https://chandlernguyen.com/zh-HK/ AI builder, lifelong learner, product creator. 18-year advertising veteran learning to code at 40 and building AI products. zh-hk Tue, 20 Oct 2026 04:30:54 GMT https://chandlernguyen.com/images/chandler_nguyen_profile.webp https://chandlernguyen.com/zh-HK/ contact@chandlernguyen.com (Chandler Nguyen) contact@chandlernguyen.com (Chandler Nguyen) Copyright 2026 Chandler Nguyen 60 https://chandlernguyen.com/zh-HK/blog/2026/10/20/knora-1-2-is-on-the-app-store/ https://chandlernguyen.com/zh-HK/blog/2026/10/20/knora-1-2-is-on-the-app-store/ Tue, 20 Oct 2026 00:00:00 GMT Knora 1.2 已經推出：十五把英語聲、四種口音、刪咗嘅風格軸，同埋修好嘅兩個音訊缺陷。笑話而家活喺把聲入面。 AI Product Build In Public contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/10/18/the-voice-is-the-joke/ https://chandlernguyen.com/zh-HK/blog/2026/10/18/the-voice-is-the-joke/ Sun, 18 Oct 2026 00:00:00 GMT 我重建咗 Knora 嘅英文聲陣容，刪咗風格軸，仲發現咗兩個連測試都話冇問題嘅音訊 bug。更新已經上架——而個教訓，其實唔關喜劇事。 AI Product Build In Public contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/10/16/my-mac-died-i-rebuilt-it-with-chrome-and-deepseek/ https://chandlernguyen.com/zh-HK/blog/2026/10/16/my-mac-died-i-rebuilt-it-with-chrome-and-deepseek/ Fri, 16 Oct 2026 00:00:00 GMT 我部 Mac 死咗，所以我剷咗個硬碟，淨係靠 Chrome 同一個 agent harness 由零開始。而家部機比之前多咗大約 60 GB 可用空間，而我都唔再以應用程式嚟思考。 AI Build in Public contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/10/06/what-is-ai-native-media-operations/ https://chandlernguyen.com/zh-HK/blog/2026/10/06/what-is-ai-native-media-operations/ Tue, 06 Oct 2026 00:00:00 GMT AI 原生媒體營運，係廣告同 marketing 團隊喺 AI 承擔大部分執行時，點樣運轉整個 campaign 生命週期——research、planning、activation、reporting、measurement 嘅方式。呢篇會講清楚 佢同「我哋有用 AI 工具」有咩分別、生命週期每個階段點變，以及人仲要負責咩。 AI contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/09/30/ai-native-media-ops-strategy-guide/ https://chandlernguyen.com/zh-HK/blog/2026/09/30/ai-native-media-ops-strategy-guide/ Wed, 30 Sep 2026 00:00:00 GMT AI 原生媒體營運嘅策略，多半係一個排序決定：先搞好 operating model 同由 research 到 measurement 嘅循環，先至去買更多工具。呢份指南梳理 strategy lane——「原生」係咩意思、 點解將 AI 疊喺舊 workflow 上面會失敗、research 同 planning 點變、以及 planner 會唔會生存—— 並指向每條問題更深入嗰篇。 AI contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/09/24/i-built-an-app-with-almost-no-ui/ https://chandlernguyen.com/zh-HK/blog/2026/09/24/i-built-an-app-with-almost-no-ui/ Thu, 24 Sep 2026 00:00:00 GMT 關於我個喜劇 App 最好嘅一次評價，嚟自一張坐咗四個朋友、擺住一輪酒嘅枱——而我第一個版本冇過到呢關。所以我先刪咗笑話呢個格式，然後再刪咗介面。下面講吓一個只靠一個手勢嘅 App 會逼出啲咩、點解 75 秒呢個上限其實係一張枱可以坐幾耐嘅約束，同埋對於日常份工係注意力指標嘅人嚟講，最唔舒服嘅嗰部分。 AI Product Build In Public contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/09/22/ai-native-media-ops-execution-guide/ https://chandlernguyen.com/zh-HK/blog/2026/09/22/ai-native-media-ops-execution-guide/ Tue, 22 Sep 2026 00:00:00 GMT Execution 係 AI 原生媒體營運成敗所在：AI 究竟承擔得到幾多工作、production、reporting 同 measurement 點變、咩一定要被 govern、以及點樣喺 90 日內 rollout。