Research: Why Some Junior Employees Work Well with AI—and Others Don’t

Ashish Agarwal, Anitesh Barua, Anu Puvvada, Fangchen Song, Wen Wen · article · 2026-07 source ↗ Reader 🖥️

Amplifier / Delegator / Apprentice Typology AI as Amplifier AI Labor Disruption Capacity to Verify AI Output Cognitive Surrender Learning Culture & Slack Systemic Apprenticeship & Skill Transmission

What separates early-career employees who create value beyond an AI baseline from those who merely match or fall below it is not their foundational skills (critical thinking, domain knowledge, AI literacy) but how they behave while working with AI — orchestrating, interrogating, and refining its output — so organizations should redesign entry-level training and performance assessment around that interaction behavior rather than around screening or upskilling foundational skills alone.

practical · mixed — Reports an empirical field study (theoretical component: a three-way performance typology and its correlates) but the whole piece is organized in service of practical organizational prescriptions — how to redesign entry-level training and assessment. It is HBR's standard 'research summary → so-what for leaders' structure.

Structure

Part	Role
Opening: entry-level tasks are being delegated to AI, raising two questions — what makes individuals add value beyond an AI baseline, and how can that be intentionally developed	context/motivation
KPMG/UT Austin field study design: 523 early-career KPMG professionals (≤18 months tenure) complete business tasks with an AI agent; performance benchmarked against an AI-only baseline and against foundational-skill measures	evidence (methodology)
Three performance profiles emerge — AI apprentices (24.1%, below baseline), AI delegators (25.8%, at baseline), AI amplifiers (50.1%, above baseline) — and foundational skills don't predict which group someone falls into	core finding
Behavioral portraits of each profile, with worked examples from the documents/recommendations task	evidence/elaboration
'Rethinking Early-Career Potential': what each profile implies leaders should do differently (unlock apprentices, surface/train delegators, scale amplifiers)	application
KPMG's own training redesign as a running case study (skill-check baseline, personalized learning pathways, AI champions network, KPMG Lakehouse intern training)	narrative evidence for the prescriptions
Three closing lessons for enterprises: rethink human value creation, make judgment visible early, shift assessment from answers to process	conclusion/prescription

Key terms

AI baseline
The quality of output an AI agent produces working alone on a task, established first in the study as a fixed benchmark; all human performance is measured as above, at, or below this line.
AI apprentices Amplifier / Delegator / Apprentice Typology
Early-career employees with foundational skills (critical thinking, domain knowledge, AI literacy) statistically indistinguishable from amplifiers, who nonetheless underperform the AI baseline because they critique AI output without translating that critique into improved results, or focus on irrelevant issues.
AI delegators Amplifier / Delegator / Apprentice Typology
Employees who scored lowest on foundational skills and who match the AI baseline by passively accepting AI output with minimal interrogation, adding little independent reasoning.
AI amplifiers Amplifier / Delegator / Apprentice Typology
Employees who exceed the AI baseline not through stronger underlying skill but by actively orchestrating the workflow — framing problems, anchoring in domain frameworks, defining evaluation criteria, interrogating and iteratively refining AI output, treating AI as a 'thought partner.'
Value beyond AI / value beyond the AI baseline
The marginal human contribution that exceeds what the AI could have produced unassisted — the organizing question the whole study and article are built around.

