資料來源#
摘要#
Fiona Fung 對技術辯論的規則是:當代理式程式碼生成讓產出幾乎免費時,「技術辯論中,程式碼勝出——建造很便宜,爭論很昂貴。」 當你能在預約一次白板會議所需的時間內生成三個真實實作,就不會再爭論採用哪種方法,而會開始比較它們。在 Claude Code 上,對應的規範是:製作原型並實際試用,而不是撰寫設計文件;減少深度規劃文件,改以在 PR 和原型中討論。
產生三個 PR 的做法#
Fung 分享了她的親身經歷:剛開始使用 Claude Code 時,她和 Boris Cherny 爭論重構方法。過去的直覺是:「我們去會議室開白板會議吧。」但她改為:「多虧 Claude,我生成了三個不同版本的 PR」——這讓技術辯論更有品質,因為她不只能比較程式碼如何實作,還能比較對被呼叫端的影響。實體成果(真實、可執行的 PR)比論點(白板上的抽象概念)承載更多資訊,而如今產出它的成本很低。
用原型取代設計文件#
同樣的經濟條件也改寫了原型製作。過去,原型製作把團隊分成兩派——快速取得回饋,以及偷工減料,最後卡在交付那些根本沒有為擴展性而打造的拋棄式程式碼。有了 Claude,第二種反對意見變弱了:「原型製作是很好的起點,因為我們可以迭代並學習,之後再更快地把原型擴展到 production。」因此 Claude Code 的規範是:有想法 → 製作原型 → 實際試用 → 然後再做決定。(參見 prototype「用來充實設計的拋棄式原型」做法。)
減少了什麼:設計文件#
團隊減少的是:「真正深入的規劃與設計文件。我們大多數的討論其實都發生在 PR 或原型中。」這是 Verification as the New Bottleneck 的下游結果——當工程頻寬不再稀缺時,那些原本用來保護頻寬的繁重前期規劃,便失去了存在的理由。重構與架構清理過去常在「我們什麼時候才找得到時間?」的預算爭奪中敗下陣來,如今也不再是瓶頸。
需要維持的張力#
這項規範與 wiki 中「先規劃」的概念形成了有益的張力。Design Concept Grilling(Matt Pocock)和 Vertical Slice Tracer Bullets 主張,建造之前必須先達成共享的設計概念。兩者的調和方式是:「建造很便宜」並不會取消設計思考——而是重新安置它。你仍然需要品味與規格(Outsource Your Thinking, Not Your Understanding),但你是透過比較已建造的成果,而不是在抽象層次爭論來形成它們。便宜的建造讓原型成為設計的媒介,而不是設計的替代品。(Karpathy 在 Vibe Coding vs. Agentic Engineering 中也取得了相同的平衡:規格仍由人類負責。)
相關連結#
- Implementation Abundance Inverts Product Work — 同樣「便宜的建造會重新安置困難部分」的邏輯,擴展到整個產品流程(策展 90 個平行建造)
- Fiona Fung — 「建造很便宜,爭論很昂貴」
- Boris Cherny — 產生三個 PR 的重構辯論夥伴
- Verification as the New Bottleneck — 上游原因:生成便宜 → 為保護頻寬而規劃失去意義
- Design Concept Grilling — 有益的張力:先設計再建造 vs. 透過建造做決定(調和方式:原型就是設計媒介)
- Vertical Slice Tracer Bullets — tracer bullets 是「建造一個薄而真實的東西來學習」的同一種直覺
- Building Is Cheap, Arguing Is Expensive — 另見 Disposable Micro-Apps:作為決策單位的拋棄式建造成果
- Disposable Micro-Apps — Thariq Shihipar 的拋棄式 UI,同樣遵循「便宜的建造成果勝過抽象論點」的邏輯,應用於編輯計畫
- Outsource Your Thinking, Not Your Understanding — 建造很便宜,但人類仍提供評判三個 PR 的品味與規格
- AI Native Product Cadence — 以原型取代文件,是更輕量 PRD 節奏的一部分
- Prototype Over PRD — Dan Carey 將「建造成果勝過抽象論點」一路推進到規格撰寫:跳過 PRD,從一段探討 why 的對話生成多種原型變體
開放問題#
- 「生成三個並比較」何時會變得浪費——決策權重達到什麼程度時,一場真正的辯論(或一份設計文件)仍比三個實作更便宜?
