[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-why-prompt-engineering-is-dead-for-ai-agents-zh":3,"article-related-why-prompt-engineering-is-dead-for-ai-agents-zh":30,"series-ai-agent-8a126231-b5e1-4c89-a338-73d1d73148ff":78},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":23,"views":27,"created_at":28,"published_at":29,"topic_cluster_id":11},"8a126231-b5e1-4c89-a338-73d1d73148ff","why-prompt-engineering-is-dead-for-ai-agents-zh","為什麼 AI Agent 時代，Prompt Engineering 已經死了","\u003Cp data-speakable=\"summary\">\u003Ca href=\"\u002Ftag\u002Fai-agent\">AI Agent\u003C\u002Fa> 的可靠性不靠更會寫提示詞，而靠上下文管理：選對資訊、控制長度、維持狀態。\u003C\u002Fp>\u003Cp>\u003Ca href=\"\u002Ftag\u002Fprompt-engineering\">Prompt engineering\u003C\u002Fa> 對 \u003Ca href=\"\u002Ftag\u002Fai-agents\">AI agents\u003C\u002Fa> 來說已經不是主戰場，因為真正決定成敗的不是措辭，而是上下文管理。Chroma 在 2025 年 7 月針對 18 個模型的研究，包含 C\u003Ca href=\"\u002Fnews\u002Fbest-solana-api-providers-for-devs-and-ai-agents-zh\">la\u003C\u002Fa>ude 4、GPT-4.1 與 \u003Ca href=\"\u002Ftag\u002Fgemini\">Gemini\u003C\u002Fa> 2.5，顯示 context 一拉長，表現就下滑，而且連簡單的檢索任務都會受影響。這代表問題不是模型「突然變笨」，而是它在錯的時間看到太多錯的資訊。\u003C\u002Fp>\u003Ch2>第一個論點\u003C\u002Fh2>\u003Cp>對 AI \u003Ca href=\"\u002Ftag\u002Fagents\">agents\u003C\u002Fa> 而言，prompt 只是包裝，真正的產品是整包上下文：指令、工具、記憶、檢索文件與任務狀態。只要這包內容雜亂，agent 就會失手，即使 prompt 寫得再工整也一樣。很多團隊把心力花在字句修飾，卻忽略了系統輸入本身才是主要變因。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778184658982-xgnl.png\" alt=\"為什麼 AI Agent 時代，Prompt Engineering 已經死了\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>Chroma 的結果之所以重要，是因為它說明退化不是線性的。context 變長不等於資訊變多、答案就更好；相反地，模型的注意力會被競爭中的 token 分散，連原本很簡單的 retrieval 任務也會變差。這不是文案問題，而是系統設計問題。\u003C\u002Fp>\u003Ch2>第二個論點\u003C\u002Fh2>\u003Cp>真正能交付穩定 agent 的團隊，不會只靠一段萬用指令。他們做的是管線：先選資料、再排序、再壓縮、再更新 context，最後才交給模型回應。這就是為什麼 context engineering 比 prompt engineering 更接近實戰：它把模型視為系統的最後一步，而不是整個系統本身。\u003C\u002Fp>\u003Cp>以客服自動化為例，把完整工單歷史、產品文件、政策全文與前一次工具輸出一次塞進去，常常比不上只給它「當前問題、相關政策片段、上一輪動作摘要」的精簡版本。這裡的關鍵不是少，而是準。對 agent 來說，正確的 300 個 token，常常勝過錯置的 3,000 個 token。\u003C\u002Fp>\u003Ch2>反方可能怎麼說\u003C\u002Fh2>\u003Cp>prompt engineering 的支持者不是完全錯。對於範圍很窄的工作流，一段精準的 system prompt 確實能大幅改善行為；清楚的格式、明確的限制、挑對的 few-shot 範例，都有實際效果。很多今天上線的 pr\u003Ca href=\"\u002Fnews\u002Fsoderbergh-ai-lennon-doc-meta-zh\">od\u003C\u002Fa>uction agent，仍然離不開這些基本功。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778184648786-twoz.png\" alt=\"為什麼 AI Agent 時代，Prompt Engineering 已經死了\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>另一個合理顧慮是，context engineering 聽起來很對，但工程成本也很高。若每個 agent 都要加 retrieval、memory m\u003Ca href=\"\u002Fnews\u002Fwhy-solana-developer-hiring-should-stop-treating-skills-as-s-zh\">ana\u003C\u002Fa>ger、summarizer 和 ranking service，團隊可能還沒做出產品，就先被基礎設施拖住。