Unverified50% confidenceFactExact time
在多轮迭代模式下每一轮执行都会重新传入累积的上下文，导致token消耗呈近似二次方增长
1
Sources
50%
Confidence
Long-term
Relevance
7/6/2026
First Seen
Sources
Harness Engineering: The Core Methodology for Advanced AI Programming
bilibili12点就睡的林同学7/3/2026
Related Claims
Unverified多智能体系统随着调用链路加长，错误累积效应和Token消耗会成倍增加72% similarUnverified多Agent架构中Agent间传递上下文时重复信息被反复处理，导致Token消耗和成本上升70% similarUnverifiedMost Agent frameworks adopt a full context replay strategy, requiring the entire conversation history to be passed in with each call, causing token consumption to grow quadratically with the number of loop iterations.70% similarUnverified/goal 的自主循环中，随着循环轮次增加，单轮的 Token 消耗会加速增长（除非实施上下文压缩策略）69% similarUnverifiedUltra模式的多代理架构会产生token乘数效应，整体token消耗可能是单代理模式的数倍乃至数十倍68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/113838API
curl https://kongchang.com/api/v1/knowledge/claims/113838MCP
get_claim(id=113838)