Agentic AI is increasingly used in engineering workflows, from PLC code generation to the orchestration and ERP. Agents can handle unstructured input flexible, but their non-deterministic behaviour can be a problem in industrial environment, where safety, quality and compliance with applicable rules are essential. Historically, automation in engineering relied on deterministic algorithms: precise, repeatable, and inherently predictable. While these systems excelled in reliability, their rigidity limited adaptability in complex, dynamic environments.

Recent work has proposed agentic Business Process Model and Notation (BPMN) concepts as a solution, wherein BPMN serves as a rule-based orchestrator. 
This approach integrates agentic tasks into a structured framework, allowing processes to remain machine-readable and deterministic at their core while enabling flexible execution where needed. 
However, integrating agentic decisions, particularly those originating from enterprise-level systems like ERP, into these rule-based workflows raises important questions about trust, reliability, and control.
This thesis addresses these challenges by investigating whether agentic and deterministic elements can be combined within industrial automation. 

GOALS:

• Conceptual design for usage of agentic and deterministic elements in industrial workflows including clear criteria for deciding which task is handled agentically, which deterministically or where hybrid architectures are sensible
• Prototypical implementation of the concepts on 2–3 industrial use cases
• Validation of the approach and compare hybrid, purely agentic, and purely deterministic variants regarding efficiency, reliability, and engineering effort.

HELPFUL PRIOR KNOWLEDGE:

• Interest in automation engineering workflows
• Prior knowledge in mechanism of agentic AI 
• Lectures IT/AT, MARO and CPM