latency-reliability-cost

This paper analyzes tradeoffs between latency, reliability, and cost in LLM-enabled agentic workflows, introducing performance models and deriving optimal resource allocation policies like water-filling token allocation.