對小產品來說，強 prompt 仍然是最快的起步方式。\u003C\u002Fp>\u003Cp>但這些都是限制，不是反駁。prompt quality 是必要條件，卻不是充分條件；一旦 agent 要跨多步驟、多工具、或多個真實資料來源行動，核心問題就會變成選擇與控制，而不是修辭。Chroma 的研究也把邊界講得很清楚：context 越長，表現越差，所以真正的解法是工程化 context window，而不是繼續堆字。\u003C\u002Fp>\u003Ch2>你能做什麼\u003C\u002Fh2>\u003Cp>如果你是工程師，別再把 prompt 當成主要抽象層，改做 context pipeline：少抓資料、提高排序品質、積極摘要、並且只保留當前步驟真的需要的記憶。如果你是 PM 或創辦人，請用真實 context 負載下的任務成功率評估 agent，不要只看 demo 好不好看。問題從來不是「提示詞寫得夠不夠聰明」，而是「agent 有沒有在正確的時間拿到正確的證據」。\u003C\u002Fp>","AI Agent 的關鍵不在於把提示詞寫得更漂亮，而在於把上下文選對、排好、壓縮好；context engineering 才是可靠性的核心。","pub.towardsai.net","https:\u002F\u002Fpub.towardsai.net\u002Fprompt-engineering-is-dead-for-ai-agents-here-is-what-actually-works-541ceda072de?gi=afda2de8cc66",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778184658982-xgnl.png","ai-agent","zh","33341479-794d-446c-aa4e-7b8aa61c72d0",[17,18,19,20,21,22],"AI agents","prompt engineering","context engineering","retrieval","context window","LLM reliability",[24,25,26],"AI agents 的主要瓶頸是上下文管理，不是提示詞措辭。","context 變長會帶來注意力分散與表現退化，尤其在檢索任務上。","可靠的 agent 需要 context pipeline，而不是更長更花俏的 prompt。",5,"2026-05-07T20:10:24.866536+00:00","2026-05-07T20:10:24.66+00:00",{"tags":31,"relatedLang":11,"relatedPosts":41},[32,34,35,37,39],{"name":18,"slug":33},"prompt-engineering",{"name":20,"slug":20},{"name":19,"slug":36},"context-engineering",{"name":21,"slug":38},"context-window",{"name":17,"slug":40},"ai-agents",[42,48,54,60,66,72],{"id":43,"slug":44,"title":45,"cover_image":46,"image_url":46,"created_at":47,"category":13},"83c2f8f6-3710-466e-b52c-473b811f0535","how-to-set-up-openclaw-safely-zh","如何安全架設 OpenClaw","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780549368665-1t2l.png","2026-06-04T05:02:21.26625+00:00",{"id":49,"slug":50,"title":51,"cover_image":52,"image_url":52,"created_at":53,"category":13},"0ba5b1a8-82c5-464a-bea5-9a2c8730da74","aws-devops-agent-turns-incident-chaos-into-triage-zh","AWS DevOps Agent 把事故排查變成三步","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780466689960-g1sv.png","2026-06-03T06:03:14.154923+00:00",{"id":55,"slug":56,"title":57,"cover_image":58,"image_url":58,"created_at":59,"category":13},"841eac88-b0f0-4a4c-9e1e-efc3b5c16281","kimi-k26-live-300-agent-workflows-zh","Kimi K2.6 上線：300 代理工作流","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780430574285-hqpn.png","2026-06-02T20:02:24.972179+00:00",{"id":61,"slug":62,"title":63,"cover_image":64,"image_url":64,"created_at":65,"category":13},"f0411957-bcdb-42d9-a267-3e90ae7d9cb1","how-to-take-a-sabbatical-at-openai-zh","怎麼申請 OpenAI