International Journal of Science, Engineering and TechnologyAuthors: Dr. Shobana V, Ms.Maheswari P, Ms. Anitha M, Ms.Divyalakshmi.S
Abstract: Enterprise Systems are now becoming increasingly reliant on autonomous decision-making capabilities that go beyond rule-based automation. This paper puts forward an extensive framework for agentic AI systems which include large language models, multi-agent systems, and confidence-based governance which allows for reliable decision-making to be achieved autonomously. The framework takes into consideration the problem of reconciling the need for autonomy and accountability in enterprises and consists of a five-layer system architecture including perception layer, reasoning layer, orchestration layer, execution layer and governance layer. A confidence-based human in the loop system decides on how much governance each decision needs depending on the level of confidence in a particular decision. The experimental evaluation on enterprise workflow systems shows considerable improvement in reducing workflow downtime, improvement in task allocation efficiency and achievement of task autonomy within strict enterprise constraints. Comparisons among the five different agentic AI frameworks show that architecture modularity and governance integration play a critical role in successful implementation of agentic AI systems in real-life settings.
DOI: http://doi.org/10.5281/zenodo.23095376