Propositions

explicitcorroboratedIn a field study of 523 early-career KPMG professionals completing business tasks with an AI agent, performance relative to an AI-only baseline split into three profiles: AI apprentices (24.1%, below baseline), AI delegators (25.8%, comparable to baseline), and AI amplifiers (50.1%, above baseline).conf: high
supports ←State Of AI-Assisted Software Development · p9 — A large-sample, cross-occupation finding (Beane) and a large-sample, single-company finding (KPMG study) converge on the same underlying phenomenon: junior/early-career engagement with AI is heterogeneous and a real risk category, not a uniform benefit.
explicitFoundational skills — critical thinking, domain knowledge, and AI literacy — did not reliably predict which performance profile an employee fell into; apprentices scored as high as amplifiers on these measures yet still underperformed.conf: high
refinesEngineering Leadership Report 2026 · p12 — Postman's survey finds 84% of leaders believe AI will broadly harm junior developers' ability to enter/grow the profession; this field study refines that uniform-pessimism framing by showing outcomes for early-career employees using AI are highly heterogeneous (half outperform an AI-only baseline) and depend on trainable behavior, not fixed talent — an actionable lever the survey-based pessimism doesn't surface.
explicitWhat differentiates the three profiles is behavior during AI interaction, not underlying skill: amplifiers orchestrate the workflow, frame problems, set evaluation criteria, and iteratively interrogate/refine AI output; delegators passively accept output with minimal intervention; apprentices critique AI responses without that critique improving the output.conf: high
extendsCognitive Surrender · p7 — Extends the posture framing ('the outcome depends entirely on the user's posture — thinking with vs. instead of thinking') into a quantified three-way empirical typology in a non-engineering, entry-level professional-services setting.
explicitcorroboratedAI does not standardize the quality of work across employees doing the same task with the same tool — it magnifies pre-existing differences in how people apply their skills.conf: high
supports ←Living and Working in the Age of Longevity · p7 — Gratton's claim that AI multiplies an experienced, good-judgment worker's output roughly six-fold matches the KPMG field study's core finding that AI does not standardize output quality across employees — it magnifies pre-existing differences in how people apply their skills.
explicitFoundational capabilities (domain knowledge, critical thinking, AI literacy) are necessary but not sufficient for creating value beyond an AI baseline.conf: high
explicitDelegators' underperformance is often invisible to organizations, because evaluating only the final deliverable can mistake passive acceptance of an already-competent AI output for productive human contribution, letting underperformance persist while appearing acceptable.conf: high
explicitcorroboratedOrganizations should shift performance assessment from evaluating only final deliverables to evaluating the process of interaction with AI — how employees frame problems, interrogate outputs, and integrate results — because deliverable-only evaluation cannot distinguish delegators from amplifiers once AI can independently produce competent output.conf: high
supportsHow to Turn Individual Talent into Organizational Excellence · p3 — Both locate talent development as a byproduct of how work is designed and embedded in real tasks, not episodic/label-driven programs — this article applies that same principle specifically to AI-mediated entry-level work.
supports ←Teaching Coding When AI Can Write the Code · p11 — Corroborates, in an educational setting, the workplace finding that once AI can independently produce competent output, evaluation must shift from grading the deliverable to observing the process of interaction with AI.
implicitestablishedThe traditional entry-level on-ramp — tasks that trained junior employees through repetition and built professional judgment — is being displaced as AI takes over analytical, information-intensive assignments, which threatens how the next generation of judgment and domain expertise gets built.conf: high
supports ←Agentic Skill Decay · p5 — Both point to the same mechanism — the entry-level on-ramp that builds judgment through repetition is being displaced by AI assistance — evidenced here by a controlled quiz-retention study rather than a field study of task performance.
supports ←Podcast: How Will We Train Developers If AI Does the Routine Work: A Conversation with Scott Hanselman · p4 — Corroborates the claim that AI is displacing the traditional repetition-based entry-level on-ramp for professional judgment.

Arguments

Organizations should redesign entry-level training and performance assessment around AI-interaction behavior (orchestration, interrogation, iterative refinement) rather than around screening or upskilling foundational skills alone. strength: medium

Explicit premises:

Foundational skills don't predict which performance profile (apprentice/delegator/amplifier) an employee falls into.
The three profiles differ by observable interaction behavior with AI, not by underlying skill level.
Deliverable-only assessment cannot distinguish a delegator (passive, at-baseline) from an amplifier (active, above-baseline).

Implicit premises:

Interaction behavior with AI is observable and coachable — legible enough to be trained and assessed, not a fixed trait.
The behaviors that make amplifiers effective can be taught to apprentices and delegators, not just selected for.
Results from a single firm (KPMG), in professional-services-style client-engagement tasks, generalize to other organizations, industries, and knowledge-work domains.

Inferences:

If interaction behavior, not credentials, drives value creation, hiring/screening criteria for entry-level roles may need to shift from testing foundational skills toward simulated AI-collaboration assessments.
The same three-way split likely exists in other AI-augmented professions (e.g., software engineering), where an adjacent literature on 'posture toward AI' (offloading vs. surrender) already describes a similar behavioral fork.

