1. Promising theoretical perspectives
1.1 Organizational knowing and epistemic transformation
A first perspective treats generative AI as an epistemic technology that changes how organizations produce, validate, and apply knowledge rather than merely as a productivity tool. Large language models generate plausible connections across domains, which can expand inquiry while simultaneously making verification and attribution more difficult. Faraj et al. [1] conceptualize LLMs as “epistemic monsters”: hybrid actors that unsettle established boundaries between human expertise, representation, and organizational knowing. This perspective is especially useful for studying professional organizations because knowledge work depends on judgments about what counts as relevant evidence, an adequate explanation, and a legitimate professional answer.
This approach can be combined with sensemaking theory. Digital technologies do not simply transmit information; they alter the materials, temporalities, embodiments, and language through which organizational actors interpret situations and coordinate action. The central theoretical issue is therefore how professionals incorporate AI-generated outputs into situated interpretation. The relevant outcome is not only task speed or accuracy, but epistemic change: whether professionals ask different questions, rely on different sources of authority, or redefine what constitutes expertise.
1.2 Human–AI collaboration as relational organizing
A second perspective moves beyond the view of AI as an autonomous organizational resource. Stelmaszak et al. [2] argue that AI should be understood as an organizing capability arising from relations among human and algorithmic actors. Its effects emerge through connectivity, codependence, and interaction rather than residing entirely in the model. This relational view is promising because the same system may produce different consequences across organizations depending on workflows, authority structures, review practices, and professional norms.
The perspective also directs attention to the social fabric of work. Baygi and Huysman describe generative AI as multipurpose, unpredictably plausible, and highly personalizable; these characteristics may reroute how employees exchange expertise, trust, and support. Under deliberate organizational cultivation, AI may strengthen coordination, whereas unmanaged adoption may produce dependence on individual “oracles,” weaken collegial exchange, or concentrate knowledge in opaque systems. The empirical contribution would be to explain how AI changes relationships through which knowledge circulates, rather than treating collaboration as a simple interaction between a user and a tool.
1.3 Complementarity, cognitive asymmetry, and task allocation
A third perspective focuses on the conditions under which human and AI capabilities are complementary. Complementarity exists when the combined team achieves outcomes that neither human nor AI can achieve independently, but such outcomes are not automatic. Hemmer et al. [3] identify information asymmetry and capability asymmetry as principal sources of complementarity. This framework allows researchers to distinguish genuine augmentation from situations in which humans merely accept AI recommendations or perform corrective work after automation fails.
Recent work on cognitive asymmetry adds a more specifically professional dimension. Yokoi et al. [4] show how human tacit and embodied knowledge can be difficult to express in the codified representations used by generative AI. They identify three coordinating practices: allocating tasks according to cognitive advantages, translating between tacit and codified forms of knowledge, and iteratively steering AI outputs. This suggests that professional value may depend less on prompt proficiency than on the ability to recognize which aspects of a problem can be formalized and which require situated judgment.
1.4 Professional learning, autonomy, and identity
A fourth perspective examines how generative AI reshapes professional development and occupational control. Existing accounts point in opposing directions. AI may support learning by providing feedback, explanations, and access to otherwise specialized knowledge; mentoring and organizational support may help employees actualize these affordances. Conversely, overreliance may reduce reflection, contextual understanding, creativity, and the development of independent expertise. Izak et al. [5] link generative AI to risks including technical dependence, declining knowledge quality, and the production of “illusory truths.”
This tension connects AI adoption to professional autonomy. Bechky and Davis argue that algorithmic management can weaken craft and community by shifting attention toward measurable outputs and scalable production, while resistance may preserve forms of judgment that are difficult to encode. The key theoretical opportunity is to study learning and autonomy jointly: a system may improve immediate performance while reducing the long-term capacity of professionals to diagnose novel problems without technological assistance.
2. Three empirical research questions
2.1 Epistemic change in professional judgment
RQ1. How does repeated use of generative AI change the criteria professionals use to evaluate, justify, and communicate knowledge claims in ambiguous work situations?
This question addresses a gap between productivity-oriented research and research on organizational knowing. It could be studied through a longitudinal mixed-method design involving professionals who perform ambiguous analytical or advisory tasks. Repeated task observations, think-aloud protocols, interviews, and analyses of work products could trace changes in source evaluation, explanation quality, confidence, and willingness to challenge AI-generated claims. The central mechanism would be a shift in epistemic authority: professionals may increasingly treat fluency and plausibility as substitutes for evidential justification, or they may develop more sophisticated verification routines.
2.2 The organizational conditions of human–AI complementarity
RQ2. Under what organizational conditions does generative AI produce complementary performance in professional teams rather than substitution, duplication, or human correction work?
This question extends complementarity theory by examining organizational arrangements rather than only individual task characteristics. A comparative study could examine teams differing in task allocation rules, review procedures, AI transparency, and professional expertise. Complementarity could be assessed by comparing human-only, AI-only, and human–AI team performance on the same set of cases, while also measuring coordination time, error detection, and the distribution of responsibility. The theoretically important mechanism is representational integration: performance should improve when teams explicitly translate tacit professional knowledge into forms that AI can use and translate AI outputs back into contextually meaningful judgments.
2.3 Generative AI and the reproduction of professional capability
RQ3. How do different modes of generative-AI use affect the development of professional expertise, autonomy, and collegial knowledge exchange over time?
This question investigates the possibility that short-term augmentation and long-term capability development may diverge. A longitudinal study could compare organizations that use AI primarily for substitution, individual assistance, collaborative problem solving, or structured mentoring. Outcomes would include independent performance on novel cases, reflective reasoning, perceived professional autonomy, reliance on colleagues, and participation in informal knowledge-sharing practices. The question is novel because it treats professional capability as a collective and temporal outcome: AI may improve current outputs while weakening the social interactions through which expertise is learned and renewed, or it may free professionals from routine work and increase opportunities for higher-order learning.
3. Theoretical integration
These perspectives can be integrated into a process model in which generative AI first alters the representations available for knowledge work, then restructures human–AI coordination, and ultimately affects professional capability and organizational relationships. The model predicts that outcomes will depend on more than model performance. They will be shaped by the ambiguity of the task, the tacit character of professional knowledge, the organization of review and accountability, and whether AI use is embedded in collective learning practices.
The literature supports these mechanisms conceptually, but it remains relatively limited in longitudinal and comparative empirical evidence. Much of the selected work consists of conceptual frameworks, reviews, or early qualitative studies; therefore, the three questions should be tested with repeated observations of actual professional work rather than with adoption-intention surveys alone. This would help distinguish immediate efficiency gains from deeper changes in epistemic standards, collaboration, and the reproduction of expertise.
