.

Before reducing junior hiring, leadership should establish what expertise the company will still need—and how it will develop or recruit that expertise under the new staffing model.

That is a more useful starting point than deciding whether AI is good or bad for entry-level employment.

What happened in one support operation

In Generative AI at Work, Erik Brynjolfsson, Danielle Li and Lindsey Raymond examined 5,172 support agents serving an unnamed Fortune 500 software company. Most worked in the Philippines. The deployment took place mainly in late 2020 and early 2021.¹

Agents already received weekly managerial feedback. The new assistant added suggested replies and relevant documentation during customer conversations; employees decided what to send. Issues resolved per hour rose about 15% overall and roughly 30% among less-experienced and lower-skilled workers.¹

The researchers also found faster chats during later system outages among workers with more prior AI exposure, suggesting some learning. But outages were not necessarily random, and chat speed does not establish broader expertise. The study did not establish long-term promotion outcomes.¹

This was an earlier system in one setting, not a forecast for every employer. It nevertheless gives leadership a reason to test whether AI makes developing beginners more economical.

The hiring data do not settle the decision

The concern about access to work is real. The Federal Reserve Bank of New York reported unemployment of about 5.6% among recent U.S. college graduates in the second quarter of 2026. Its early-career category covers ages 22 to 27. Those figures describe employment conditions, not what caused them.²

A warning from payroll records

Stanford’s Canaries in the Coal Mine? working paper identifies a more specific warning. Its August 2026 revision, using ADP payroll records through June, found that employment among workers ages 22 to 25 in AI-exposed occupations was 19% below where it would have been had it kept pace with less-exposed peers. Reduced hiring accounted for most of the divergence.³

That is not a finding that AI eliminated 19% of young workers’ jobs. The authors call the evidence descriptive rather than causal. Some divergent trends predated generative AI, the patterns weakened after accounting for education, and the differences were more pronounced in their payroll sample than in national surveys. They found no evidence of widespread, economy-wide displacement.³

Why causation remains uncertain

A separate New York Fed analysis of job advertisements found that hiring weakness in highly AI-exposed occupations began before ChatGPT’s release, without a clear additional break afterward. It also found that junior and senior postings within highly exposed occupations moved broadly in parallel.⁴

These studies examine different populations and measures. They cannot be combined into a single estimate of jobs lost to AI.

For a company deciding its next hiring cycle, the evidence supports investigation, not a blanket rule. Cutting junior positions simply because other employers appear to be doing so would be a poor substitute for examining the company’s own work.

What, exactly, was the junior employee learning?

There is a difference between an assignment being routine and it being useless.

Consider an illustrative staffing review at a healthcare provider. A junior analyst must explain why overtime has risen at one facility. Suppose an approved AI tool organizes the records and prepares a first draft.

The valuable part of the assignment may come afterward. Does the analyst check whether vacancies explain the increase? Ask a supervisor about training schedules? Discover that an apparent staffing problem is concentrated in a few shifts? Revise the recommendation when the first explanation falls apart?

Those are opportunities to practice judgment. They need not depend on manually assembling every spreadsheet.

But a redesigned role that asks the analyst only to forward a finished report could remove much of that practice. The output might look better while the employee learns less about the operation.

PwC’s 2026 Global AI Jobs Barometer describes employers seeking traditionally senior-level skills in AI-exposed entry-level roles. It also identifies the tension between removing routine work and expecting judgment earlier in a career. Its findings come from job-advertisement analysis, not an experiment showing that AI caused employers’ requirements to change.⁵

The practical response is to be specific about what a junior assignment should teach.

An employer can automate much of the first while deliberately preserving the second. That does not happen merely by adding “critical thinking” to a job description.

Nor should organizations preserve repetitive work out of habit. If an assignment teaches little, replace it. The standard should be whether employees acquire capabilities the business needs, not whether they endure the same apprenticeship their managers did.

Count the work that moves

The business case also needs to follow work beyond the first draft.

Suppose an analyst saves an hour preparing a report, but a manager spends additional time checking the result. There may still be a worthwhile saving. It cannot be calculated from the analyst’s time alone.

The same applies to corrections, exceptions and the cost of maintaining the tool. A faster draft is not automatically a cheaper completed assignment.

