[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-nx-polygraph-ai-agent-bottlenecks-zh":3,"article-related-nx-polygraph-ai-agent-bottlenecks-zh":32,"series-industry-79038fd5-cf90-4ab7-a5e6-e5b15665b8b4":77},{"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":24,"views":28,"created_at":29,"published_at":30,"topic_cluster_id":31},"79038fd5-cf90-4ab7-a5e6-e5b15665b8b4","nx-polygraph-ai-agent-bottlenecks-zh","Nx Polygraph 盯住 AI 代理卡點","\u003Cp data-speakable=\"summary\">Nx Polygraph 用來找出 AI \u003Ca href=\"\u002Fnews\u002Fai-writes-code-teams-own-debt-zh\">寫碼\u003C\u002Fa>代理在 monorepo 裡卡住的地方。\u003C\u002Fp>\u003Cp>這份清單看完，你可以更快判斷 4 件事：代理到底卡在上下文、依賴路徑、驗證，還是工作流設計。對已經在多語言、多套件倉庫裡試 AI 寫碼的團隊來說，這比單純換\u003Ca href=\"\u002Fnews\u002Fcodex-third-party-model-integration-guide-zh\">模型\u003C\u002Fa>更實際。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>項目\u003C\u002Fth>\u003Cth>看什麼\u003C\u002Fth>\u003Cth>適合誰\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>Polygraph\u003C\u002Ftd>\u003Ctd>代理在 monorepo 的瓶頸位置\u003C\u002Ftd>\u003Ctd>大型代碼庫團隊\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>\u003Ca href=\"https:\u002F\u002Fnx.dev\u002F\">Nx\u003C\u002Fa>\u003C\u002Ftd>\u003Ctd>任務圖與工作區編排\u003C\u002Ftd>\u003Ctd>平台工程與多套件團隊\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>\u003Ca href=\"https:\u002F\u002Fopentelemetry.io\u002F\">OpenTelemetry\u003C\u002Fa>\u003C\u002Ftd>\u003Ctd>追蹤與可觀測性訊號\u003C\u002Ftd>\u003Ctd>想量化代理行為的團隊\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Agent harnesses\u003C\u002Ftd>\u003Ctd>執行與驗證迴圈\u003C\u002Ftd>\u003Ctd>在做代理工作流的人\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Synthetic monorepos\u003C\u002Ftd>\u003Ctd>可控的測試倉庫\u003C\u002Ftd>\u003Ctd>要先做實驗再上線的團隊\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>1. Polygraph 先把卡點畫出來\u003C\u002Fh2>\u003Cp>Polygraph 的重點不是再講一次 AI 會寫程式，而是把代理進到真實倉庫後，究竟在哪一步變慢、迷路或失去上下文，直接攤開來看。它把問題從「模型不夠強」改寫成「倉庫結構哪裡讓代理不好做事」。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782405177271-be2u.png\" alt=\"Nx Polygraph 盯住 AI 代理卡點\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>這對團隊很重要，因為代理失敗常常不是單點錯誤，而是資料夾層級、依賴關係、檔案\u003Ca href=\"\u002Fnews\u002Fvllm-sglang-vmlx-local-llm-runtimes-zh\">選擇\u003C\u002Fa>和驗證流程一起造成的。\u003C\u002Fp>\u003Cul>\u003Cli>標出代理最常停住的區域\u003C\u002Fli>\u003Cli>找出依賴密集、上下文切換多的路徑\u003C\u002Fli>\u003Cli>看驗證在哪些步驟失效\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>2. Nx 把 monorepo 關係變成可讀圖譜\u003C\u002Fh2>\u003Cp>Polygraph 背後的基礎是 \u003Ca href=\"https:\u002F\u002Fnx.dev\u002F\">Nx\u003C\u002Fa>，它本來就擅長理解 monorepo 的任務圖、套件關係和工作區編排。這讓它不只是看程式碼，而是看改動會牽動哪些 app、library 和 test。\u003C\u002Fp>\u003Cp>如果你的團隊已經在管理很多套件，Nx 的價值在於把隱性的變更成本變成可討論的規格，讓你知道代理到底是卡在能力，還是卡在倉庫太複雜。\u003C\u002Fp>\u003Cul>\u003Cli>任務圖能顯示變更波及範圍\u003C\u002Fli>\u003Cli>工作區依賴關係可直接追蹤\u003C\u002Fli>\u003Cli>build、test、lint 可一起編排\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>3. OpenTelemetry 讓代理行為可追蹤\u003C\u002Fh2>\u003Cp>文章真正指向的下一步，是把代理行為當成可以觀測的系統。這時 \u003Ca href=\"https:\u002F\u002Fopentelemetry.io\u002F\">OpenTelemetry\u003C\u002Fa> 這類工具就很有價值，因為它能把代理走過哪些步驟、在哪裡耗時、哪個檢查失敗記錄下來。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782405175090-t6gx.png\" alt=\"Nx Polygraph 盯住 AI 代理卡點\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>對工程團隊來說，這比事後猜測可靠得多。