A Process Harness for Uplifting Legacy Workflows to Agentic BPM: Design and Realization in CUGA FLO
Fabiana Fournier, Lior Limonad
Read on arXiv →Key claim
Enhances legacy workflows with an agentic decision-making layer.
In plain English
Imagine you're managing a complex business process, like loan approvals, where you need to balance strict rules with the flexibility to adapt to unique cases. Traditionally, you might rely on a rigid workflow engine that executes tasks in a predetermined way, but this can lead to inefficiencies when unexpected situations arise. For instance, if a customer has a unique request that doesn't fit neatly into the established process, the system might struggle to handle it appropriately, leading to delays or errors. This is what's called a lack of adaptability in workflow management.
To address these challenges, the authors propose a solution that adds a new layer around the existing workflow engine. This layer, called a process harness, introduces three types of agents that work together to enhance the system's capabilities. The TaskAgent focuses on executing knowledge-intensive tasks, the DecisionAgent helps route cases based on specific criteria, and the FlowAgent manages the overall flow of the process, allowing for adjustments as needed. By doing this, the process harness maintains the structural authority of the original workflow engine while enabling more intelligent decision-making at critical points.
The practical impact of this approach is significant: it allows organizations to modernize their workflows without completely replacing their existing systems. By integrating this agentic layer, businesses can achieve a balance between compliance with established rules and the flexibility to adapt to new situations, ultimately leading to more efficient and responsive operations.
The process harness introduces a new agentic layer for workflow management, extending existing BPM frameworks.
The claims are supported by a concrete implementation and demonstration on a specific use case.
Deep reliability assessment
The methodology supports the paper as a conceptual architecture and design realization: TDF defines roles, policies, and execution semantics, and CUGA FLO is illustrated through a loan approval workflow. It does not support claims about reliability, productivity, compliance improvement, or general applicability across real legacy BPM systems because there is no quantitative evaluation, baseline comparison, or production deployment evidence in the provided text.
Reproducibility
No: no open-source code, dataset, benchmark suite, or executable artifact is mentioned; the paper describes the TDF model and a loan approval demonstration but does not provide enough implementation detail to independently reproduce results.
Key figure
The key architectural diagram depicts a process harness as a policy-governed agentic overlay around a deterministic BPM/workflow engine, with TaskAgent, DecisionAgent, and FlowAgent invoked at designated control points while the engine maintains control of the process structure.
