It is argued that agentic AI governance is four problems, not one, each with a mature governing science, and a five-level maturity model with a non-compensatory bottleneck-weighted index and assessment instrument operationalizes CASE as a scientific rather than process maturity model, grounded in production enterprise agentic platforms.
Abstract
Enterprises are deploying autonomous AI agents faster than they can govern them, and prevailing approaches stretch a single discipline, typically DevSecOps built for deterministic automation, across every scale of agency. We argue that agentic AI governance is four problems, not one, each with a mature governing science. The CASE framework assigns Control theory to the individual agent (intent as setpoint, guardrails as feedback, evaluation as observation), complex Adaptive systems theory to agent collectives (where emergence makes single-agent assurance non-compositional), Supervisory cybernetics to human-agent teams (where the Law of Requisite Variety shows unaided human oversight fails structurally), and Engineering operations to fleets (extending error budgets to decision quality so autonomy becomes a controlled variable). We formalize each layer, derive cross-layer coupling conditions, including a zero-touch deployment paradox where excellence at one-layer strains the others, and trace twenty-plus enterprise controls to their classical constructs. Three empirical studies validate the thesis: 82 percent of documented production agent failures are multi-layer trajectories; none of 22 ecosystem tools offers full Layer 2 (emergence) coverage; and all 35 scored public deployments fall in the lowest maturity band. We name this mismatch, risk realized at the emergence layer against capability barely offered and practice absent, the Emergence Gap. A five-level maturity model with a non-compensatory bottleneck-weighted index and assessment instrument operationalizes CASE as a scientific rather than process maturity model, grounded in production enterprise agentic platforms. As EU AI Act Article 14 makes effective human oversight a legal requirement, only architectures satisfying requisite variety can make oversight real rather than ceremonial.
Agentic artificial intelligence represents a significant shift in organisational AI because such systems can interpret goals, plan multi-step workflows, use tools, access data, and execute actions with varying levels of autonomy. This policy paper examines whether organisations are ready to “hand over the keys” to these systems. It argues that the answer is conditional and domain-specific: many organisations are ready to use agentic AI as a supervised assistant in bounded, low-risk and reversible workflows, but most are not ready to delegate broad discretionary authority in high-stakes contexts. Drawing on supplied source materials and recent research on reliability, governance, risk alignment, oversight, security, AI management systems, and enterprise adoption, the paper develops a six-dimensional readiness model that covers technical reliability, authority and identity, data and tool governance, human oversight, accountability, and workforce legitimacy. The main finding is that agentic AI readiness is constrained not only by model performance but also by socio-technical governance gaps, including tool misuse, prompt injection, memory poisoning, privilege abuse, cascading failures and ineffective human review. The paper concludes that organisations should adopt staged delegation, maintain agent registries, assign unique agent identities, apply least privilege, red-team agentic workflows, formalise oversight and prohibit high-stakes autonomy until assurance evidence is substantially stronger.
Vincent English, Marthinus Van den Berg· International Business Resea...· 0 citations
This paper proposes two complementary artifacts: an Agency Justification Record (AJR) helps teams decide when an agent is warranted over simpler alternatives and an Agentic Delegation Policy (ADP) captures what must be specified for safe and effective development.
Chetan Arora, Andreas Vogelsang, Abbishek Sharma· arXiv.org· 0 citations
A unified, taxonomy-driven, and deployment-oriented survey of agentic AI systems, synthesizing recent advances through a modular reference architecture and a four-dimensional taxonomy that characterizes agents along the axes of autonomy, tool use, collaboration, and safety–governance is presented.
Sparsh Bajoria, Shreyanshu Ranjan, Adhitya M et al.· Cognitive Computation· 0 citations
With the advent of Artificial Intelligence (AI), the world of enterprise automation has radically changed to an AI multi-agent ecosystem with coordination across functional teams and the capacity to make autonomous decisions. Despite this, many companies are still discontinuing the implementation of AI, with partial integration into their processes, weak systems integration, and a lack of a sense of network in some business units. It introduces the concept of the traditional enterprise transforming into an intelligent, autonomous enterprise with the help of AI in logistics, knowledge management, finances, HR, cybersecurity, compliance, customer support, and operational analytics, and also introduces the Multi-Agent Enterprise Framework (MAEF) as the scalable architecture. The proposed architecture has four layers: shared memory, human in the loop, policy-driven control, and orchestration layer, which are necessary for safe, transparent, and trustworthy cooperation between the set of specialized agents. Training is conducted in a highly realistic business environment that includes several departments, numerous workflow requests, and is evaluated and tested against standard automated and single-agent AI systems. Experimental results show that workflow automation and task completion time have been enhanced, cross-department collaboration has been effective, operational efficiency has been achieved, and resources are used optimally; meanwhile, the governance and compliance requirements are met. Agreeing with these conclusions, it seems that enterprise-wide multi-agent systems are a good building block for digital enterprises capable of adapting, scaling, and operating autonomously, on which future intelligent businesses would be able to operate.
Swaroop Suresh Borukar· International Research Journ...· 0 citations
This paper argues that the introduction of agentic AI requires a substantial expansion of traditional enterprise architecture principles to address new behavioral, security, and governance risks emerging from non-deterministic AI systems interacting with heterogeneous operational platforms-ERP, HCM, CLM, asset management, workflow engines, and domain-specific applications.
Elizabeth Koumpan, Vimal Dimpi· AHFE International· 0 citations
Multi-agent frameworks built on large language models (LLMs) routinely entangle three logically distinct concerns: who is on the team (organization), how members align (coordination), and which algorithm fuses their work (collaboration protocol). IMACS (Intelligent Multi-Agent Collaboration System) separates the three into orthogonal, independently swappable layers. Classic organizational theory (Belbin roles, Mintzberg coordination, RACI accountability) becomes executable, validated configuration, and the framework places six published collaboration algorithms behind a common interface while exposing roles, coordination, and accountability as independently configurable factors. We use this separation to conduct controlled comparisons in which organizational assignments vary while the collaboration protocol is held fixed. It also turns protocol choice into a variable that can be learned: Adaptive Org Routing, a contextual-bandit meta-protocol, selects a protocol per task under an explicit quality-cost tradeoff, outperforms every fixed protocol in a controlled study, and trains online on real benchmark and LLM-judge rewards. The ablations expose a mechanism. Accountability placement changes outcomes exactly when the protocol routes the deliverable through the accountable agent, and the winning placement flips across model families, so organizational design cannot be hard-coded; it must be revalidated, or learned, for each model binding.
Huan-Wei Chen, Xiang Song, Jian Jin et al.· arXiv.org· 2 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
