Workplace AI companions matter for relationships, not just tasks: a new framework argues that cognitive, emotional and developmental competencies arise through sustained worker–AI interaction and should shape design and organizational policy.

Main Finding

Workplace AI companions generate value not just through task capabilities but through sustained, evolving relationships with workers; their value is best understood via a relational competency framework comprising cognitive, emotional, and developmental domains. These competencies are relational achievements that emerge from ongoing worker–AI interaction, with important implications for design, wellbeing, and organizational outcomes.

Key Points

• Framework: Relational competency organized into three domains
  • Cognitive competency: supports reasoning, decision support, and task performance.
  • Emotional competency: enables affective engagement, trust, and regulation of worker emotions.
  • Developmental competency: reflects mutual learning, adaptation, and relationship evolution over time.
• Competencies are not fixed system attributes but outcomes of interaction—shaped by context, worker characteristics, and organizational practices.
• Shifts analytical focus from "what AI can do" (capabilities) to "how human–AI relationships are sustained" (processes, trajectories, and co-development).
• Interdisciplinary synthesis: integrates insights from HCI, organizational studies, sociotechnical systems, and AI/ML literature.
• Practical stakes: design choices, managerial practices, and policy affect whether AI companionship produces productivity gains, wellbeing benefits, or harms.

Data & Methods

• Approach: Systematic literature review and conceptual synthesis across disciplines concerned with workplace AI companions.
• Scope: Studies of AI systems with which workers form sustained relationships (companion-like systems embedded in organizational life).
• Methodological output: A theoretical/reconceptualized framework (relational competency) derived from cross-cutting empirical and conceptual findings in the reviewed literature.
• Note: The paper is a conceptual contribution rather than an original empirical study; it synthesizes existing empirical work to produce testable propositions.

Implications for AI Economics

• Productivity and complementarities
  • Relational competencies can create complementarities between AI and labor beyond pure task automation—affective trust and developmental learning may amplify productivity gains.
  • Economic models should allow AI value to depend on interaction intensity, relationship age, and mutual adaptation, not just task performance metrics.
• Human capital and skill formation
  • Developmental competence implies on-the-job learning mediated by AI companions; AI can act as a form of organizational capital that builds worker skills (or deskills them, depending on design).
  • Consider long-run effects on wage trajectories, training investments, and career dynamics.
• Labor demand and job design
  • Adoption effects may differ by task type: relationally rich tasks may see increased complementarity, whereas routine tasks may be substituted.
  • Firms may reorganize work around sustained human–AI pairings (team composition, monitoring regimes, incentive design).
• Firm-level returns and adoption heterogeneity
  • Returns to AI investments depend on managerial practices that foster relational competencies (onboarding, calibration, feedback loops), producing heterogeneity across firms and sectors.
  • Important to model adoption as joint investment in technology and relational infrastructure.
• Welfare, wellbeing, and externalities
  • Emotional competency affects worker wellbeing, stress, and turnover—non-monetary welfare effects should be incorporated into policy and cost–benefit analyses.
  • There may be spillovers (peer effects, customer experience) from worker–AI relationships that standard productivity metrics miss.
• Measurement and empirical strategy recommendations
  • Key variables to collect: measures of trust/affective engagement, interaction frequency/duration, learning/adaptation rates, task accuracy, productivity, turnover, and subjective wellbeing.
  • Favor longitudinal and panel designs to capture developmental competence and causal dynamics; randomized deployments or staggered rollouts can identify effects.
  • Consider production-function extensions that include relational AI capital and interaction terms for worker skill and relationship maturity.
• Policy and regulation
  • Policies (disclosure, transparency, labor protections) influence relational dynamics and hence economic outcomes; regulation that shapes interaction norms may alter the balance of benefits/harms.
• Research gaps relevant to economists
  • Quantify the marginal product of relational competencies and their persistence.
  • Identify distributional effects across worker skill levels, occupations, and firm types.
  • Explore optimal investment mixes (technology vs. relational/organizational capital) and contract/incentive designs that internalize relational externalities.

Suggested next empirical steps for researchers in AI economics: build panel datasets around deployments of companion-like systems, collect both objective performance and subjective relational metrics, exploit randomized or phased rollouts to estimate causal effects, and incorporate relational-capital terms into firm production and wage-setting models.