For a hiring decision, management should compare the cost of work that meets an agreed standard. It should also distinguish between hours freed and expenditure actually removed. Time might be used to reduce a backlog, improve service or train staff rather than reduce payroll. Those can be valuable uses, but they are different business cases.

Training deserves the same treatment.

If fewer junior employees will receive more intensive coaching, that coaching needs a budget and space in managers’ schedules. If the plan is to recruit experienced people later, include recruitment costs, time to fill vacancies and a realistic allowance for onboarding.

Buying experience is a legitimate strategy. Treating its future availability as an unquestioned assumption is not.

A smaller junior intake could also be appropriate because the organization expects to need fewer managers or different specialist skills. The relevant question is not whether the old staffing structure survives. It is whether the proposed one can support the work the business expects to do.

Test the redesign before changing the intake

A useful pilot should test a staffing proposition, not just demonstrate that employees can produce more with AI.

Set up the test

For a recurring administrative workflow, an initial 90-day pilot could provide a starting point. Longer cycles would require longer observation. The purpose would be to compare current performance with a redesigned, AI-assisted role that includes explicit opportunities to practice and receive feedback.

The business-unit leader should own the decision. Finance should validate the costs. Experienced practitioners and human resources should define what competent performance looks like. Relevant risk and technology teams should approve the tools, data and review arrangements.

Where feasible, use a phased rollout with a comparison group. Keep track of differences in task complexity and employee experience so that an easier workload is not mistaken for better performance. Record material changes to the AI system during the test.

Measure the results

The review should cover more than output volume:

These are proposed measures, not a validated Elyonik Vision scoring system. Their thresholds should reflect the work being tested.

For proficiency checks, use simulated or historical cases where practical. Ask the employee to identify what is wrong with an apparently plausible answer, explain which evidence matters and decide when to seek help. Subject-matter experts should assess the work against criteria established before the pilot begins.

The objective is not to force employees to operate without tools for the sake of it. It is to establish that they can use those tools responsibly and recognize when the normal workflow is failing.

Make the hiring decision

The hiring decision should then follow the results.

If costs fall, quality holds and employees become ready for broader responsibility, the organization has a stronger basis for redesigning the role. It may decide to hire fewer people, expand output with the same team or redirect capacity elsewhere.

If output increases but senior review consumes the savings, the staffing case needs revision. If employees submit polished work but cannot identify serious errors, the development model needs attention before their responsibility expands. Material quality or safety problems should trigger corrective action rather than be averaged away by faster turnaround.

A short pilot will not establish the effect on a five-year talent pipeline. Management should therefore schedule later reviews of retention, progression and readiness for more demanding roles. The initial hiring decision should remain open to revision as that evidence arrives.

Experience should not become an unfunded entry requirement

There is a wider consequence when employers expect beginners to arrive with more experience.

If the employer no longer provides paid opportunities to acquire it, someone else must bear the cost: an educational institution, a previous employer, the individual or their family.

That could change who can realistically qualify. An applicant who can afford an unpaid placement or months of independent portfolio work faces a different set of options from someone who needs a salary immediately.

This is a conditional concern, not a measured estimate of inequality caused by AI. It applies beyond recent graduates to career changers and people returning to employment, although the age-specific studies discussed here should not be generalized to those groups.

Employers should examine whether revised requirements identify genuine readiness or simply favor applicants who had more opportunity to acquire it elsewhere. Paid learning opportunities can be part of that review without obliging a company to preserve every existing junior position.

The next hiring proposal needs a development plan

The argument is not that companies owe the old career ladder permanent protection.

An employer may reasonably conclude that it needs fewer junior employees. AI may help the remaining employees develop faster. A different organizational structure may require less of some expertise and more of another.

Those possibilities belong in the analysis.

What should not pass without scrutiny is a proposal that counts the savings from fewer beginners while assuming the future supply of experienced people will take care of itself.

The next hiring proposal should identify the capabilities the company expects to need, how people will acquire them, what that will cost and who is responsible for checking whether the plan works.

“We will hire someone experienced” may be the answer. Leadership should still ask where that person’s experience is supposed to come from.

Cite this article

Alazzawi, A. (2026, September 29). Everyone Wants Experience. Who Will Provide It? Elyonik Vision.

BibTeX citation · RIS citation · Download PDF

Related research

Can AI Make Organizations Move Faster Than Humans Can Govern Them?