AI reshapes team dynamics in two distinct ways: when used for specialized knowledge work it complements peer consultation and is linked to fewer knowledge-silo problems, while in coordination tasks it directly improves communication quality without reducing human interaction.

Main Finding

AI adoption does not have a single uniform effect on team social health. Its association with community smells depends on the type of collaborative work: - For specialization (knowledge-seeking) activities, AI adoption is associated with increased specialization-oriented peer interaction, and that increased peer interaction is in turn associated with fewer knowledge-related community smells (less knowledge/communication fragmentation and less expertise/cultural misalignment). In other words, AI complements peer consultation indirectly via a mediated effect. - For coordination (communication/alignment) activities, AI adoption is directly associated with communication outcomes (it complements human communication) without changing interaction frequency; information-sharing outcomes are also directly related to AI use. Thus AI’s social effect is contingent on task type and how it is used.

Key Points

• Theoretical framing: Transactive Memory Systems (TMS). Focus on two TMS dimensions: Specialization and Coordination (Credibility/trust was excluded deliberately).
• Measurement: Developed and validated instruments for Human–AI and Human–Human interaction along Specialization and Coordination dimensions.
• Community smells: Constructs derived from the literature and validated via expert surveys and exploratory factor analysis (examples include knowledge/communication fragmentation, expertise/cultural misalignment, communication fragmentation, unhealthy interaction, and information-sharing problems).
• Modeling: Five Partial Least Squares Structural Equation Models (PLS-SEM) tested hypotheses linking AI adoption → human interaction (Human–Human and Human–AI) → community smells.
• Main empirical patterns:
  • Specialization path: AI → more specialization-oriented peer interaction → fewer knowledge-related smells (indirect/mediated complementarity).
  • Coordination path: AI → direct association with communication-related outcomes and information sharing (direct complementarity, no change in interaction frequency).
• Contribution: an empirically validated, TMS-grounded model and a reusable measurement instrument for studying AI-mediated teamwork in software engineering.

Data & Methods

• Sample: Survey data from 152 software professionals actively using AI tools in collaborative team settings.
• Instrument development: Community-smell constructs adapted from existing catalogs in the literature, validated through expert surveys and Exploratory Factor Analysis (EFA).
• Main analysis: Partial Least Squares Structural Equation Modeling (PLS-SEM) to estimate five structural models focused on the Specialization and Coordination dimensions of TMS.
• Scope/limitations noted by authors:
  • Cross-sectional, perception-based survey data (associations, not causal claims).
  • Focused on two TMS dimensions; did not model interpersonal credibility/trust in human teammates (distinct from trust in AI).
  • Sample size moderate (n=152); models reported and validated but further longitudinal/experimental work is needed.

Implications for AI Economics

• Complementarity vs substitution: Evidence suggests AI often complements human work rather than displacing interpersonal interactions. For specialization tasks, AI can increase peer consultation (complementary effect), reducing costly knowledge bottlenecks. For coordination tasks, AI directly affects communication quality without reducing contact frequency—again pointing to complementarity.
• Human capital and skill formation: If AI use increases peer consultation in specialization activities, it may support knowledge diffusion and reduce concentration of expertise (lower "truck factor"), lowering organizational risk and potentially reducing returns to a few star individuals. Conversely, because information-sharing is directly shaped by AI interaction, firms should monitor whether reliance on AI changes incentives for documentation and knowledge codification—affecting on-the-job learning and long-run skill accumulation.
• Productivity and quality externalities: Reduced knowledge fragmentation and improved communication quality should lower coordination costs, reduce rework, and decrease technical debt—translating into measurable productivity and quality gains. However, these gains depend on governance and how AI is integrated; poor governance could create new social frictions with economic costs.
• Organizational investment priorities: Firms should invest not only in AI tools but also in governance (documentation practices, protocols for Human–AI workflows), training (to use AI discerningly), and social-health monitoring. These investments influence the realized economic returns from AI adoption.
• Labor-market and wage implications: If AI adoption makes knowledge more broadly accessible within teams, the premium for narrow, concentrated expertise may fall, while demand could rise for integrative skills (coordination, AI-augmented evaluation, and governance). Policymakers and HR should anticipate shifting skill premiums.
• Research & measurement: Economists studying AI impacts should measure social/organizational channels (knowledge fragmentation, communication quality) as mediators between AI adoption and productivity outcomes. Longitudinal and quasi-experimental designs are needed to estimate causal effects and to quantify economic magnitudes (e.g., cost savings from reduced technical debt, changes in turnover risk).

Practical takeaways for managers and economists: treat AI as a potentially task-contingent complement, monitor both documented knowledge governance and informal peer interactions, and allocate resources toward training and coordination mechanisms that preserve or strengthen transactive memory (who-knows-what) as AI is introduced.