Agent Behavior Mining: Generative AI Agent Governance in Business Processes

The paper introduces "Agent Behavior Mining," a governance capability designed to enhance the observability and traceability of decision-making in generative AI agents used in business processes. It presents an event data model that translates detailed agent activities into standardized process logs, allowing for the detection of policy deviations and quantification of operational variability in a multi-agent order-to-cash scenario. This approach is significant for practitioners as it addresses the governance challenges posed by the non-deterministic nature of AI agents, emphasizing the need for behavioral transparency to foster trust in AI-driven business processes.