資料來源#
摘要#
機器學習領域最廣為人知的教育者與實務工作者之一：Google Brain 創始負責人、Coursera 共同創辦人、Baidu 前首席科學家、DeepLearning.AI 與 Landing AI 創辦人、AI Fund 管理普通合夥人，以及 Stanford 兼任教授。他撰寫 The Batch，這是一封每週發行的公開信，也是該領域流通量最高的實務工作者平台之一。在 X 上使用 @AndrewYNg。
本 wiki 透過一封 2026 年 6 月的公開信認識他；該信也交叉發布於 X，主題是他如何運用 coding agents 建構 0 到 1 的產品。
2026 年 6 月的公開信#
這封信是在 Boris Cherny 與 Peter Steinberger 走紅後，「loop engineering」成為流行語的回應——同樣是 Loop Engineering 建立其論述的兩段引言。Ng 的切入點是指出：每個人都在最佳化其中一個迴圈，但產品其實依賴三個迴圈運作。
- The Three Loops of AI-Native Building — agentic coding loop（由 agent 完成閉環，數分鐘）、developer feedback loop（由人類完成閉環，數十分鐘至數小時）、external feedback loop（由市場完成閉環，數小時至數週）。每個外層迴圈都比它所包含的迴圈慢一到兩個數量級。「這些迴圈不只指引我如何建構軟體，也指引我決定要建構什麼軟體。」
- Context Advantage, Not Taste — 這封信最尖銳的一句話，也是對 wiki 核心開放問題的直接重新詮釋。「許多人將這項人類貢獻描述為『品味』，但我更傾向把它理解為人類擁有脈絡優勢，因為這為協助 AI 系統變得更好提供了更清晰的路徑……只要人類知道某些 AI 不知道的事，就需要 human-in-the-loop。」
- QA 是消失的工作。「去年，許多開發者（包括我在內）都在為 coding agents 擔任 QA 職能……隨著 coding agents 更有能力測試自己的程式碼，我們需要投入這項職能的時間已大幅減少。」人類是被提升離開 QA，而不是被移出迴圈。這與 Verification as the New Bottleneck 及 Faros's telemetry 的觀察相反；範圍差異（個人的 0 到 1 建構與 production 組織）可能正是兩者的調和點。
- Evals 是一種反應，而非預防措施。「如果你發現系統反覆遇到某些問題，為 agent 建立一組 evals 就會變得有用。」這比把 Cat Wu's "ten great evals" as the spec 當作規格更便宜，也更偷懶。
- 工程師投入產品工作。「越來越多工程師開始扮演部分產品管理角色……最困難的部分，是塑造產品願景，並在建構與取得使用者回饋之間取得平衡。」這是 Engineer PM Convergence 的第三份獨立報告，並指出一種具名的失敗模式：工程師把自己喜歡的迴圈跑過頭。
他反覆使用的例子，是一款他在某個週末為女兒打造的打字練習 app；coding agent 無人看管地「運作了大約一小時，使用網頁瀏覽器多次檢查自己建構的成果，之後才回來找我」。這是軼聞，不是測量結果——信中的一切都屬於 practitioner-opinion。
相關連結#
- The Three Loops of AI-Native Building — 他的分類法；也是 wiki 描繪 loop engineering 未觸及之處的地圖
- Context Advantage, Not Taste — 他將人類殘餘角色重新詮釋為資訊不對稱，而非某種能力
- Loop Engineering — 他正在定位並劃定邊界的紀律；他將命名這套方法歸功於 Boris Cherny 與 Peter Steinberger
- Engineer PM Convergence — 在 Cat Wu 與 Boris Cherny 之後，對同一角色融合現象的第三個獨立觀點
- Research Taste as the Human Bottleneck — 他「脈絡優勢」的旁論認為這個頁面提出了錯誤問題
- Andrew Ambrosino — 在 Ng 發表迴圈分類的同一週，發表了「loops are so last week」；只有在把 harness-loops 與 product-loops 混為一談時，兩者才會產生分歧
資料來源#
- Thread by @AndrewYNg — Andrew Ng，《The Batch》（2026-06-30），取自 X 的節錄（
practitioner-opinion）
Cited by 25
- Peter Steinberger×3
Austrian developer best known as the founder of PSPDFKit (a widely-licensed PDF SDK) which he built…
- Context Advantage, Not Taste×2
In a June 2026 letter otherwise devoted to a loop taxonomy, Andrew Ng makes an aside that quietly…
- Engineer PM Convergence×2
Three Loops Of Ai Native Building — Andrew Ng's third-vantage report, plus the failure mode:…
- Loop Engineering×2
Two weeks after Osmani's essay, Andrew Ng responded to loop engineering "becoming a hot buzzphrase…
- Research Taste as the Human Bottleneck×2
Context Advantage Over Taste — the third answer: Andrew Ng argues taste is not a capability but an…
- The Three Loops of AI-Native Building×2
Andrew Ng's June 2026 letter in The Batch opens by observing that "loop engineering" had become a…
- The Abstraction Barrier
Context Advantage Over Taste — the generative asymmetry this page defends, stated as the boundary…
- Acceleration Whiplash