Narratives

In the documents-and-recommendations task, apprentices "spent most of the conversation requesting, rewriting, or reorganizing information, or pursuing less relevant questions" rather than using AI to test their own thinking.
→ Illustrates the apprentice paradox — capable people who fail to translate critique into improved output.
Amplifiers, in the same task, "treated AI as a thought partner, asking it to test assumptions, explain its reasoning, consider alternative explanations, and identify missing evidence before refining the final recommendation."
→ Gives concrete behavioral content to the abstract claim that orchestration/interrogation drives outperformance.
Delegators "provided the relevant materials and desired deliverable, then largely accepted the AI's output with minimal intervention."
→ Illustrates passive acceptance and grounds the claim that delegator underperformance is invisible when only the deliverable is judged.
KPMG's own internal training program — skill-check baseline, personalized learning pathways combining courseware/podcasts/on-the-job activities/an AI champions network, and its National Intern Training at KPMG Lakehouse evolving toward scenario-based exercises for Audit interns.
→ Functions as a proof-of-concept case study for the article's prescriptions, showing them as already-in-motion practice rather than purely theoretical recommendations.
Quote from Alison Levin (NBC Universal): "The best AI strategy in the world won't matter if we stop producing people who can actually lead this industry forward."
→ Rhetorical device underscoring the stakes of the disappearing on-ramp premise, borrowed from outside the KPMG study itself.

Questions addressed

✅ What enables early-career employees to add value beyond an AI baseline in AI-enabled workflows?
Interaction behavior with AI — orchestrating the workflow, framing problems, setting evaluation criteria, interrogating and iteratively refining output — not the level of foundational skill.
✅ How can those value-creating capabilities be intentionally developed?
Contextual, task-based, continuous training embedded in real workflows; making senior judgment visible and narratable; workflow-based (not deliverable-only) assessment. Presented via KPMG's in-progress program, with results described as preliminary ("early results suggest").
⭕ Does training or intervention actually move an employee from one profile to another (e.g., delegator to amplifier), and if so, how long does that take?

Gaps

incompleteThe article asserts that behavior-focused, workflow-embedded training builds 'amplifier' behaviors, citing KPMG's own program as a test case, but reports no before/after outcome data, control group, or effect size — only "early results suggest." The load-bearing premise underneath every prescription (that these behaviors are teachable and that training measurably shifts people between profiles) is asserted, not demonstrated.
Read next: Longitudinal or intervention studies testing whether AI-collaboration training changes an individual's performance-profile classification over time; a follow-up KPMG/UT Austin publication with pre/post data.
incompleteThe study never connects its behavioral typology to the existing literature on posture toward AI tools (offloading vs. surrender) — a delegator's 'minimal intervention, high acceptance' pattern and an apprentice's 'critique that doesn't improve output' pattern look like specific instances of a more general posture distinction (thinking with vs. instead of thinking) documented elsewhere, but the article treats its typology as sui generis rather than as an instance of that broader mechanism.
Read next: Corpus material on cognitive offloading vs. cognitive surrender, and on why AI's declarative tone and surface-correctness make passive acceptance the low-effort default.
incompleteThe 24.1/25.8/50.1 split and the amplifier/delegator/apprentice behaviors are derived from a single company (KPMG) on client-engagement-style professional-services tasks; the article never addresses whether these proportions or behaviors would replicate in other industries, task types (e.g., software engineering, where an adjacent 'AI as amplifier' literature already exists), or more senior career stages.
Read next: Cross-industry or cross-domain replications of the typology, especially in software engineering where similar amplifier/dysfunction-multiplier framings and junior-pipeline concerns already circulate.

Closes gaps elsewhere

Vibe coding and agentic engineering are getting closer than I'd like · g4 — Willison's essay raised the apprenticeship problem (if AI absorbs the tasks juniors learn on, how do they develop the expertise that makes AI an amplifier?) without addressing it. This field study offers a partial, adjacent answer: entry-level value creation isn't gated by foundational skill acquisition at all, but by learned interaction behavior with AI — reframing the apprenticeship problem as one of teaching workflow orchestration rather than only protecting hands-on skill-building tasks.