内容库 / 论文

arXiv:2608.24358 · Agent 智能体

The Handoff Tax: Continuing Non-Native Trajectories in LLM Agents

The Handoff Tax: Continuing Non-Native Trajectories in LLM Agents

类型：arxiv
标识：2608.24358
链接：https://arxiv.org/abs/2608.24358
主分类：agent
形态：method
TLDR：Coding agents perform long-running tasks spanning dozens of model calls, tool uses, and code edits. As these runs unfold, users face a practical cost-quality trade-off: escalating to a stronger model when a cheaper one struggles, or downshifting once the hard reasoning is complete. Each switch requires the receiver to continue a non-native trajectory produced by another model. We study how this handoff affects quality and cost, and how varying the trajectory information inherited by the receiver changes the outcome. Using pairs of low-cost, low-capability (LC) and high-cost, high-capability (H
待LLM分类：否
来源文件：
/inbox/tom/_candidates/2026-08-27-agent-rag-longcontext-candidates.json

相关 · 按语义匹配

仓库 AIDASLab/Awesome-LLM-Values-and-Pluralistic-Alignment

仓库 awabahmad469/composite-endpoints-audit

仓库 Sai-Kartheek-Reddy/HateMirage-ICON2026

笔记 flyP 轻量精读 · Cameron R. Wolfe《Agent Evaluation: A Detailed Guide》· v2 覆盖稿

笔记 Tom 文献雷达 · Agent RAG LongContext · 2026-07-07 14:30

仓库 sandesh20lamichhane/when-should-ids-adapt

arXiv

2608.24358

TLDR

Coding agents perform long-running tasks spanning dozens of model calls, tool uses, and code edits. As these runs unfold, users face a practical cost-quality trade-off: escalating to a stronger model when a cheaper one struggles, or downshifting once the hard reasoning is complete. Each switch requires the receiver to continue a non-native trajectory produced by another model. We study how this handoff affects quality and cost, and how varying the trajectory information inherited by the receiver changes the outcome. Using pairs of low-cost, low-capability (LC) and high-cost, high-capability (H

相关解读

arXiv ↗ 中文全文