Ethnographies show AI rarely just replaces experts: deployments either fail, become instruments of surveillance and control, create hybrid expert+AI markets, or are ignored — producing unequal labor and market outcomes determined by tacit knowledge and institutional power.

Main Finding

Ethnographic and grounded studies across sociology, management, and STS show that human‑expert ↔ AI interactions fall into four qualitatively distinct arrangements — (1) attempts at complete automation (“no human in the loop”) that largely fail; (2) AI as exploitive/bossy (algorithmic management and platform surveillance), which dominates sociological attention; (3) AI generating new expert markets/roles and productive co‑productions (“yay”); and (4) expert dismissal or ignoring of AI (“nah”). Together these cases show that the economic effects of AI on expert work are heterogeneous, shaped by tacit knowledge, institutional power, trust, market design and professional identity — not just technical capability.

Key Points

• Spectrum of collaboration: conceptualizes human–AI relations as a continuum from full replacement to active co‑production to outright dismissal.
• Failure of full automation: attempts to remove humans often fail because of tacit/contextual knowledge, social accountability, trust, legal/regulatory constraints, and boundary work by professions.
• Algorithmic management (most studied): platforms and firms use AI to monitor, allocate tasks, evaluate performance and discipline workers — producing new forms of control, surveillance externalities, and asymmetric power over labor.
• New markets & roles (“yay”): AI enables new expert services, productized expertise, and hybrid roles (expert + AI operator/designer), creating demand for complementarities and re‑skilling, though also creating professional risks.
• Ignoring/dismissal (“nah”): in many domains experts simply ignore AI recommendations because of professional norms, perceived low reliability, reputational risk, or because AI does not map onto what counts as expertise in practice.
• Cross‑cutting themes: entanglement of technical and social systems; expertise as a form of tacit capital; power, governance and data ownership shape outcomes; context matters — same AI produces different labor/economic effects across settings.

Data & Methods

• Methods: ethnography, participant observation, semi‑structured interviews, archival and field studies; grounded theory and comparative case analysis drawn from sociology, management studies, and STS.
• Typical evidence: detailed workplace case studies (platforms, hospitals, legal firms, professional services), close observation of expert workflows and interactions with AI, document and log analysis where available.
• Strengths: deep contextualized insight into mechanisms, norms, and power dynamics that quantitative aggregate studies miss.
• Limitations: limited external generalizability, potential selection on studied sites, and less emphasis on causal identification and economy‑wide quantification — motivating mixed‑methods followups.

Implications for AI Economics

• Task decomposition & substitution/complementarity
  • Models should treat expertise as bundled tacit and social capital; many tasks are not fully automatable because they require contextual judgement or social accountability.
  • Expect heterogeneous substitution: routine components get automated while tacit, interactive, trust‑dependent tasks persist or grow in value.
• Labor market effects
  • Wage and employment impacts will be uneven: skill polarization in some sectors; creation of premium hybrid roles (expert + AI operator/designer); downward pressure where platforms extract rents.
  • Algorithmic management can reduce worker bargaining power and increase effective labor supply elasticity via surveillance and reputational scoring.
• Market structure and rents
  • Platforms and firms that control data and evaluation algorithms can capture surplus (monopsony/rent extraction); antitrust and governance issues arise.
  • Reputation/trust markets become more important — premium for explainability, verifiability, and liabilities borne by humans.
• Measurement and empirical agenda
  • Need microdata linking task assignments, AI use, wages, and firm outcomes; combine ethnographic insights to correctly classify tasks as tacit vs codifiable.
  • Natural experiments, field experiments, and difference‑in‑differences exploiting staggered AI adoptions can quantify causal effects; complement with workplace ethnographies to unpack mechanisms.
• Policy & regulation
  • Policies should target power asymmetries (platform governance, data portability, algorithmic transparency), protect worker privacy, and support re‑skilling into complementary roles.
  • Liability and audit rules matter: where professional accountability persists, incentives to ignore unreliable AI may be welfare‑enhancing.
• Modeling suggestions for economists
  • Incorporate institutional constraints, trust, and legitimacy premiums into models of automation.
  • Model multi‑task agents with tacit knowledge accumulation costs and endogenous investment in human capital that complements AI.
  • Endogenize platform governance and data ownership to study rent extraction and distributional consequences.

Suggested research priorities for AI economics informed by this review - Build mixed‑methods studies: pair ethnographic case studies with matched administrative/transactional data. - Task‑level decomposition across occupations to measure which elements are codifiable vs tacit. - Longitudinal studies of wage trajectories for hybrid roles created by AI. - Experiments on algorithmic transparency, worker monitoring, and reputation systems to assess behavioral and market responses. - Policy evaluation work on governance interventions (auditability, data portability, worker rights on platforms).

Short takeaway: AI’s economic impact on expert work is context‑dependent; simplistic “automate or not” predictions miss critical institutional, tacit, and power factors. Economics should integrate ethnographic mechanisms into empirical and theoretical models to predict who gains, who loses, and how markets and institutions will adapt.