呢份指南梳理執行 lane， 並指向每條問題更深入嗰篇。 AI contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/09/16/ai-native-media-ops-agency-model-guide/ https://chandlernguyen.com/zh-HK/blog/2026/09/16/ai-native-media-ops-agency-model-guide/ Wed, 16 Sep 2026 00:00:00 GMT 對 agency 嚟講，AI 嘅問題唔係要唔要採用啲工具，而係當每個競爭者都有咗呢啲工具之後你賣咩。 呢份指南梳理 agency-model lane——該保留、改變、停止賣咩，AI 原生行家喺價錢同人頭上點比、 點樣為工作定價、以及持久嘅護城河喺邊——並指向每條問題更深入嗰篇。 AI contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/09/14/i-gave-my-9-agent-marketing-platform-away-as-open-source/ https://chandlernguyen.com/zh-HK/blog/2026/09/14/i-gave-my-9-agent-marketing-platform-away-as-open-source/ Mon, 14 Sep 2026 00:00:00 GMT STRAŦUM v1 而家以 MIT 授權公開咗。九個 Agent、兩個 Postgres schema、十種語言——仲有一張出街嗰陣冇開 Row Level Security 嘅表，任何揸住喺瀏覽器度載入嗰條 key 嘅人都掂得到。以下係我送咗啲咩出去、有邊啲嘢我唔會再照同一個做法做，同埋點解你大可以 fork 佢、整一個你自己嘅版本。 AI Product Build In Public contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/09/09/ai-native-vs-ai-assisted-marketing-teams/ https://chandlernguyen.com/zh-HK/blog/2026/09/09/ai-native-vs-ai-assisted-marketing-teams/ Wed, 09 Sep 2026 00:00:00 GMT AI 原生意思係 model 產出第一 pass、人 review 同決定；AI 輔助意思係人產出工作、偶爾伸手去 拎 model 加速其中一部分。同樣嘅工具，唔同嘅重心。下面係個區分、一張對比表，同一個簡短嘅 診斷，幫你判斷團隊實際屬於邊一種。 AI Marketing contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/09/02/the-parade-problem-ai-media-workflows/ https://chandlernguyen.com/zh-HK/blog/2026/09/02/the-parade-problem-ai-media-workflows/ Wed, 02 Sep 2026 00:00:00 GMT 巡遊問題，係當你將 AI 外掛到一個從來冇重建過嘅 workflow 上面時發生嘅事：第一步快咗， 身後嘅樽頸照舊，收工時間一啲都冇早到。呢篇講佢點解會發生、大多數 AI pilot 之後仍然生存嘅 四個樽頸，以及點樣判斷你係喺幫架花車加速定係喺重新設計條路線。 AI Marketing contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/08/27/how-much-of-media-work-ai-can-do-the-75-80-principle/ https://chandlernguyen.com/zh-HK/blog/2026/08/27/how-much-of-media-work-ai-can-do-the-75-80-principle/ Thu, 27 Aug 2026 00:00:00 GMT 喺我觀察過嘅 media task 入面，AI 大約做 75–80% 嘅工作——喺 production、拼裝同搵規律上 最多，喺 negotiation、關係同最終判斷上最少。佢係用嚟搵出你必須留返俾人嗰 20% 嘅 heuristic， 唔係 benchmark。下面係佢嘅形狀，以及點樣 audit 你自己嘅 task。 AI Paid Media contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/08/21/how-ai-changes-media-research-personas-and-data/ https://chandlernguyen.com/zh-HK/blog/2026/08/21/how-ai-changes-media-research-personas-and-data/ Fri, 21 Aug 2026 00:00:00 GMT AI 主要唔係令 media research 快啲——而係令佢持續。Persona 唔再係一份起草一次嘅文件， 而變成一個會運轉嘅 model；持久嘅優勢由公開 research（而家對所有人都平）轉移到你專有嘅 first-party data 同 memory。呢篇講噉會改變咩，以及點解 governance 係前提而唔係文書。 AI Marketing contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/08/14/ai-media-planning-and-measurement-design/ https://chandlernguyen.com/zh-HK/blog/2026/08/14/ai-media-planning-and-measurement-design/ Fri, 14 Aug 2026 00:00:00 GMT AI media planning 意思係 model 對住同一份 brief 同 budget 草擬幾個結構化方案，planner 喺幾個講得通嘅 plan 度揀同 refine，而唔係拼裝一份 slide。