arXiv 2601.02695

EvoRoute: Experience-Driven Self-Routing LLM Agent Systems

By Guibin Zhang, Haiyang Yu, et al.

Published 2026-01-06

Wiki summary

Explore the paper's summary, context, and related research on Papiers.

Complex agentic AI systems, powered by a coordinated ensemble of Large Language Models (LLMs), tool and memory modules, have demonstrated remarkable capabilities on intricate, multi-turn tasks. However, this success is shadowed by prohibitive economic costs and severe latency, exposing a critical, yet underexplored, trade-off. We formalize this challenge as the Agent System Trilemma : the inherent tension among achi…

View the original paper on arXiv