AI is slowing the hiring of young workers in jobs where it can do the tasks. Experienced workers show no comparable effect so far. That creates a pipeline problem for every firm built on a pyramid. Every reviewer, manager, partner, and licensed signer started as a junior. A firm that stops hiring juniors today has fewer seniors in ten years. The fix is not to protect the old junior job. It is to redesign it, so juniors learn by reviewing and explaining AI output instead of producing first drafts.
The Evidence: The Effect Falls on New Entrants
Stanford. Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen study ADP payroll data for millions of US workers. In their August 2026 revision, employment of 22- to 25-year-olds in AI-exposed jobs is 19% below where it would be if it had kept pace with less-exposed jobs. Experienced workers "show no comparable gap." The decline comes mainly from fewer hires, not more exits. It appears where AI substitutes for tasks. Where AI complements workers, employment is flat or rising.
The authors are careful. They call these "early, descriptive indicators" rather than causal estimates. The pattern weakens after controlling for education, and some trends began before generative AI.
Harvard. Hosseini and Lichtinger studied résumé and job posting data covering about 62 million workers at 285,000 firms (SSRN). At firms that adopted generative AI, junior employment fell about 7.7% within six quarters compared with non-adopters. Senior employment held steady. Again, the cause was slower hiring. Tasks that AI can do are disappearing from junior job postings.
The wider market. The New York Fed reports that recent college graduates had an unemployment rate of about 5.6% and an underemployment rate of 42% in Q2 2026. The Yale Budget Lab finds no broad disruption of the labor market from AI. Both can be true. The overall market is stable, while the entry rung of exposed jobs thins.
The essay The Commanding Heights by Will Manidis and Joe Roberts cites the same Stanford findings. It describes junior reviewers of agent output as people in "the training environment" for their own replacement. That outcome is a design choice, not a law.
The Pipeline Was Already Thin
Professional pipelines were narrowing before AI. The AICPA's 2025 Trends Report shows first-time CPA exam candidates fell from 48,004 in 2016 to 28,082 in 2024. Accounting degrees fell about 31% from their 2015-16 peak.
There are signs of change. Accounting enrollment rose for the third straight year in 2026, according to the Journal of Accountancy. In May 2025, the AICPA and NASBA approved a new licensure path that accepts a bachelor's degree plus two years of experience.
Law shows a smaller entry class too. NALP reports that the average 2L summer class per office fell from 10 in 2022 and 2023 to 8 in 2025, and the median number of offers per office reached an all-time low of four. NALP does not attribute this to AI. Still, a smaller entry class means fewer senior associates and partners later.
AI Can Speed Learning or Stop It
The same tool can make juniors better faster or keep them from learning. The studies show what decides it.
AI speeds up novices. In a study of 5,179 customer support agents, Brynjolfsson, Li, and Raymond found AI raised productivity 14% on average and 34% for novices. Agents with two months of tenure performed as well as agents with six months who did not use AI. The gains held even when the AI was offline, so the agents had learned.
Delegation stops learning. In an Anthropic trial, 52 mostly junior engineers learned a new programming library. The group with AI scored 50% on a skills quiz. The group coding by hand scored 67%. The largest gap was in debugging. How people used AI mattered most.
Guardrails help. In a study of nearly 1,000 high school students, Bastani and coauthors found that students who practiced with an open AI chatbot scored 17% lower on exams once the AI was removed. A version that gave only hints largely removed the harm.
The pattern across these studies is consistent. AI that explains helps people learn. AI that does the work for them does not.
Seven Ways to Redesign the Junior Role
- Make juniors explain AI output before they use it. No AI draft is accepted without a short written explanation by the junior of what it does and why. The "explain" groups in the Anthropic trial kept most of their learning.
- Use hint mode for learning tasks. Configure firm AI tools to give hints and questions, not answers, during training work.
- Schedule work without AI. Hold regular exercises without AI, such as a tie-out, a memo from scratch, or a debugging session. Grade them. Skill loss shows up only when the AI is gone.
- Make review the junior's core job. Juniors review AI output against a checklist. Seniors review the juniors' reviews, and look at what they missed. This builds the judgment a future signer needs. See human-in-the-loop agent design for review patterns.
- Move juniors to judgment work sooner. If AI compresses the first months of learning, as the support study found, rotate juniors into client and judgment work earlier.
- Keep the entry rung, change its tasks. Hire for the tasks AI complements, such as review, client contact, and exception handling. Do not delete the role because AI does its old first drafts.
- Track your own pipeline every year. Count juniors hired, juniors promoted to reviewer, and reviewers per signer. A falling ratio is an early warning.
Build AI Work That Trains Your People With OpenNash
OpenNash puts AI to work in your firm in a way your people learn from. Our forward-deployed engineers:
- Design workflows with a review step that juniors run and seniors audit.
- Set up AI tools by role, with hint modes for training and full assistance for production.
- Log every review, so you can see what juniors catch and what they miss.
- Train your team to use AI to learn faster, not to skip the learning.
Book a call with an OpenNash forward-deployed engineer or email us at hello@opennash.com. In the first session, we review your top AI use cases and how your teams are structured. Then we show you which ones deliver the highest business value and the fastest ROI.