Agentic AI confronts firms not with isolated obstacles but an interdependent sociotechnical web: most barriers are amplified legacy transformation problems, but a few—error propagation in multi-agent systems, role ambiguity, and accountability diffusion—are novel. The central managerial challenge is governance calibration: enabling valuable autonomy while constraining risks requires joint optimization across social and technical subsystems.

BARRIERS TO AGENTIC AI ENTERPRISE TRANSFORMATION

Roman Kuziv · March 31, 2026 · Economic Synergy

openalex review_meta n/a evidence 7/10 relevance DOI Source PDF

A narrative review classifies 29 barriers to enterprise transformation under agentic AI into five dimensions, finds most are amplified versions of prior digital-transformation problems while three are agentic-specific, and frames the central challenge as a governance calibration (joint optimization) problem across sociotechnical subsystems.

This study identifies, classifies, and critically analyzes barriers to enterprise transformation under the influence of agentic AI – autonomous software that leverages large language models (LLMs) to perceive its environment, reason through complex tasks, plan and execute actions, and use tools to achieve goals with minimal human oversight.A critical narrative literature review of 30 sources (2019–2026) was conducted. Barriers were identified inductively through open and axial coding; Sociotechnical Systems (STS) theory was then applied as an interpretive lens to map the resulting dimensions onto social and technical subsystems and analyze cross-subsystem interactions.Twenty-nine barriers were classified into five dimensions: technological (7), organizational (7), human (6), governance and regulatory (4), and economic (5). Each barrier was assessed for agentic specificity. Three barriers were identified as agentic-specific (error propagation in multi-agent systems, role ambiguity, accountability diffusion). At the same time, the remaining 26 are carried over from prior digital transformation waves – 22 in amplified form and 4 unchanged. STS mapping based on root-cause analysis revealed that 12 barriers originate in the technical subsystem and 17 in the social subsystem, with governance serving as the social subsystem's primary mechanism for managing the technical subsystem. Five interaction mechanisms were identified, with the majority propagating across the subsystem boundary.Agentic AI transformation barriers constitute an interdependent sociotechnical system rather than isolated obstacles. The governance calibration problem – balancing control with the autonomy that gives agentic AI its value – emerges as the STS joint optimization challenge: governance, as the social subsystem's mechanism for managing the technical subsystem, must simultaneously enable and constrain autonomous operation.The taxonomy provides a diagnostic framework for identifying priority barrier dimensions and understanding cross-dimensional amplification mechanisms. The agentic-specificity classification helps organizations distinguish challenges that require novel approaches from those that are addressable with established practices.

Summary

Main Finding

The paper develops a five-dimensional, sociotechnically grounded taxonomy of 29 barriers to enterprise transformation driven by agentic AI (autonomous, LLM-powered multi‑agent systems). It shows that most barriers are not isolated but form an interdependent sociotechnical system; only three barriers are agentic‑specific (error propagation in multi‑agent systems, role ambiguity, accountability diffusion), while the rest are inherited from earlier digital transformations (22 amplified, 4 unchanged). Mapping barriers onto Sociotechnical Systems (STS) subsystems reveals 12 originating in the technical subsystem and 17 in the social subsystem, with governance playing the central role as the social mechanism for managing technical capabilities. The critical STS challenge identified is the governance‑calibration problem: jointly optimizing control and agentic autonomy.

Reference: Kuziv, R. (2026). Barriers to agentic AI enterprise transformation. ECONOMIC SYNERGY (1 (19)). DOI: https://doi.org/10.53920/ES-2026-1-18. JEL: O33, M15, L86.

