Answers are grounded in this paper and cite their sources.
Beane used the term shadow learning for the norm-bending practices through which surgical trainees built skill when approved means of learning failed them. This article extends that construct from the individual to the collective level and applies it to generative AI. It theorises shadow learning as the constitutive and subversive process through which employees, working outside formal training, build shared repertoires and informal standards around unsanctioned AI use, and in doing so change what their organisation can know. Drawing on critical organisational learning theory, affect studies and sociomaterial analysis, the article advances three arguments. Informal AI learning forms an affective infrastructure, sustained by the emotional labour of concealing practice from managers while sharing it with trusted peers. Epistemic authority within this infrastructure follows differential proficiency and position in peer networks, not use of the tools, and can diverge from formal rank. The consequences are ambivalent: the infrastructure can spread capability beyond the few who first find it, while exposing participants to self-exploitation and to appropriation of their work without acknowledgement. Uncritical delegation of tasks to AI falls outside the construct. The argument is conceptual; two constructed vignettes serve as theoretical illustrations; and the propositions await empirical study.
No takes yet. Share an insight, caveat, or question.
Dittmar et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: