Workflow-Governed Automated Validation of AI Agent Task Assignments in Microsoft Dynamics 365 ERP
Abstract
AI agents are increasingly deployed in enterprise resource planning (ERP) systems to automate workflow tasks. In platforms such as Microsoft Dynamics 365, autonomous task execution raises unresolved issues around authorization boundaries, data access scope, and accountability. We propose a workflow-governed validation framework that borrows role-based authorization, structured approval routing, and audit trail generation from the ERP workflow engine and applies them as a pre-execution validation layer for agent task requests. The framework intercepts each request, checks it against the existing role-permission policy, and either approves, escalates, or rejects the action without modifying the underlying ERP security model. A proof-of-concept prototype in Microsoft Dynamics 365 uses a procurement scenario where an agent purchase order request exceeds its configured approval limit. The proof-of-concept and analytical evaluation demonstrate policy-conformant routing and model-level audit completeness, while requests within the configured authorization scope remain automated. The architecture is designed for portability to other major ERP systems, subject to platform-specific implementation and validation.