Key Points

Taxonomy: 29 barriers classified into 5 dimensions
- Technological: 7 barriers
- Organizational: 7 barriers
- Human: 6 barriers
- Governance & Regulatory: 4 barriers
- Economic & Financial: 5 barriers
Agentic specificity:
- Agentic‑specific barriers (3): error propagation in multi‑agent systems; role ambiguity; accountability diffusion.
- Inherited but amplified barriers (22) and unchanged inherited barriers (4).
STS mapping and origin:
- 12 barriers traced to the technical subsystem (tools, infrastructure, data, architectures).
- 17 barriers traced to the social subsystem (roles, processes, culture, governance).
- Governance is the primary social mechanism for managing technical subsystems.
Interaction structure:
- Barriers interact across subsystem boundaries; the study identifies five interaction mechanisms (majority cross the social‑technical boundary), producing self‑reinforcing cycles that amplify transformation failure risks (e.g., investing in infrastructure while governance or absorptive capacity remains weak).
Governance‑calibration problem:
- Key joint‑optimization challenge: calibrating governance intensity so it constrains unacceptable risks without eliminating the autonomy that produces agentic AI value (too little governance → trust/risk failures; too much → “agent washing” or suppressed value).
Human/organizational insights:
- Agentic AI induces continuous, open‑ended change (no fixed “after” state), undermining classical change management models and increasing risk of change fatigue and oversight cognitive costs.
Practical diagnostic use:
- The taxonomy is intended as a diagnostic framework to prioritize intervention dimensions and distinguish where novel solutions are required (agentic‑specific) versus where established practices can be adapted.

Data & Methods

Methodological approach: Critical narrative literature review (suitable for emerging topics where systematic review corpus is limited).
Corpus: 30 sources published 2019–2026, combining peer‑reviewed research, top‑tier conference papers, and methodologically transparent industry reports (examples cited include Deloitte, Gartner, WEF, Makarius et al., Papagiannidis et al., Sapkota et al.).
Analysis:
- Inductive barrier identification via open and axial coding.
- Root‑cause analysis to map barriers onto STS social and technical subsystems.
- Agentic‑specificity coding to classify barriers as agentic‑specific vs. inherited (and whether inherited are amplified).
- Interpretive application of Sociotechnical Systems (STS) theory to analyze cross‑subsystem interactions and identify joint‑optimization challenges.
Limitations noted by the author:
- Emerging field with limited peer‑reviewed empirical deployments of agentic AI (most evidence from early pilots and industry surveys).
- Narrative review tradeoffs: depth and theoretical framing prioritized over exhaustive systematic coverage.

Implications for AI Economics

Adoption & diffusion dynamics:
- The taxonomy explains the gap between high exploration/pilot rates and low production deployment (cited survey: 30% exploring, 38% piloting, 11% in production). Economic models of AI diffusion should incorporate sociotechnical constraints and feedback loops (governance, absorptive capacity) rather than only technology cost curves.
Investment and cost‑benefit assessment:
- Agentic AI imposes different infrastructural and organizational investment profiles (real‑time integration, shared memory, identity/security, continuous monitoring). Cost estimates must include governance, continuous change management, and cognitive/oversight costs.
Market structure and vendor behavior:
- “Agent washing” risk implies information asymmetries and market signaling problems. Regulators and purchasers may need standards to distinguish genuinely agentic capabilities from rebranded automation, affecting procurement and competitive dynamics.
Regulatory and compliance costs:
- Regulatory fragmentation across jurisdictions raises compliance complexity and costs, affecting multinational firms’ adoption decisions and possibly creating regulatory arbitrage or uneven market advantages.
Labor markets and productivity:
- Continuous scope expansion and role ambiguity can shift task boundaries and create accountability diffusion, altering how labor is compensated, insured, and regulated. Economists should model transitional frictions (reskilling costs, change fatigue, oversight productivity losses) when projecting net welfare impacts.
Policy and public economics:
- The governance‑calibration problem suggests policy roles beyond ex post liability: standards for governance intensity, certification of agentic systems, interoperability and auditability requirements, and incentives for organizational investments in absorptive capacity (training, shared processes).
Modeling recommendations:
- Incorporate sociotechnical joint‑optimization constraints into models of firm adoption and social welfare (endogenize governance choice; include externalities from error propagation in multi‑agent networks).
- Empirical research needed to quantify barrier‑specific impacts (e.g., cost of governance calibration, expected loss from error cascades) to calibrate macro and microeconomic models of agentic AI.
Research priorities:
- Quantitative, empirical studies of production‑grade agentic AI deployments to measure barrier prevalence and economic impacts.
- Comparative policy analysis of governance regimes and their effects on adoption, innovation, and market entry.
- Microfoundations for accountability diffusion and multi‑agent error externalities to feed into regulation and firm strategy.