團隊規劃不足嘅係 measurement： 如果個 plan 被生成、measurement 卻唔變，你只係更快咁產出多啲選項，喺當中學唔到多啲。 將兩者一齊設計。 AI Paid Media contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/08/11/deepseek-v4-pro-two-months-cost-savings/ https://chandlernguyen.com/zh-HK/blog/2026/08/11/deepseek-v4-pro-two-months-cost-savings/ Tue, 11 Aug 2026 00:00:00 GMT 六月：$57.01。七月：$33.09。透過 OpenCode 轉用 DeepSeek V4 Pro 兩個月之後，我嘅 AI 成本跌咗 75%。Commit 歷史證明我出咗更多嘢，冇少到——786 個 commits，橫跨 4 個產品。 呢度係完整嘅 breakdown。 AI Product Build In Public contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/08/07/activation-when-ai-drives-media-production/ https://chandlernguyen.com/zh-HK/blog/2026/08/07/activation-when-ai-drives-media-production/ Fri, 07 Aug 2026 00:00:00 GMT 當 AI 推動 production，activation 由製作 asset 轉向監督一條 pipeline：variant、copy 同 adaptation 被成量產出，review 成為真正嘅 constraint。Operator 嘅工作變成設定 guardrail、 定義 review standard，以及知道幾時叫停部機器。 Paid Media AI contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/07/31/reporting-as-intelligence-not-assembly/ https://chandlernguyen.com/zh-HK/blog/2026/07/31/reporting-as-intelligence-not-assembly/ Fri, 31 Jul 2026 00:00:00 GMT 當 AI 拼裝份 report——拉取、關聯、排版、敘述——analyst 嘅價值就由做 chart 轉向讀 chart。 Reporting 變成 intelligence，而唔係拼裝。呢篇講部機器攞走咩、咩留返俾人，以及點樣重建 一份 report，令會議時間花喺解讀上。 AI Analytics contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/07/25/90-day-ai-native-media-ops-rollout/ https://chandlernguyen.com/zh-HK/blog/2026/07/25/90-day-ai-native-media-ops-rollout/ Sat, 25 Jul 2026 00:00:00 GMT AI 原生媒體團隊嘅 90 日 rollout：第 1–30 日整好個 constraint、令一份 brief 變得 machine-readable，第 31–60 日端到端跑嗰個階段並建立 review standard，第 61–90 日 measure 同決定擴唔擴展。一個階段、一位 owner、擴展之前先有證據。 AI Strategy contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/07/19/agency-design-decision-framework/ https://chandlernguyen.com/zh-HK/blog/2026/07/19/agency-design-decision-framework/ Sun, 19 Jul 2026 00:00:00 GMT Agency design 決策框架將每一項服務歸入保留、改變或停止賣。個軸係價值坐喺邊——拼裝定判斷—— 以及 AI 對交付經濟學做咗咩。呢篇講點樣分類你所賣嘅嘢，以及佢對周圍嘅人同定價意味住咩。 AI Marketing contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/07/16/i-evaluated-ai-podcast-pipeline-six-fixes-what-improved/ https://chandlernguyen.com/zh-HK/blog/2026/07/16/i-evaluated-ai-podcast-pipeline-six-fixes-what-improved/ Thu, 16 Jul 2026 00:00:00 GMT 我構建咗一個評估框架，喺 7 種語言嘅 40 個真實集數上為我嘅 AI 播客管線評分。 審查步驟反而令腳本變差（3.34 → 2.91）。所有引用 URL 都睇落係假嘅（得分：1.28）。 中文係最弱嘅語言（2.88）。六項修復、一日同一個新模型之後——呢度係 前後對比數字同我做嘅每一個更改。 AI Product contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/07/11/will-ai-replace-media-planners/ https://chandlernguyen.com/zh-HK/blog/2026/07/11/will-ai-replace-media-planners/ Sat, 11 Jul 2026 00:00:00 GMT 唔會——AI 唔取代 media planner。佢取代 planning 份工入面嘅拼裝，留低判斷。草擬 plan、 分配 channel budget、產出每週視圖而家大體可以自動化。決定邊個 constraint 重要——並據此 押上一條建議——就唔可以。呢個角色係集中，而非消失。 