claude官方揭秘:为啥不需要多Agent架构?《Building multi-agent systems: when and how to use 》①
多agent系统:最不容易翻车的子 Agent 是谁?-《Building multi-agent systems: when and how to …》③
不同类型 Agent 怎么评估?四类 Agent 的评分器侧重点:Anthropic《Demystifying evals for AI agents》④
多智能体架构全景-主agent如何动态派生subagent:精读《How we built our multi-agent research system》②
单体Agent起步,拒绝造火箭:Anthropic《Building Effective AI Agents: Architecture Patterns》①
Agent 评估工具要不要自建?如何理性选择框架:Anthropic《Demystifying evals for AI agents》⑦
Agent Teams 正在逼近什么样的未来?一个高度自主的工程时代:《Building a C compiler with a team of parall
Agent架构选型:好架构都是逼出来的|claude《Building Effective AI Agents:Architecture Patterns》④
从原型实验出发,拆解 Claude Agent Teams 的基础架构:《Building a C compiler with a team…》①
不要构建多agent系统:精读cognition《Don't Build Multi-Agents》
多agent的提示词设计原则:精读《How we built our multi-agent research system》③
Google总结了5种Skill设计模式,让Agent稳定输出
没有评估的 Agent,注定不可规模化:Anthropic《Demystifying evals for AI agents》①
构建长效Agent——功能清单、增量开发与端到端测试:精读《Effective harnesses for long-running agents》②
为什么 Agent Memory 不是上下文工程的子集? Memory in the Age of AI Agents》③
区分agent framework与agent harness
Workflow模式之:Prompt Chaining-为什么单LLM能做的任务,偏要拆开? | 精读《Building Effective Agents》③
Anthropic构建高效agent的第一性原理:精读《Building effective agents》①
Agent评估中的评分器Grader怎么做:Anthropic《Demystifying evals for AI agents》②
多Agent:蜜糖还是砒霜?——claude《Building Effective AI Agents:Architecture Patterns and Im