The junior seat is disappearing.
AI can make a senior employee more productive while making it harder for you to become one.
The apprenticeship problem
Conceptual diagram, not a measured staffing ratio.
A good jobs report can still be bad news for your entrance.
You are 24. You have a degree, or you are close to finishing one. You are willing to work. You apply for the analyst job, the associate job, the junior developer job. Then you read that the economy is adding jobs and wonder why nobody seems to need you.
The August 2026 employment report, released September 4, puts that tension in view. America added 162,000 payroll jobs; unemployment held at 4.1%. Food services and drinking places added 59,000 jobs, and local government education added 42,000. Information employment fell by 23,000. Professional and business services changed little. These are seasonally adjusted figures, and the latest estimates are preliminary. Source: BLS August report.
Those are real jobs and real livelihoods. But an economy that adds jobs does not necessarily add the particular first jobs through which graduates enter professional careers. The report does not identify AI as the cause of any industry’s losses. It tells us where employment changed, not why an individual hiring manager passed on a junior candidate.
The 65% number needs its full label.
SignalFire’s June 2026 talent report estimates these declines across its group of 12 major technology companies, including Alphabet, Microsoft, Apple, Amazon and Meta. Its proprietary talent dataset is not a census of all American employers. Source and methodology: SignalFire.
The distinction matters. Saying “tech hiring is down 65%” exaggerates the overall collapse and hides the more consequential imbalance: the entrance has contracted much more sharply than hiring as a whole.
There is weakness beyond that group. The latest available BLS turnover report covers July, not August: professional and business services recorded 188,000 fewer hires than in June. That is a decline in monthly hiring, not 188,000 net job losses. Source: BLS July JOLTS.
LinkedIn’s August report, also measuring July, puts its national hiring rate 26% below February 2020. That rate counts members reporting new employers relative to membership. It is neither a total employment count nor a statistic specifically for professional and business services. Source: LinkedIn.
We should resist bundling all of this into “professional jobs are below pre-pandemic levels.” Employment, job advertisements and actual hires measure different things. For someone trying to get in, the flow of new hires can matter more than how many people already have seats.
The job you lose can be the job
that was never offered.
The spreadsheet was also a classroom.
“Spreadsheet jockey” sounds dismissive. Yet in many office careers, basic production work helped pay for an apprenticeship. Pull the numbers. Reconcile the columns. Prepare the slides. Summarize the documents. Give a senior person something to review.
The assignment produced two things: a deliverable for the company and experience for the junior. The senior caught an assumption you missed. A customer asked a question your model could not answer. You learned that a perfectly formatted spreadsheet can describe the wrong problem.
Here is the substitution mechanism to watch. If AI can produce an adequate first pass and an experienced employee can check it, a manager may decide to keep the senior and leave the junior requisition unfilled. The work continues. The apprenticeship does not.
This is an explanation of how hollowing out can happen, not a claim that every junior was only doing clerical work or that AI can reliably perform every assignment. Budgets, financing conditions, earlier overhiring and the availability of experienced applicants also affect hiring decisions.
But “AI helps employees” and “AI reduces opportunities for beginners” can both describe the same decision. The benefit lands with the employee who already knows what good work looks like. The cost lands with someone who needs a chance to learn.
The research gets uncomfortably close to 24.
In an August 12, 2026 revision of Canaries in the Coal Mine?, Stanford researchers Erik Brynjolfsson, Bharat Chandar and Ruyu Chen report a widening employment gap for workers ages 22–25 in occupations exposed to AI, using ADP payroll data through June.
Employment in that group stood about 19% below the level implied by keeping pace with less-exposed peers. That is a relative shortfall, not a 19% unemployment rate. The researchers find the adjustment operates mainly through reduced hiring, and experienced workers show no comparable gap. They do not find widespread displacement across the economy.
The authors explicitly call these descriptive indicators, not causal estimates. Some differences predate generative AI, results weaken with education controls, and the patterns are stronger in their ADP sample than in national surveys. Read the revised study summary and limitations.
That is enough to take the warning seriously without pretending the cause has been settled. A hiring freeze does not need to produce a dramatic layoff announcement to change the beginning of your adult life.
Show where your judgment enters.
If you are a young man measuring yourself against the offer your friends already got, a shrinking entrance can feel like a personal verdict. It isn’t a reliable measure of your worth. It is still a practical problem you have to navigate.
My advice is to make your contribution visible at the point where an AI answer needs to meet reality. A polished output alone leaves an employer wondering how much you understand. Show the question you chose, the evidence you checked, the error you caught and the decision you changed.
A sample assignment / open each review
“Explain why sales fell.”
1. The first-pass spreadsheet
AI groups transactions, creates a chart and drafts a summary. This is useful production. By itself, it tells a reviewer little about your ability to judge whether the analysis is sound.
2. The check that changes the answer
You discover that one week of returns was recorded twice, and a new store was included in only one comparison period. You correct both, document the changes and explain why the original conclusion was misleading. This is an illustrative example.
3. The decision someone can use
You separate what the corrected data establishes from what still needs investigation. You recommend a specific next check and explain what result would change your recommendation. A reviewer can now assess your thinking.
Use public or permitted data. Choose one manageable problem in a field you understand. Keep your workings and explain them without the chatbot open. Ask someone who does the work to challenge your assumptions. A small, defensible analysis is more revealing than a grand demonstration you cannot explain.
In interviews, ask who will review your work, how often you will receive feedback and what responsibility you can earn in six months. Those answers help you evaluate whether the role will develop you. “We use AI” tells you much less.
You may also need a job that pays the bills while you keep looking. There is no shame in that. These steps can improve the evidence you present; they cannot guarantee an offer or fix an industry that has cut its training budget.
Someone has to train the next senior.
Companies have a problem here too. If every employer wants experienced judgment and fewer employers finance its development, the apparent efficiency comes with a future bill.
A better apprenticeship can put juniors alongside AI while seniors review their reasoning, expose them to customers and assign progressively harder decisions. That requires time and accountability. Buying software does not supply either automatically.
The August jobs report leaves room for national resilience and a damaged entrance to coexist. The danger for a new graduate is that the routine assignment disappears before anyone replaces the learning it provided.
You deserve a chance to become experienced. Employers need to decide whether they are still willing to create one.