當代理在某個 repo pattern 上反覆失敗，trace 和 span 可以直接告訴你問題出在檔案探索、依賴掃描，還是測試驗證。\u003C\u002Fp>\u003Ccode>trace: agent_start -> file_discovery -> dependency_scan -> edit -> test -> verify\u003C\u002Fcode>\u003Ch2>4. Agent harnesses 決定代理能不能收斂\u003C\u002Fh2>\u003Cp>單靠提示詞，代理很難在複雜倉庫裡穩定完成多步驟任務。\u003Ca href=\"\u002Ftag\u002Fagent\">Agent\u003C\u002Fa> harnesses 指的是包住代理的執行規則、測試、檢查和回饋機制，目的是讓它在真實系統裡不亂跑。\u003C\u002Fp>\u003Cp>這也是這篇內容最實用的地方：瓶頸正在從「生成能力」轉向「受控執行」。如果沒有 harness，再強的模型也可能在大型代碼庫裡做出看似合理、實際不可驗證的改動。\u003C\u002Fp>\u003Cul>\u003Cli>每次改動後自動跑測試\u003C\u002Fli>\u003Cli>先查檔案層，再查倉庫層依賴\u003C\u002Fli>\u003Cli>驗證失敗就阻擋合併\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>5. Synthetic monorepos 適合先做實驗\u003C\u002Fh2>\u003Cp>Polygraph 也暗示了一種更安全的測試方式：先用 synthetic monorepos 做實驗。這種控制過的倉庫結構，能讓團隊觀察代理在哪些語言混搭、依賴圖或目錄形狀上最容易出問題。\u003C\u002Fp>\u003Cp>對平台工程和開發工具團隊來說，這像一個可重複的實驗台。你可以先比較不同 repo 形狀下的代理表現，再決定要不要把流程推進到正式代碼庫。\u003C\u002Fp>\u003Cul>\u003Cli>倉庫結構可控\u003C\u002Fli>\u003Cli>測試結果可重跑\u003C\u002Fli>\u003Cli>方便比較不同工作流\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>怎麼挑\u003C\u002Fh2>\u003Cp>如果你已經在大型 monorepo 裡試 AI 寫碼，先看 \u003Ca href=\"https:\u002F\u002Fnx.dev\u002F\">Nx\u003C\u002Fa> 和 Polygraph，因為它們最直接回答「卡在哪」。如果你更在意量化問題，\u003Ca href=\"https:\u002F\u002Fopentelemetry.io\u002F\">OpenTelemetry\u003C\u002Fa> 式追蹤更適合先上。若你正在設計代理流程，先補 harness，再用 synthetic monorepos 做壓力測試。\u003C\u002Fp>\u003Cp>最實際的順序是：先畫出倉庫與依賴，再加驗證，最後才談擴大代理使用範圍。這樣你比較能判斷，問題是模型、環境，還是流程本身。\u003C\u002Fp>","4 個重點看 Nx Polygraph 如何找出 AI 寫碼代理在 monorepo 變慢的原因，並判斷該先補觀測、驗證還是工作流。","thenewstack.io","https:\u002F\u002Fthenewstack.io\u002Fnx-polygraph-synthetic-monorepo-agents\u002F",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782405177271-be2u.png","industry","zh","251c627e-83e7-43b2-9163-0bd3d8c5d539",[17,18,19,20,21,22,23],"Nx Polygraph","AI coding agents","monorepo","OpenTelemetry","agent harnesses","synthetic monorepos","developer tools",[25,26,27],"Polygraph 的核心價值是把代理在 monorepo 的卡點具體化，而不是只歸咎模型能力。","Nx 擅長把任務圖與依賴關係攤開，適合大型工作區先做結構分析。","OpenTelemetry、harness 和 synthetic monorepos 分別對應觀測、驗證與前置測試。",0,"2026-06-25T16:32:24.073843+00:00","2026-06-25T16:32:24.066+00:00","caa87b65-9bbc-46fe-bba8-4f4158dd2d8b",{"tags":33,"relatedLang":36,"relatedPosts":40},[34],{"name":18,"slug":35},"ai-coding-agents",{"id":15,"slug":37,"title":38,"language":39},"nx-polygraph-ai-agent-bottlenecks-en","Nx Polygraph targets AI agent bottlenecks","en",[41,47,53,59,65,71],{"id":42,"slug":43,"title":44,"cover_image":45,"image_url":45,"created_at":46,"category":13},"47ecd595-8782-403d-b091-64e0fec5e176","ai-companies-must-earn-trust-on-jobs-zh","AI 公司要贏，先證明自己不會掏空工作","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782416873244-86n5.png","2026-06-25T19:47:25.696056+00:00",{"id":48,"slug":49,"title":50,"cover_image":51,"image_url":51,"created_at":52,"category":13},"1709aaa0-6b69-402d-954c-9b367d30a5f0","microsoft-ai-education-report-adoption-support-zh","微軟：AI 已成教室日常","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782415075444-xfph.png","2026-06-25T19:17:27.883368+00:00",{"id":54,"slug":55,"title":56,"cover_image":57,"image_url":57,"created_at":58,"category":13},"85621665-982c-44b8-aa53-9d7352e51dac","ruffle-keeps-flash-games-playable-zh","Ruffle 