Action-Oriented Programming and Automatic Agent Generation for Adaptive Data Collection in Decentralized Data Ecosystems

The semiconductor manufacturing industry depends on effective data collection and analysis for critical processes such as Root Cause Analysis (RCA) and Risk Assessment (RA). Both processes involve software-driven data collection and subsequent analysis by domain experts to support informed decision-making. However, the increasing complexity, volume, and decentralized nature of manufacturing data pose significant challenges for effective data collection. Data is distributed across multiple systems with varying formats and ownership, making conventional programming paradigms and manual data collection scripts inadequate for handling this decentralized data landscape. To address these challenges, this study proposes integrating Action-Oriented Programming (AcOP) with Automatic Agent Generation (AAG) as a novel solution. AcOP emphasizes actions as fundamental execution units, separating system behavior and data. Complementing this, AAG uses large language models (LLMs) to autonomously generate intelligent agents, which manage these actions and perform preliminary data analysis with domain-specific knowledge. Our experimental setup compares three microservice applications supporting RCA and RA: Object-Oriented Programming (OOP), AcOP, and AcOP integrated with AAG. Evaluation results indicate that AcOP improves modularity, adaptability, and error handling in decentralized systems. Integrating AAG enhances automation, provides a flexible, low-maintenance solution for data collection and analysis pipelines, and promotes autonomous microservice architectures in data-intensive environments.