AI Marketing contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/07/07/how-i-write-system-prompts-that-get-enforced/ https://chandlernguyen.com/zh-HK/blog/2026/07/07/how-i-write-system-prompts-that-get-enforced/ Tue, 07 Jul 2026 00:00:00 GMT 我花咗幾日時間為AI podcast主持編寫系統提示。佢哋逐字閱讀咗——然後幾乎全部無視晒。解決方案唔係更好嘅措辭——而係明白LLM處理規則嘅方式，取決於規則喺pipeline中嘅位置。 AI Product contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/07/04/ai-native-agency-vs-traditional-agency/ https://chandlernguyen.com/zh-HK/blog/2026/07/04/ai-native-agency-vs-traditional-agency/ Sat, 04 Jul 2026 00:00:00 GMT AI 原生 agency 同傳統 agency 唔係同一門生意配一份更大嘅工具 budget。個分別在於「默認」 坐喺邊、工作點樣配人、以及客戶實際喺買咩。傳統賣一個團隊同佢嘅工時；AI 原生賣一套 operating model 同判斷，由部機器做第一 pass。 AI Agency contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/06/28/what-stays-in-house-vs-agency-ai-era/ https://chandlernguyen.com/zh-HK/blog/2026/06/28/what-stays-in-house-vs-agency-ai-era/ Sun, 28 Jun 2026 00:00:00 GMT 喺 AI 時代，內部 vs agency 嘅決定唔再主要係成本問題。保留啲會複利嘅層——brand 判斷、 first-party data、以及令 AI 隨時間變好嘅客戶 memory——將彈性 production 同專業執行交俾 agency。決定性嘅問題已經由價錢轉向複利。 AI Marketing contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/06/26/after-i-shipped-prova-i-rewrote-the-curriculum-two-more-times/ https://chandlernguyen.com/zh-HK/blog/2026/06/26/after-i-shipped-prova-i-rewrote-the-curriculum-two-more-times/ Fri, 26 Jun 2026 00:00:00 GMT 我喺四月份推出嘅產品，引擎係啱嘅。六月份定稿嘅產品，骨架係啱嘅。喺中間：兩次完整嘅課程重寫，一個喺一個禮拜內殺死咗成個架構嘅框架，同埋一條我由三月起就應該問嘅問題。 AI 產品 contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/06/20/how-many-people-does-an-ai-native-agency-need/ https://chandlernguyen.com/zh-HK/blog/2026/06/20/how-many-people-does-an-ai-native-agency-need/ Sat, 20 Jun 2026 00:00:00 GMT 同樣嘅產出，AI 原生 agency 需要嘅人比傳統 agency 少，而個數字由 review 能力、唔係 production 量決定。形態係一個細而 senior 嘅 pod，承載過去需要一個更大團隊嘅工作。更少、 更 senior、更 accountable，就係全部設計。 AI Agency contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/06/14/how-to-price-ai-agency-services/ https://chandlernguyen.com/zh-HK/blog/2026/06/14/how-to-price-ai-agency-services/ Sun, 14 Jun 2026 00:00:00 GMT 俾判斷同結果定價，唔好俾工時定價。AI 打破咗時間同成本之間嘅連繫，所以按小時定價會因為 你變快而懲罰你。AI 原生嘅修法係：為你客戶實際喺買嘅嘢——個決定、個標準、個結果——定價， 將 production 當作你控制嘅一項輸入成本。 AI Agency contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/06/10/offline-mode-didnt-need-ai-it-needed-boring/ https://chandlernguyen.com/zh-HK/blog/2026/06/10/offline-mode-didnt-need-ai-it-needed-boring/ Wed, 10 Jun 2026 00:00:00 GMT 我差啲就為 DIALØGUE 個 iOS app 整咗一個聰明、專門為 AI 而設嘅離線模式。後來我問咗一個更悶嘅問題——iOS 究竟已經畀咗啲乜嘢我？——結果先發現，真正嘅工夫係要忍住「想表現得聰明」嘅衝動。呢篇就講「下載一集，然後喺飛機上聽」背後嗰套標準、毫不華麗嘅機制，以及點解揀悶嗰條路先至係老手嘅決定。 AI Build in Public contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/06/07/is-an-ai-native-agency-cheaper/ https://chandlernguyen.com/zh-HK/blog/2026/06/07/is-an-ai-native-agency-cheaper/ Sun, 07 Jun 2026 00:00:00 GMT 一間 AI 原生 agency 按單位產出通常平啲，總花費上卻好少平啲。