[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-multi-agent-systems":3},{"tag":4,"articles":11},{"id":5,"name":6,"slug":7,"article_count":8,"description_zh":9,"description_en":10},"b2141a97-74e4-424c-a2c0-ae8ee60211ac","multi-agent systems","multi-agent-systems",9,"多代理系統把規劃、工具呼叫與驗證拆成多個角色，適合長任務、科學流程與自動化工作流；但也帶來記憶污染、協調成本與 token 開銷，因此遞迴式設計、Harness 架構與可演化流程成為重點。","Multi-agent systems split planning, tool use, and verification across roles, making them useful for long-running tasks, scientific workflows, and automation. The trade-offs are coordination cost, context pollution, and token use, which is why recursive designs, harness engineering, and adaptive pipelines matter.",[12,21,28,36,43],{"id":13,"slug":14,"title":15,"summary":16,"category":17,"image_url":18,"cover_image":18,"language":19,"created_at":20},"74839bf5-7853-4a42-86d6-142e1a95ba9a","llm-only-social-networks-emergent-behavior-zh","LLM 社群會長什麼樣","這篇研究把一個 Facebook 風格的社群放滿 LLM agent，觀察 14 天內 184,203 則貼文與 465,136 則留言，想看 AI 社群會冒出什麼行為。","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778050866080-cmuz.png","zh","2026-05-06T07:00:32.886126+00:00",{"id":22,"slug":23,"title":24,"summary":25,"category":17,"image_url":26,"cover_image":26,"language":19,"created_at":27},"6581dbcf-4d19-4d97-bc10-371b2e66aab1","recursive-multi-agent-systems-token-efficiency-zh","遞迴多代理系統可省 token","RecursiveMAS 把多代理協作改成潛在空間的遞迴計算，主打更少 token、更快推論，摘要宣稱平均準確率提升 8.3%。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777442638259-m4uc.png","2026-04-29T06:03:39.163322+00:00",{"id":29,"slug":30,"title":31,"summary":32,"category":33,"image_url":34,"cover_image":34,"language":19,"created_at":35},"35b17db6-a915-4a3c-87c6-733fbb7f5a31","harness-engineering-long-running-multi-agent-systems-zh","長跑型多代理系統的 Harness 設計","長跑型多代理系統最怕記憶污染。這篇看 Harness Engineering 怎麼用新 process、JSON 任務與 Claude Code，切開 Planner 和 Generator。","ai-agent","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775629509415-9j0g.png","2026-04-08T06:24:33.356723+00:00",{"id":37,"slug":38,"title":39,"summary":40,"category":17,"image_url":41,"cover_image":41,"language":19,"created_at":42},"360e622b-d834-4641-a7d3-5f60f9797146","mimosa-evolving-multi-agent-science-workflows-zh","Mimosa 讓科學代理流程自己進化","Mimosa 想解決科學代理系統太死板的問題：先自動組出工作流、執行、評分，再根據結果迭代調整。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775113458428-zcwq.png","2026-04-02T05:30:26.892841+00:00",{"id":44,"slug":45,"title":46,"summary":47,"category":48,"image_url":49,"cover_image":49,"language":19,"created_at":50},"b17b5b59-66b9-422e-9e36-99599aa614b5","ai-and-tech-trends-to-watch-in-2026-zh","2026 科技趨勢：AI 進入實戰","IBM 對 2026 的觀察很直接：多代理系統會開始進入正式環境，AI 硬體焦點從堆算力轉向效率，量子運算也要面對一次可驗證的實際考驗。重點不再是最大模型，而是能不能在企業裡穩定、便宜、可治理地跑起來。","industry","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1774517182966-phe7.png","2026-03-26T07:59:50.711356+00:00"]