Skip to main content Skip to main navigation

Publication

Understanding Agentic Systems as Business Processes: A Vision for Process-Driven GenAI Engineering

Andreas Emrich; Jan Gronewald; Michael Frey; Anna K. Hildebrandt; Andreas Hildebrandt; Peter Loos
In: Tobias Greff; Peter Pfeiffer (Hrsg.). Joint Proceedings of the WI 2025 Workshops Regulation of AI Systems and Gen AI for Business Process Management. Internationale Tagung Wirtschaftsinformatik (WI-2025), Workshop on Gen AI for Business Process Management, located at WI-2025, September 13-17, Münster, Germany, CEUR Workshops, 2025.

Abstract

Agentic Systems based on Generative AI address the shortcomings of current LLM approaches by integrating external tools, context and memory, and interacting with the user. In a way, they allow for business processes that leverage the potential of Generative AI. However, current LLM Operation Platforms (LLMOps) as the commercial platforms LangSmith, Phoenix and the OSS solutions Langfuse or OpenLLMetry do not regard these agent interactions as business processes. This paper develops a vision how this class of systems can be extended with methods of Business Process Management (BPM) and Process Mining (PM), in order to allow for an assessment of Agentic System flows in a similar fashion as regular business processes. The paper explores motivational process scenarios in Agentic Systems, compares current LLMOps Platforms and their suitability to capture the required process metrics. The paper concludes with an architectural vision of a process-driven GenAI Engineering Platform and gives on outlook on future implementations.