sabbatical","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780398216422-8fi7.png","2026-06-02T11:02:25.74372+00:00",{"id":67,"slug":68,"title":69,"cover_image":70,"image_url":70,"created_at":71,"category":13},"37a5e429-4235-439c-9b05-bb377085462c","8-steps-build-production-rag-with-langchain-zh","8 步驟打造可上線的 LangChain RAG","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780178597493-4hz7.png","2026-05-30T22:02:48.14022+00:00",{"id":73,"slug":74,"title":75,"cover_image":76,"image_url":76,"created_at":77,"category":13},"e73c041b-852b-44c3-85aa-0f1e2e5848e3","ai-agents-hit-chaos-mode-claude-code-openclaw-zh","Claude Code＋OpenClaw 讓 AI 代理失控升溫","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780160576178-yqcs.png","2026-05-30T17:02:25.725767+00:00",[79,84,89,94,99,104,109,114,119,124],{"id":80,"slug":81,"title":82,"created_at":83},"4ae1e197-1d3d-4233-8733-eafe9cb6438b","claude-now-uses-your-pc-to-finish-tasks-zh","Claude 開始幫你操作電腦","2026-03-26T07:20:48.457387+00:00",{"id":85,"slug":86,"title":87,"created_at":88},"5bede67f-e21c-413d-9ab8-54a3c3d26227","googles-2026-ai-agent-report-decoded-zh","Google 2026 AI Agent 報告解讀","2026-03-26T11:15:22.651956+00:00",{"id":90,"slug":91,"title":92,"created_at":93},"2987d097-563f-46c7-b76f-b558d8ef7c2b","kimi-k25-review-stronger-still-not-legend-zh","Kimi K2.5 評測：更強，但還不是神作","2026-03-27T07:15:55.277513+00:00",{"id":95,"slug":96,"title":97,"created_at":98},"95c9053b-e3f4-4cb5-aace-5c54f4c9e044","claude-code-controls-mac-desktop-zh","Claude Code 也能操控 Mac 了","2026-03-28T03:01:58.58121+00:00",{"id":100,"slug":101,"title":102,"created_at":103},"dc58e153-e3a8-4c06-9b96-1aa64eabbf5f","cloudflare-100x-faster-ai-agent-sandbox-zh","Cloudflare 的 AI 沙箱跑超快","2026-03-28T03:09:44.142236+00:00",{"id":105,"slug":106,"title":107,"created_at":108},"1c8afc56-253f-47a2-979f-1065ff072f2a","openai-backs-isara-agent-swarm-bet-zh","OpenAI 挺 Isara 的 agent swarm …","2026-03-28T03:15:27.513155+00:00",{"id":110,"slug":111,"title":112,"created_at":113},"7379b422-576e-45df-ad5a-d57a0d9dd467","openai-plan-automated-ai-researcher-zh","OpenAI 想做自動化 AI 研究員","2026-03-28T03:17:42.090548+00:00",{"id":115,"slug":116,"title":117,"created_at":118},"48c9889e-86df-450b-a356-e4a4b7c83c5b","harness-engineering-ai-agent-reliability-2026-zh","駕馭工程：從「馬具」到「作業系統」，AI Agent 可靠性的終極密碼","2026-03-31T06:42:53.556721+00:00",{"id":120,"slug":121,"title":122,"created_at":123},"96d8e8c8-1edd-475d-9145-b1e7a1b02b65","mcp-explained-from-prompts-to-production-zh","MCP 怎麼把提示詞變工作流","2026-04-01T09:24:39.321274+00:00",{"id":125,"slug":126,"title":127,"created_at":128},"f2ca7720-b471-4ce5-9336-2a9ac2a876fd","amazon-bedrock-agents-multi-agent-workflows-zh","Amazon Bedrock Agents 進入多代理工作流","2026-04-01T09:30:29.945429+00:00"]