Agentic AI and Multi-Agent Systems: Governance, Reliability, and the Enterprise Trust Gap in Autonomous Workflow Environments

Dickson Mdhlalose

EconStor Preprints from ZBW - Leibniz Information Centre for Economics

Abstract: The emergence of agentic artificial intelligence (AI) systems capable of autonomous planning, multi-step reasoning, tool invocation, and real-world task execution represents one of the most consequential technological transitions of the mid-2020s. While organisational interest in agentic AI has surged, with McKinsey (2025) reporting that 78% of enterprises have initiated agentic AI experiments, production-scale deployments remain dramatically underdeveloped, with fewer than 24% of pilot programmes successfully transitioning to operational status. This paper investigates the structural forces underpinning the enterprise trust gap: the chasm between agentic AI's demonstrated experimental potential and its reliable, governed, production-scale deployment. Drawing on a systematic review of 60 peer-reviewed publications, technical reports, regulatory documents, and industry surveys spanning 2020 to 2026, this research identifies four core domains of failure: (a) reliability deficits arising from emergent behaviours and misaligned agent objectives; (b) governance lacunae in multi-agent orchestration and oversight; (c) inadequacy of evaluation benchmarks for long-horizon agentic tasks; and (d) legacy system integration challenges that impede observability and safe termination. The study argues that the trust gap is not merely a technical problem but a sociotechnical governance crisis requiring coordinated intervention across organisational, regulatory, and engineering domains. Original contributions include a taxonomy of agentic failure modes, a comparative framework analysis, and a multi-layered governance model for enterprise deployment. Actionable recommendations are directed at AI developers, enterprise adopters, and policymakers, emphasising the urgency of robust observability infrastructure, standardised evaluation protocols, and institutionalised human-in-the-loop oversight mechanisms.

Keywords: Agentic AI; Multi-agent systems; AI governance; Enterprise AI; Trust gap (search for similar items in EconPapers)
JEL-codes: D82 K20 L86 M15 O33 (search for similar items in EconPapers)
Date: 2026
References: Add references at CitEc
Citations:

Downloads: (external link)
https://www.econstor.eu/bitstream/10419/341499/1/A ... S_Research_Paper.pdf (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:zbw:esprep:341499

Access Statistics for this paper

More papers in EconStor Preprints from ZBW - Leibniz Information Centre for Economics Contact information at EDIRC.
Bibliographic data for series maintained by ZBW - Leibniz Information Centre for Economics ().