How this editorial can be challenged
Can an industry keep gaining from experienced workers while removing the work that produces them?
Automation can reduce the immediate cost of simple coding tasks. Firms then post for engineers who can own architecture, security and failures, but hire fewer beginners to perform supervised work. The private saving is visible now; the missing future cohort is spread across employers and years.
AI tools may make beginners productive faster and let small teams hire more people, while new technical roles appear as demand for software expands. A projection of growing employment is consistent with that optimistic path.
That outcome is plausible, but it requires deliberate paid training and evidence that novices are moving into independent roles. A decade-long occupation forecast alone cannot show whether the entry ladder is functioning today.
BLS projections are conditional, CNBC's posting trend is not a census of hiring, and Stanford's descriptive payroll gap does not establish that AI caused the early-career divergence. IMF Asia modeling is a scenario, not observed regional outcomes.
Sustained growth in paid junior hiring, apprenticeship completion, mentorship time and promotion into independent engineering roles across firms using AI heavily would weaken the pipeline concern, especially if matched to comparable firms.
The headline that sounds like reassurance
The U.S. Bureau of Labor Statistics projects 10% growth from 2025 to 2035 for software developers, quality-assurance analysts and testers. That is faster than average. CNBC's reporting on a recovery in developer postings may make the picture sound stronger still, although a rebound from a low base is not the same as a decade of uninterrupted hiring.
I want to ask a different question. If employers seek more experienced engineers but provide fewer entry-level chances, where will the experienced engineers come from? A job forecast totals an occupation. It does not tell a student whether a first rung exists.
The missing production process
Experience is produced, not purchased from thin air. Someone has to debug a modest task, get corrected, learn the limits of a shortcut and gradually take responsibility for work that matters. Those early assignments can look inefficient on a spreadsheet because the senior person must review them.
A firm using AI can save money by having a model handle the first draft and assigning a senior engineer only the hard decision. That might be sensible for today's project. If every firm does it without replacing the training function, the industry consumes a stock of expertise without replenishing it. This is an economic mechanism, not a measured universal outcome.
A signal, not a verdict
Stanford researchers using payroll data report a 19% relative employment shortfall for workers ages 22 to 25 in highly AI-exposed occupations as of June 2026, compared with a path based on less-exposed peers. Experienced workers show no comparable gap in that analysis. The researchers stress that these are descriptive patterns and cannot isolate AI as the cause.
That caveat matters. Interest rates, hiring cycles, remote work and changed employer preferences may also shape the entry market. But the gap is large enough to demand a direct measure: among employers using AI heavily, how many beginners are paid to work, mentored and promoted into independent roles?
Asia is a mirror, not a copy
The IMF's new Asia book describes an economy facing AI, aging, fragmentation and uneven productivity. A separate IMF model warns that structurally prepared economies may adopt AI sooner and capture gains before less-prepared ones. It also models a tilt toward capital owners relative to younger wage-dependent workers.
This is not proof that one country or class has already lost. It shows why a rising regional GDP line may coexist with weakened access for the people who need wages and training. The same distinction appears inside a software team: productivity can rise while the opportunity to become productive falls.
The strongest objection
The optimistic case is real. AI can help novices learn faster, reduce the cost of starting a company and create more software to build, secure and maintain. Employers may find new junior roles centered on verification, integration, product judgment or human coordination. BLS's positive outlook is consistent with that possibility.
But the optimistic path is not automatic. An assistant that completes a task can either teach a beginner or remove the beginner from the task. The difference is how work is organized: supervised responsibility, feedback, visible progression and a wage. A course or certificate cannot substitute for all of that.
The paradox worth measuring
A company can report higher output per engineer and still be quietly borrowing from the future. The immediate gain belongs to the team that ships faster. The cost of an undersupplied cohort of experienced engineers would arrive later and be shared across many employers. That lag makes the problem easy to miss in quarterly metrics.
The counterintuitive test of AI's success may be whether it creates more people capable of challenging its answers. If the occupation grows and the first rung grows with it, this argument weakens. If the upper rungs brighten while the entrance disappears, the boom may be eating the very human capacity that makes it useful.
Read the reporting
Opinion is ours. The factual record is linked below.
U.S. Bureau of Labor Statistics — software occupation outlook Stanford Digital Economy Lab — early-career employment gap CNBC — developer job-posting report IMF — AI and Economic Divergence in Asia