慳到嘅體現為彈性同速度—— 更多 variant、更多市場、更快嘅變化——先於體現為一張更細嘅單。而最平嗰份報價，往往正係 review standard 靜靜雞滑落咗嗰一份。 AI Agency contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/06/01/native-isnt-a-skin-rebuilding-dialogue-ios/ https://chandlernguyen.com/zh-HK/blog/2026/06/01/native-isnt-a-skin-rebuilding-dialogue-ios/ Mon, 01 Jun 2026 00:00:00 GMT 幾個月前，我將 DIALØGUE 個 iOS app 當成 web 版嘅移植咁出咗街，之後又幾乎全部重做成一個真正 native 嘅 app——三個 tab、上鎖畫面音頻、同步字幕、捱得住爛網絡嘅 offline，仲有 Siri——因為一個被壓細塞入手機嘅 web app，講到尾都仲係一個 web app。 AI Build in Public contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/05/30/what-agencies-should-stop-selling-ai-era/ https://chandlernguyen.com/zh-HK/blog/2026/05/30/what-agencies-should-stop-selling-ai-era/ Sat, 30 May 2026 00:00:00 GMT 停止賣被當成技能定價嘅拼裝——啲價反映佢過去所需人工、而非而家所交付價值嘅服務。 噉意味住手工 report 包、按量嘅零件生產項、作為獨立服務嘅送審同 QA、以及成量嘅通用內容。 難嘅部分唔係決定，而係退出時唔失去個客戶。 AI Agency contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/05/25/opus-for-design-codex-for-execution-rebuilt-my-site/ https://chandlernguyen.com/zh-HK/blog/2026/05/25/opus-for-design-codex-for-execution-rebuilt-my-site/ Mon, 25 May 2026 00:00:00 GMT TRANSMISSION做完咗佢嘅工。四個半月之後，係時候行落去。趁住Memorial Day長週末，我重建咗成個網站，但更值得講嘅其實係個workflow：Claude Opus 4.7負責design嘅判斷，GPT-5.5上面嘅Codex負責execution，加上`/goal`功能，Codex一次可以自己跑接近四個鐘。 AI Build in Public Design contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/05/23/how-to-start-an-ai-native-agency/ https://chandlernguyen.com/zh-HK/blog/2026/05/23/how-to-start-an-ai-native-agency/ Sat, 23 May 2026 00:00:00 GMT 由一條你可以端到端跑嘅窄 wedge 開始開一間 AI 原生 agency，而唔係一份 full-service menu。 揀一種你可以砌個 context layer、喺單個客戶身上驗證個 operating model、並定低一個質素標準嘅 工作——然後才擴展。廣度係你賺返嚟嘅，唔係你開局就拿出來嘅。 AI Agency contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/05/18/what-publicis-is-really-buying-for-2-2b-liveramp-deal/ https://chandlernguyen.com/zh-HK/blog/2026/05/18/what-publicis-is-really-buying-for-2-2b-liveramp-deal/ Mon, 18 May 2026 00:00:00 GMT Publicis 已經同意以大約 22 億美金 enterprise value 收購 LiveRamp。我覺得 最有意思嘅問題，唔係呢單 deal 可唔可以取代 walled gardens，或者拯救 open web。唔係。更好嘅問題係：喺封閉生態之外，advertisers 到底仲需要啲咩。 Marketing AI contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/05/16/how-to-govern-marketing-data-for-ai/ https://chandlernguyen.com/zh-HK/blog/2026/05/16/how-to-govern-marketing-data-for-ai/ Sat, 16 May 2026 00:00:00 GMT 為 AI govern marketing 數據，歸結為 model 睇到任何嘢之前你做出嘅四個決定：佢可以睇咩、 可以喺邊度跑、邊個擁有產出、以及你保留幾耐。先砌 memory layer、後 govern 嘅團隊， 最終會喺律師嘅監督下重建佢。 AI Marketing contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/05/13/6-months-3-game-projects-0-shipped/ https://chandlernguyen.com/zh-HK/blog/2026/05/13/6-months-3-game-projects-0-shipped/ Wed, 13 May 2026 00:00:00 GMT 我一邊做 full-time job，一邊花咗 6 個月試住成為 game developer。 我做咗三個 project，寫咗 684 個 commits，最後真正推出嘅 game 係 0 個。 呢篇係講發生咗咩、我學到咩，以及點解最後乜都未推出。 