Three Loops Of Ai Native Building — the telemetry that outranks Andrew Ng's self-report: he claims…
- Agent Loop Pattern
Three Loops Of Ai Native Building — Andrew Ng's taxonomy places this primitive: it closes the…
- Andrew Ambrosino
Three Loops Of Ai Native Building — Andrew Ng published a three-loop taxonomy the same week…
- The Automation–Optimism Link
Three Loops Of Ai Native Building — a candidate instance of the gap: Andrew Ng self-reports that…
- Boris Cherny
Three Loops Of Ai Native Building — Andrew Ng credits him (with Peter Steinberger) for making "loop…
- Evals as Product Spec
Three Loops Of Ai Native Building — the productive disagreement on when to write evals: Andrew Ng…
- Implementation Abundance Inverts Product Work
Context Advantage Over Taste — what the new bottleneck is: Andrew Ng argues the "taste" curation…
- Jagged Intelligence (Ghosts, Not Animals)
Context Advantage Over Taste — Andrew Ng displaces the framing this page supplies for taste: "a…
- Entities — People, Orgs, Tools & Projects
Andrew Ng — Founder of DeepLearning.AI and AI Fund, founding lead of Google Brain, co-founder of…
- OpenClaw
Open-source personal AI agent and harness created by Peter Steinberger (openclaw.ai) — the project…
- Outsource Your Thinking, Not Your Understanding
Context Advantage Over Taste — Andrew Ng locates the residual human role in what they know rather…
- Task Time-Horizon Scaling
Three Loops Of Ai Native Building — one anecdotal point on the curve: Andrew Ng's coding agent…
- Thariq Shihipar
Context Advantage Over Taste — his unknown knowns are Andrew Ng's context advantage seen from…
- Transformative Creativity
Context Advantage Over Taste — the limit Andrew Ng's frame doesn't reach: Boden level-3 creation of…
- Unknowns as the Agentic Bottleneck
Three Loops Of Ai Native Building — vision→spec lossiness is this problem stated as a loop: Andrew…
- Verification as the New Bottleneck
Three Loops Of Ai Native Building — the direct dissent. Andrew Ng reports the opposite motion:…
- Vibe Coding vs. Agentic Engineering
Three Loops Of Ai Native Building — Andrew Ng published a loop taxonomy the same week Ambrosino…
- Why AI Lags at Design
Context Advantage Over Taste — the test case for Andrew Ng's reframe: if design taste is a closable…
Related articles
- The Three Loops of AI-Native Building
Andrew Ng's nested-loop taxonomy for 0-to-1 products: the agentic coding loop (minutes, agent-closed), the developer fe…
- Harness Shrinkage as Models Improve
Prompt scaffolding shrinks each model release; Cat Wu's pruning discipline; Boris Cherny "100 lines of code a year from…
- Boris Cherny
Creator of Claude Code at Anthropic; phone-driven workflow with hundreds of agents; primary advocate of `/loop` primiti…
- Claude Code
Anthropic's agentic coding product; created by Boris Cherny late 2024; TypeScript/React on Bun (itself Claude-rewritten…
- Context Advantage, Not Taste
Andrew Ng's reframing of the residual human contribution: not 'taste' but an information asymmetry — 'so long as the hu…