If you want, I can: - Extract the 29 barrier labels into a compact checklist. - Translate this summary into Ukrainian. - Draft suggested empirical metrics and variables to operationalize the taxonomy for quantitative research.

Assessment

Paper Typereview_meta Evidence Strengthn/a — This is a critical narrative literature review and conceptual taxonomy rather than an empirical study that estimates causal effects; it synthesizes and interprets existing literature rather than producing new causal evidence. Methods Rigormedium — The study uses inductive open and axial coding on a moderately sized corpus (30 sources) and applies Sociotechnical Systems theory for interpretive mapping, which is appropriate for conceptual taxonomy work; however, it is a narrative (not systematic) review, lacks reported protocol details (search strategy, inclusion/exclusion criteria, inter-coder reliability), and therefore is vulnerable to selection and interpretation bias. SampleA critical narrative literature review of 30 sources published 2019–2026 (mix of academic articles, conference papers, and industry/ policy reports) addressing agentic AI and digital/enterprise transformation; barriers were identified via inductive coding and analyzed through a Sociotechnical Systems lens. Themesorg_design governance human_ai_collab GeneralizabilityLimited by narrative (non-systematic) selection of 30 sources — possible selection and publication bias, Findings reflect the state of the literature through 2026 and may omit very recent empirical work or unpublished industry practices, Heterogeneity of organizational contexts in source material (sectors, firm sizes, geographies) reduces precision for any specific context, Conceptual taxonomy and interpretive STS mapping are transferable as analytic tools but not directly predictive of firm-level outcomes, Focus on agentic AI may not generalize to non-agentic AI deployments or narrowly scoped automation technologies

Claims (13)

Claim	Direction	Confidence	Outcome	Details
A critical narrative literature review of 30 sources (2019–2026) was conducted. Other	null_result	high	study_design	n=30 0.4
Barriers were identified inductively through open and axial coding. Other	null_result	high	analytic_method	n=30 0.24
Sociotechnical Systems (STS) theory was applied as an interpretive lens to map dimensions onto social and technical subsystems and analyze cross-subsystem interactions. Other	null_result	high	analytic_framework_application	n=30 0.24
Twenty-nine barriers were identified and classified into five dimensions: technological (7), organizational (7), human (6), governance and regulatory (4), and economic (5). Organizational Efficiency	null_result	high	number_and_classification_of_barriers	n=30 0.24
Three barriers were identified as agentic-specific: error propagation in multi-agent systems, role ambiguity, and accountability diffusion. Governance And Regulation	negative	high	agentic_specific_barriers	n=30 0.24
The remaining 26 barriers are carried over from prior digital transformation waves — 22 in amplified form and 4 unchanged. Organizational Efficiency	negative	high	novelty_vs_carried_over_of_barriers	n=30 22 amplified; 4 unchanged 0.24
STS mapping based on root-cause analysis revealed that 12 barriers originate in the technical subsystem and 17 in the social subsystem. Organizational Efficiency	null_result	high	subsystem_origin_of_barriers	n=30 12 technical; 17 social 0.24
Governance serves as the social subsystem's primary mechanism for managing the technical subsystem. Governance And Regulation	null_result	high	role_of_governance_between_subsystems	n=30 0.04
Five interaction mechanisms were identified, with the majority propagating across the subsystem boundary. Organizational Efficiency	null_result	medium	interaction_mechanisms_and_propagation	n=30 five mechanisms; majority propagate across boundary 0.14
Agentic AI transformation barriers constitute an interdependent sociotechnical system rather than isolated obstacles. Organizational Efficiency	null_result	high	interdependence_of_barriers	n=30 0.04
The governance calibration problem — balancing control with the autonomy that gives agentic AI its value — emerges as the STS joint optimization challenge: governance must simultaneously enable and constrain autonomous operation. Governance And Regulation	null_result	high	governance_calibration_challenge	n=30 0.04
The taxonomy provides a diagnostic framework for identifying priority barrier dimensions and understanding cross-dimensional amplification mechanisms. Adoption Rate	positive	high	usefulness_of_taxonomy_for_diagnosis	n=30 0.04
The agentic-specificity classification helps organizations distinguish challenges that require novel approaches from those that are addressable with established practices. Organizational Efficiency	positive	high	practical_utility_of_agentic_specificity_classification	n=30 0.04