Pythia: Exploiting Workflow Predictability for Efficient Agent-Native LLM Serving

TL;DR

Pythia leverages the structured predictability of multi-agent workflows to optimize LLM serving, significantly enhancing throughput and reducing delays compared to traditional systems.

Contribution

This paper introduces Pythia, a novel multi-agent serving system that exploits workflow semantics to improve efficiency in LLM deployment.

Findings

Identified key bottlenecks like low cache hit rates and resource contention.

Demonstrated Pythia's substantial improvements over existing baselines.

Analyzed production traces to inform system design.

Abstract

As LLM applications grow more complex, developers are increasingly adopting multi-agent architectures to decompose workflows into specialized, collaborative components, introducing structure that constrains agent behavior and exposes useful semantic predictability. Unlike traditional LLM serving, which operates under highly dynamic and uncertain conditions, this structured topology enables opportunities to reduce runtime uncertainty$\unicode{x2015}$yet existing systems fail to exploit it, treating agentic workloads as generic traffic and incurring significant inefficiencies. Our analysis of production traces from an agent-serving platform and an internal coding assistant reveals key bottlenecks, including low prefix cache hit rates, severe resource contention from long-context requests, and substantial queuing delays due to suboptimal scaling. To address these challenges, we propose Pythia, a multi-agent serving system that captures workflow semantics through a simple interface at the serving layer, unlocking new optimization opportunities and substantially improving throughput and job completion time over state-of-the-art baselines.