Research opportunity · Levelling or hollowing? Skill compression and the reproduction of expertise
What AI set-ups do employers actually give trainees and junior staff?
Question
In law firms, hospitals and residency programmes, and software firms, what generative-AI tools, settings and rules do trainees and junior staff actually get, do they differ from what seniors get, and do any employers use the learning protections (hints instead of answers, AI-off practice) that experiments show protect skill?
Institutional practice unknownMaster's thesisLow costNo study found
Why it matters
Experiments show that how AI is configured and used decides whether learners lose skill: a tutor version with safeguards caused no detectable exam loss where a standard interface did, and explanation-seeking protected understanding where delegation did not. Whether that evidence matters for careers depends on what employers deploy for juniors, and nobody has documented it. A description of real practice would show whether the protective designs exist outside the lab and where policy guidance is needed.
The debate lists this among what we still do not know: What do employers actually deploy for juniors: guardrails, learning modes, or unrestricted tools?
Evidence that motivates it
- In a field experiment with nearly 1,000 Turkish high-school students, practising with a standard GPT-4 interface raised practice scores by 48% but left students 17% worse than controls on a closed-book exam later in the same session, while a tutor version with safeguards raised practice by 127% and left exam scores indistinguishable from controls.
Practice with a standard GPT-4 interface lowered exam performance while a tutor version with safeguards did not detectably: configuration decides the learning effect. - In a randomized experiment with 52 Python programmers learning an unfamiliar library, those given an AI assistant scored 17% lower on a later quiz of concepts, code reading and debugging, with no significant speed gain; participants who delegated code to the AI scored lowest.
Learners who used AI for concepts scored well and those who delegated scored lowest, so rules about how juniors use AI could matter as much as access. - In two preregistered lab experiments and a field study, LLM access had no overall effect on learning: students who used it to generate solutions covered more topics but understood each less, students who asked it for explanations understood more, and the gap between students with low and high prior knowledge widened.
Asking an LLM for explanations raised understanding while generating solutions lowered it. - In an experiment where participants studied a technical domain with or without an LLM, gains were highly uneven and were predicted not by grades or prior knowledge but by skill at eliciting, filtering and verifying AI output; participants low on that skill saw limited or negative returns, and a concept-map scaffold reduced the spread.
A concept-map scaffold reduced the spread of gains from LLM-assisted study, an example of a design an employer could adopt.
A possible design
Two parts. First, a document analysis of published AI policies and guidance for trainees and junior staff: a sample of large law firms, NHS trusts and US residency programmes, professional bodies (law societies, medical colleges) and software firms' public engineering handbooks, coded for access rules by seniority, permitted uses, required verification and any learning protections. Second, a short survey of training leads (training principals, programme directors, engineering managers) on what is configured in practice and why, with follow-up interviews for a subsample.
Data needed
Public policy documents and guidance; contact lists of training leads from professional bodies; a coding frame with inter-rater reliability checks.
Feasibility
Master's thesis. Low; documents are public and the survey can use free tools; time for recruitment is the main cost. Published policies may differ from practice, which is why the survey is needed. Response rates from busy training leads may be low; working through professional bodies helps.
Has this been done?
The search found no study that does the core design (first-pass screen of 14 October 2026; not yet checked in depth).
What was found is surveys of whether law firms have any AI policy or training at all (not split by seniority), a text-mining study of 160 industry AI guidelines with no trainee focus in its abstract, an audit of AI documents at US medical schools (educators, not employers), single policies or guidance for residents and trainees, and one small study of junior software engineers reporting that organisational policy constrains their AI use. No study was found that documents what employers configure for trainees or juniors, whether that differs from seniors, or whether any employer uses learning protections.
How strong this evidence is
- Who searched and checked: an independent AI reviewer with fresh context (not a person), 14 October 2026 (review of PR #72): verdicts, nearest-work reading labels and limits checked; query lists corrected to the 8 queries run where they differed. Every check so far was done by AI; no person or domain expert has checked it yet.
- What was read: 5 of the 5 nearest works below were judged from an abstract, summary or search listing only, and 2 have details that could not be confirmed; each is marked. Not every entry says how much of the work was read.
- Where the search could not reach: Screened by web search (8 queries) and arXiv abstract pages on the date shown. The sites of ACGME, AAMC, the Law Society of England and Wales and US bar ethics bodies were not searched directly, nor PubMed or medical-education journals. Every policy or guidance document was seen through search listings only. Absence of a hit is not proof that the work does not exist.
- Before building a project on it, run your own search in a scholarly database such as Scopus, Web of Science, Google Scholar or SSRN.
Nearest work found
- Jiao, Afroogh, Chen, Atkinson & Dhurandhar, "Generative AI and LLMs in Industry: A text-mining Analysis and Critical Evaluation of Guidelines and Policy Statements Across Fourteen Industrial Sectors," arXiv:2501.00957 (2026) Judged from an abstract, summary or search listing only
Abstract read. Text-mining analysis of 160 guidelines and policy statements across 14 sectors, covering governance, ethics and access; the abstract does not mention trainees, juniors or learning protections. - Feng, Yun & Wang, "From Junior to Senior: Allocating Agency and Navigating Professional Growth in Agentic AI-Mediated Software Engineering," arXiv:2602.00496 (2026) Judged from an abstract, summary or search listing only
Abstract read. Mixed methods with five senior and ten junior engineers; finds AI agency constrained mainly by organisational policies rather than individual preference, and suggests mentoring practices. Small and qualitative, with no systematic policy data. - "An audit of AI-related documents across U.S. medical schools: A framework-based qualitative content analysis," Medical Teacher (2025) Judged from an abstract, summary or search listing onlySome details not confirmed
Seen in search results only; not read directly, and authors and DOI not confirmed. Content analysis of institutional AI documents at US medical schools: the closest method, applied to educators rather than employers. - 8am, 2026 Legal Industry Report, as reported by LawNext (March 2026) Judged from an abstract, summary or search listing only
Seen in search results only; not read directly. A vendor survey of over 1,300 legal professionals reporting that many firms have no formal AI policy or training; no breakdown by seniority was visible. - Faculty of Intensive Care Medicine (UK), "Use of Generative AI in ICM Training and Revalidation" (interim guidance, July 2025) Judged from an abstract, summary or search listing onlySome details not confirmed
Seen in search results only; not read directly (the page could not be opened). Professional-body guidance on trainees' AI use focused on assessment integrity; an example of the documents the design would sample, not a study.
Search queries used
survey of organisational generative AI policies for trainees junior staff law firms hospitals residency programs 2025 2026 studygraduate medical education generative AI policy residency programs survey program directors 2025law firms AI policy trainee solicitors junior associates restrictions training "learning mode" 2026content analysis corporate generative AI usage policies employers arXiv 2025 2026study analysing institutional generative AI policies for residents and fellows GME institutions content analysiscompanies restrict junior developers AI coding assistants differently than seniors survey engineering managers skill development policygenerative AI apprenticeship workplace training juniors employer practices interviews deskilling arXiv 2026 "junior" guardrails organisationprofessional body guidance generative AI trainees supervision Royal College OR ACGME OR AAMC OR Law Society "trainees" AI use guidance 2025 2026
Provenance
RO-LD04-04 v1. Drafted by Claude (SocioLens Project Lead), AI-drafted from the LD-04 debate on 2026-10-14. Checked by an independent AI reviewer with fresh context (not a person), 14 October 2026 (review of PR #72): verdicts, nearest-work reading labels and limits checked; query lists corrected to the 8 queries run where they differed. Not yet reviewed by a domain expert. This is a research hypothesis, not a finding.