Unverified50% confidenceOpinionExact time
SWE-bench的任务设计本质上是单次任务，Agent在干净环境中处理独立Issue，掩盖了跨会话记忆缺失的问题
1
Sources
50%
Confidence
Long-term
Relevance
7/17/2026
First Seen
Sources
Related Claims
VerifiedMulti-Agent architecture fundamentally avoids the attention dilution problem of a single instance by dividing tasks so that each Agent only processes context within its own area of responsibility.74% similarUnverifiedIn a multi-agent architecture, the Orchestrator is responsible for task decomposition and scheduling, delegating specific subtasks to specialized Sub-Agents for parallel or sequential execution; this is the design pattern of mainstream AI Agent frameworks such as AutoGPT, LangGraph, and OpenAI Swarm.68% similarUnverifiedSWE-agent and Devin are typical implementations of the plan-execute-reflect Agent architecture.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/544311API
curl https://kongchang.com/api/v1/knowledge/claims/544311MCP
get_claim(id=544311)