AI Product contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/05/08/how-to-measure-ai-marketing-impact/ https://chandlernguyen.com/zh-HK/blog/2026/05/08/how-to-measure-ai-marketing-impact/ Fri, 08 May 2026 00:00:00 GMT Measure 被改變嘅決定同 cycle time，唔好 measure 產出量。一旦 model 令產出變平，產出就係 一個虛榮 metric。真正重要嘅訊號係工作推進幾快、需要幾多 review、AI 輔助嘅版本幾常真係 更好、以及隨之而嚟咩業務結果。 AI Marketing contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/05/04/i-rolled-back-2724-lines-after-one-ai-audio-change-broke-production/ https://chandlernguyen.com/zh-HK/blog/2026/05/04/i-rolled-back-2724-lines-after-one-ai-audio-change-broke-production/ Mon, 04 May 2026 00:00:00 GMT 我 push 咗一個本來應該係升級嘅改動去我嘅 podcast 平台語音系統。 六日之後，幾幾個 commit 之後，我刪走咗 2,724 行 code，rollback 返去之前用到嘅版本。 以下係成件事嘅經過，以及佢教我關於測試 production AI 改動嘅教訓。 AI Build in Public contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/05/01/which-marketing-tasks-to-automate/ https://chandlernguyen.com/zh-HK/blog/2026/05/01/which-marketing-tasks-to-automate/ Fri, 01 May 2026 00:00:00 GMT 先自動化高量、低判斷、可回退嘅 task，將低量、高判斷、唔可回退嗰啲留返俾人。Workflow audit 按頻率、判斷同回退性俾每個 task 排序。大部分團隊自動化錯咗嘢，因為佢哋按工具睇落幾容易嚟揀， 而唔係按工作點樣表現。 AI Marketing contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/04/25/ai-creative-production-for-paid-media/ https://chandlernguyen.com/zh-HK/blog/2026/04/25/ai-creative-production-for-paid-media/ Sat, 25 Apr 2026 00:00:00 GMT AI 移除咗付費媒體入面作為 constraint 嘅 production 量，將個樽頸移到 concept 質素、 brand rules 同 review。Variant 唔再貴；判斷邊個 variant 值得跑才貴。贏嘅團隊係嗰啲 砌起咗 review standard 嘅，唔係生成最多 asset 嘅。 Paid Media AI contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/04/22/im-using-claude-code-for-everything-else-but-coding/ https://chandlernguyen.com/zh-HK/blog/2026/04/22/im-using-claude-code-for-everything-else-but-coding/ Wed, 22 Apr 2026 00:00:00 GMT 取消 Claude Max 19 日之後，我嘅使用模式已經好清楚。xHigh 嘅 GPT-5.4 Codex 坐咗去 coding 嗰個位，xHigh 嘅 Opus 4.7 Claude Code 就攞走咗我張枱上面其餘所有位置。 AI contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/04/17/ai-campaign-localization/ https://chandlernguyen.com/zh-HK/blog/2026/04/17/ai-campaign-localization/ Fri, 17 Apr 2026 00:00:00 GMT AI campaign localization 將份工由翻譯移到 transcreation 同 QA。交付物係母語細節，唔係字數； model 係引擎，人擁有關個 brand 同市場。將 AI localization 當成翻譯 task 嘅團隊，會交付 流暢卻錯失咗要點嘅文案。 AI Marketing contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/04/16/two-years-after-my-7-andrew-ng-courses-the-2026-path-id-actually-take/ https://chandlernguyen.com/zh-HK/blog/2026/04/16/two-years-after-my-7-andrew-ng-courses-the-2026-path-id-actually-take/ Thu, 16 Apr 2026 00:00:00 GMT 到而家仲有人 send message 嚟問我：2023 年我推介嗰 7 堂 Andrew Ng 嘅 課程，喺 2026 年係咪仲係啱條路？簡短版：大部分仲值得上，但外面包住佢 哋嗰張 roadmap 已經唔同咗。呢篇係我 2026 年嘅 update，每堂課畀我嘅最新 判斷，再加上畀 builder 同 operator 兩條唔同嘅叉路。 