讓 Flash 遊戲續命的 5 個關鍵","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782414176328-r7r3.png","2026-06-25T19:02:27.387704+00:00",{"id":60,"slug":61,"title":62,"cover_image":63,"image_url":63,"created_at":64,"category":13},"155a5305-f45e-4ea2-8661-7d0a4e613de4","jalapeno-turns-openai-into-chip-designer-zh","Jalapeño 讓 OpenAI 變晶片公司","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782407899341-wl80.png","2026-06-25T17:17:56.450808+00:00",{"id":66,"slug":67,"title":68,"cover_image":69,"image_url":69,"created_at":70,"category":13},"7d73898b-ddb7-4326-a8a5-94d1afb5311c","anthropic-overseas-data-center-push-right-move-zh","Anthropic 海外資料中心擴張是對的：算力已是全球戰略資產","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782406974421-rxl5.png","2026-06-25T17:02:28.557827+00:00",{"id":72,"slug":73,"title":74,"cover_image":75,"image_url":75,"created_at":76,"category":13},"50ca9d65-eecc-41af-aa36-aa270522fde4","ai-writes-code-teams-own-debt-zh","AI 寫碼快，債還是團隊背","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782404291011-yahb.png","2026-06-25T16:17:35.300372+00:00",[78,83,88,93,98,103,108,113,118,123],{"id":79,"slug":80,"title":81,"created_at":82},"ee073da7-28b3-4752-a319-5a501459fb87","ai-in-2026-what-actually-matters-now-zh","2026 AI 真正重要的事","2026-03-26T07:09:12.008134+00:00",{"id":84,"slug":85,"title":86,"created_at":87},"83bd1795-8548-44c9-9a7e-de50a0923f71","trump-ai-framework-power-speech-state-preemption-zh","川普 AI 框架瞄準電力、言論與州權","2026-03-26T07:12:18.695466+00:00",{"id":89,"slug":90,"title":91,"created_at":92},"ea6be18b-c903-4e54-97b7-5f7447a612e0","nvidia-gtc-2026-big-ai-announcements-zh","NVIDIA GTC 2026 重點拆解","2026-03-26T07:14:26.62638+00:00",{"id":94,"slug":95,"title":96,"created_at":97},"4bcec76f-4c36-4daa-909f-54cd702f7c93","claude-users-spreading-out-and-getting-better-zh","Claude 用戶更分散，也更會用","2026-03-26T07:22:52.325888+00:00",{"id":99,"slug":100,"title":101,"created_at":102},"bd903b15-2473-4178-9789-b7557816e535","openclaw-raises-hard-question-for-ai-models-zh","OpenClaw 逼問 AI 模型價值","2026-03-26T07:24:54.707486+00:00",{"id":104,"slug":105,"title":106,"created_at":107},"eeac6b9e-ad9d-4831-8eec-8bba3f9bca6a","gap-google-gemini-checkout-fashion-search-zh","Gap 把結帳搬進 Gemini","2026-03-26T07:28:23.937768+00:00",{"id":109,"slug":110,"title":111,"created_at":112},"0740e53f-605d-4d57-8601-c10beb126f3c","google-pushes-gemini-transition-to-march-2026-zh","Google 把 Gemini 轉換延到 2026 年 3…","2026-03-26T07:30:12.825269+00:00",{"id":114,"slug":115,"title":116,"created_at":117},"e660d801-2421-4529-8fa9-86b82b066990","metas-llama-4-benchmark-scandal-gets-worse-zh","Meta Llama 4 分數風波又擴大","2026-03-26T07:34:21.156421+00:00",{"id":119,"slug":120,"title":121,"created_at":122},"183f9e7c-e143-40bb-a6d5-67ba84a3a8bc","accenture-mistral-ai-sovereign-enterprise-deal-zh","Accenture 攜手 Mistral AI 賣主權 AI","2026-03-26T07:38:14.818906+00:00",{"id":124,"slug":125,"title":126,"created_at":127},"191d9b1b-768a-478c-978c-dd7431a38149","mistral-ai-faces-its-hardest-year-yet-zh","Mistral AI 迎來最硬的一年","2026-03-26T07:40:23.716374+00:00"]