Novice risk work: How juniors coaching seniors on emerging technologies such as generative AI can lead to learning failures
Citation. Kellogg, K. C., Lifshitz, H., Randazzo, S., Mollick, E., Dell’Acqua, F., McFowland III, E., Candelon, F., & Lakhani, K. R. (2025). Novice risk work: How juniors coaching seniors on emerging technologies such as generative AI can lead to learning failures. Information and Organization, 35(1), 100559. https://doi.org/10.1016/j.infoandorg.2025.100559
Summary. This paper studies why junior professionals may fail to become reliable teachers for senior professionals learning generative AI. Based on interviews with 78 junior BCG consultants after a GPT-4 field experiment, the authors argue that juniors can recommend “novice AI risk mitigation tactics” that under-address emerging-technology risks. The paper shifts the explanation from senior status threat alone toward the risks GenAI poses to valued managerial outcomes such as accuracy, explainability, contextualization, and active user engagement.
Key claims
- Junior professionals expected to educate senior professionals about effective GenAI use, but the main obstacle they described was not senior status threat; it was senior concern about GenAI risks to accuracy, explainability, contextualization, and users’ active engagement with outputs.
- The authors identify three forms of novice AI risk mitigation tactics: recommendations grounded in limited understanding of GenAI capabilities, recommendations that overestimate changes to human routines rather than system design, and recommendations that overestimate project-level interventions rather than deployer- or ecosystem-level interventions.
- Juniors’ limited understanding was plausible because the technology was nascent, fast-changing, accessible without coding or infrastructure ownership, and the juniors had no formal training, no work-setting experience, and little nonwork experience with it at the time of the study.
- The paper argues that emerging technology risks require attention to system design and deployer/ecosystem interventions, including reliable-use conditions, audit tools, prompt libraries, model alignment checks, onboarding, cognitive forcing functions, task-allocation design, and developer-level choices about model behavior.
- For AI adoption, the learning problem is not simply whether lower-status workers can teach higher-status workers; it is whether the would-be teachers have enough technical and organizational understanding to diagnose where human routine changes are insufficient.
Evidence & limitations
- Design: qualitative interview study with 78 junior consultants at Boston Consulting Group in July-August 2023, conducted soon after participants used GPT-4 in a business problem-solving field experiment.
- Evidence strength: useful for theorizing how novice professionals reason about GenAI risk and senior-manager concerns in early adoption; the interviews were close to the task experience and anchored in a real consulting-work experiment.
- Limitations: single professional-services context, junior consultants only, early GPT-4 period, and evidence about recommended tactics rather than observed long-term coaching, adoption, or organizational learning outcomes. The Drive PDF is an HBS working-paper version under the earlier title; Crossref shows the later published article title and DOI used here as canonical provenance.