Paper Overview
Field: Machine Learning Authors: Ya-Ting Yang, Quanyan Zhu Published: 2026-05-26 arXiv: 2505.21640
Summary
Modern AI systems increasingly rely on workflows composed of multiple interacting agents, some powered by large language models (LLMs) and others by conventional computational modules. This paper analyzes the fundamental tradeoffs between latency, reliability, and cost in LLM-enabled agentic workflows.
The authors introduce performance models for both LLM and non-LLM agents that capture the relationship between computational effort and output quality, incorporating the impact of reasoning and output tokens for LLM agents using a parametric exponential reliability function.
Key Contributions
- Performance models for LLM and non-LLM agents relating computational effort to output quality
- A parametric exponential reliability function capturing the effect of reasoning and output tokens
- Analysis of sequential workflow design under latency and cost constraints
- A water-filling token allocation policy for optimal token distribution
- Characterizations of optimal workflow reliability in terms of shadow prices
Original Abstract
Modern AI systems increasingly rely on workflows composed of multiple interacting agents, some powered by large language models (LLMs) and others by conventional computational modules. This paper analyzes the fundamental tradeoffs between latency, reliability, and cost in LLM-enabled agentic workflows. We introduce performance models for both LLM and non-LLM agents that capture the relationship between computational effort and output quality, incorporating the impact of reasoning and output tokens for LLM agents using a parametric exponential reliability function. Then, we study the design of sequential workflows under latency and cost constraints. Main results include a water-filling token allocation policy and characterizations of optimal workflow reliability in terms of shadow prices.