Beyond Agentic AI: Exploring Deterministic, Agent-Based, and Hybrid Intelligence Architectures
A Comparative Research Agenda on Reliability, Latency, and Cost Trade-offs
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.