AI Learning contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/04/13/the-code-was-the-easy-part-of-shipping-prova/ https://chandlernguyen.com/zh-HK/blog/2026/04/13/the-code-was-the-easy-part-of-shipping-prova/ Mon, 13 Apr 2026 00:00:00 GMT 當 Prova 開始對 progression、billing、auth 同 state 作出 real promises， 我先覺得佢真係開始成為一個 product。最難嘅部分唔係 generate code，而係 build 起圍住呢啲 promises 嘅 contracts。 AI Product contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/04/09/what-agencies-actually-need-from-ai-is-not-more-content/ https://chandlernguyen.com/zh-HK/blog/2026/04/09/what-agencies-actually-need-from-ai-is-not-more-content/ Thu, 09 Apr 2026 00:00:00 GMT 我一路都見到 AI 工具向 agencies 推銷「內容產量」。但如果你真係處理過 client relationships，你會知道更難嘅問題其實係 trust：隔離、權限、 context，仲有絕對唔可以令一個 client 嘅 thinking 漏到去另一個 client 嗰邊。 AI contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/04/08/building-a-client-memory-data-moat/ https://chandlernguyen.com/zh-HK/blog/2026/04/08/building-a-client-memory-data-moat/ Wed, 08 Apr 2026 00:00:00 GMT 客戶 memory data moat，係一間 AI 原生店圍繞一個客戶砌起嘅 context layer——brief、結果、 決定、brand rules——令份工作複利、而唔係重新開始。競爭者複製得到你嘅工具；佢哋複製唔到 你累積嘅 context。嗰個係持久嘅優勢。 AI Agency contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/04/07/i-tried-to-clip-my-course-into-a-youtube-video-heres-why-i-rebuilt-it-instead/ https://chandlernguyen.com/zh-HK/blog/2026/04/07/i-tried-to-clip-my-course-into-a-youtube-video-heres-why-i-rebuilt-it-instead/ Tue, 07 Apr 2026 00:00:00 GMT 我以為只要將課程其中一個模組剪輯一下，修剪幾個過場，就可以當作 YouTube 影片。我錯了。製作 The Parade Problem 令我明白到，好嘅內容再利用唔係剪片 咁簡單。而係要為唔同嘅承諾、唔同嘅觀眾、唔同嘅頭三十秒重新設計個概念。 AI Product contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/04/03/why-i-cancelled-claude-max-after-13-months-and-what-im-testing-with-codex-next/ https://chandlernguyen.com/zh-HK/blog/2026/04/03/why-i-cancelled-claude-max-after-13-months-and-what-im-testing-with-codex-next/ Fri, 03 Apr 2026 00:00:00 GMT 我用咗 13 個月、總共畀咗 US$1,892.38 之後，正式取消咗 Claude Max。呢 唔係慶功文，而係一個 30 日實驗：睇下當 Codex 變成主要工具之後，我仲可 唔可以用同一個節奏繼續推進 STRATUM、DIALOGUE、課程平台同呢個網站。 AI contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/03/31/dropping-my-200-claude-code-plan-after-two-weeks-with-codex/ https://chandlernguyen.com/zh-HK/blog/2026/03/31/dropping-my-200-claude-code-plan-after-two-weeks-with-codex/ Tue, 31 Mar 2026 00:00:00 GMT 距離我上次比較兩個星期之後，兩個工具都出咗重大更新。Codex 挑戰咗我嘅產品策略，而 Claude Code 做唔到。Claude Code 就推出咗 Agent Teams 同 AutoMemory。結果係：我準備退訂 $200/月嘅 Max 計劃——而且用更少錢，出更好嘅成果。 AI contact@chandlernguyen.com (Chandler Nguyen) https://chandlernguyen.com/zh-HK/blog/2026/03/30/prompt-library-vs-operating-model/ https://chandlernguyen.com/zh-HK/blog/2026/03/30/prompt-library-vs-operating-model/ Mon, 30 Mar 2026 00:00:00 GMT 一個 prompt library 話俾人知該打咩字。一個 operating model 話俾佢哋知該自動化咩、該留 返俾人咩、工作點樣喺兩者之間流動、以及邊個為產出負責。只有其中一樣會複利。大部分以為 自己有 AI 策略嘅團隊，其實有一個 prompt library。 AI Strategy contact@chandlernguyen.com (Chandler Nguyen)