Abstract
As large language models are increasingly integrated into automated systems, contemporary architectures default almost universally to autonomous agent loops. However, iterative agent execution introduces unpredictable latency, compounding failure modes, and high token overhead. This working paper formulates a structured research agenda investigating when deterministic state machines, specialized tool callers, and hybrid architectures outperform unbounded agentic loops on representative enterprise tasks. We outline the theoretical trade-offs, evaluation metrics, and initial experimental methodology for the YR-001 research project.
Publication Integrity & Transparency Notice
This publication is currently an active Working Paper / Research Proposal. Full technical reports and empirical evaluation datasets will be attached as benchmarks are executed under the YR-001 research project. We do not claim peer-reviewed conference acceptance until formal review is concluded.
Cite this Publication
BibTeX Format@techreport{padwal2026beyondagentic,
title={Beyond Agentic AI: Exploring Deterministic, Agent-Based, and Hybrid Intelligence Architectures},
author={Padwal, Aditya},
institution={YARA Research},
year={2026},
number={YR-WP-2026-001},
url={https://yararesearch.com/publications/beyond-agentic-ai-architectures}
}