新闻
TRACE-ROUTER: Task-Consistent and Adaptive Online Routing for Agentic AI
arXiv cs.AI · 发布于 · 阅读约3分钟
30秒读懂
- 发生了什么
- TRACE-Router routes agentic AI tasks to different LLMs once at admission, then pins all calls to that model, updating routing policy using final task outcome.
- 为何重要
- Enterprise engineers deploying multi-model AI systems where task-level accuracy and latency matter more than optimizing individual LLM calls independently.
- 注意
- Paper is recent preprint with limited real-world deployment data; effectiveness depends on having meaningful delayed task-level feedback signals available.
收听本摘要
文章节选
-->
Computer Science > Artificial Intelligence
arXiv:2607.22465v1 (cs)
[Submitted on 24 Jul 2026 (this version), latest version 27 Jul 2026 ( v2 )]
Title: TRACE-ROUTER: Task-Consistent and Adaptive Online Routing for Agentic AI
Authors: Ritik Raj , Souvik Kundu , Sarbartha Banerjee , Dheemanth Joshi , Ishita Vohra , Tushar Krishna
View a PDF of the paper titled TRACE-ROUTER: Task-Consistent and Adaptive Online Routing for Agentic AI, by Ritik Raj and 5 other authors
View PDF HTML (experimental)
Abstract: Routing to select large language models (LLMs) with different cost-quality trade-offs has become a fundamental deployment feature of enterprise AI. Existing routers, primarily make independent routing decisions for each LLM call. However, agentic applications execute as long-horizon workflows whose quality is determined only by a delayed, task-level outcome. This mismatch prevents per-call routers from correctly attributing feedback to individual routing decisions. Towards mitigating this, we present TRACE-Router, a task-level routing framework that aligns routing with the unit of supervision. TRACE-Router assigns each task to a model once at admission using a contextual bandit, pins all subsequent LLM calls to the selected backend, and updates its policy using the task's terminal reward, jointly accounting for accuracy and latency. By leveraging delayed task feedback,
节选自原文。请前往来源阅读全文。
- agent
- agentic
- llm
- language model
The Agent Architect
每周一个模式、一个权衡、一个生产事故案例。为构建智能体系统的人准备的每周简报。
每周一封邮件,一键退订。您的地址仅用于发送简报。