Agentic Orchestration: A Process-Level Governance Construct for AI-Enabled Organisational Decision Processes — Definition, Architecture, and a Measurement-Development Protocol

Abstract

AbstractArtificial intelligence now participates in managerial decisions by generating options, ranking alternatives, and shaping professional judgment. Research explains delegation, reliance, automation, accountability, and human oversight, but treats these elements largely separately, leaving organisations without a construct for describing how a particular decision process is governed. This conceptual and methodological paper develops Agentic Orchestration as a process-level governance configuration: the arrangement through which an organisation distributes decision tasks, assigns practically exercisable decision rights, provides the epistemic conditions for evaluating AI contributions, and establishes accountability and contestability across human and machine participants. A focused theory synthesis across five research streams yields four constituent dimensions, with accountability and contestability measured as distinct subdimensions, and positions orchestration adaptivity as a separate layer rather than a fifth dimension. This two-layer specification distinguishes configuration as a state from revision as a capability and so avoids conflation with dynamic capabilities. Agentic Orchestration is differentiated from trust, reliance, human-in-the-loop design, meaningful human control, algorithmic management, sociomateriality, dynamic capabilities, and general AI-governance maturity. Because the dimensions are non-interchangeable, a Type II reflective–formative specification is advanced as one candidate model to be compared with four alternatives, with identification obtained through redundancy analysis. The paper presents construct boundaries, a nomological network with five propositions, a 45-item provisional pool, and a five-stage validation protocol. No claim of empirical validation is made. The contribution is to convert a diffuse debate about human–AI agency into an observable organisational construct. Status: Working paper — not peer reviewed.