- 如果設計討論存在於 PR/原型中,未來的讀者要在哪裡找到理由——「我們為何選擇這個」的知識能否留存,還是會遇到 Code as Source of Truth 同樣的過時問題?
推導#
- The PRD-Replacement Spectrum at AI-Native Speed — 作為光譜中點的「建造以做決定」;本頁提出的理由缺失問題,是該光譜第一個尚未解決的債務
- Where Does the Why Live? — 針對本頁理由開放問題的聚焦回答:PR 中的討論會埋沒 why;沒有任何候選位置能持久保存它
資料來源#
Cited by 16
- The PRD-Replacement Spectrum at AI-Native Speed×9
The orphaned rationale (job #3). If design lives in PRs and prototypes and the PRD is deleted after…
- Prototype Over PRD×5
Build to decide · Building Is Cheap Arguing Is Expensive (Fiona Fung) · Generate three real PRs and…
- Where Does the Why Live?×3
Design discussion gets buried in merged PRs — Building Is Cheap Arguing Is Expensive open question:…
- Open Questions Backlog×2
Building Is Cheap Arguing Is Expensive: If design discussion lives in PRs/prototypes, where is the…
- Playbook Boundary Conditions: the Devil's-Advocate Substrate and the Prototype's Edge×2
Q1: Problem Solution Fit Discipline, Claude Character As Product, Harness Shrinkage As Models…
- Boris Cherny
Building Is Cheap Arguing Is Expensive — debated refactoring approaches with Fiona Fung by…
- Code as Source of Truth
Building Is Cheap Arguing Is Expensive — if code wins debates, the repo (not docs) is the source of…
- Design Concept Grilling
Building Is Cheap Arguing Is Expensive — productive tension: Fiona Fung's "generate three PRs and…
- Disposable Micro-Apps
Building Is Cheap Arguing Is Expensive — throwaway built artifacts as the unit of decision; cheap…
- Fiona Fung
Norms rewrote themselves — TDD lost its tax; refactoring stopped being a budget fight; design docs…
- Implementation Abundance Inverts Product Work
Building Is Cheap Arguing Is Expensive — the same "cheap building relocates the hard part" logic,…
- AI Coding Practice
Building Is Cheap Arguing Is Expensive — "In technical debate, code wins": generate three PRs vs…
- Outsource Your Thinking, Not Your Understanding
Building Is Cheap Arguing Is Expensive — generating PRs to settle a debate still requires…
- Verification as the New Bottleneck
Building Is Cheap Arguing Is Expensive — the upstream half: generation is cheap, so verification…
- Vertical Slice Tracer Bullets
Building Is Cheap Arguing Is Expensive — a tracer bullet is the "build a thin real thing to learn"…
- Vibe Coding vs. Agentic Engineering
Building Is Cheap Arguing Is Expensive — cheap generation is what lets prototypes settle technical…
Related articles
- 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…
- HTML as the New Markdown
Thariq Shihipar's thesis: as models improve, thousand-line markdown plans overwhelm the *human*; HTML artifacts (visual…
- Prototype Over PRD
Dan Carey's prototype-replaces-PRD method: record a why-not-what conversation, transcribe it, hand the transcript to Cl…
- Claude Code
Anthropic's agentic coding product; created by Boris Cherny late 2024; TypeScript/React on Bun (itself Claude-rewritten…
- Compute Allocator
The human's evolving role: deciding what's worth spending compute on; ~1% of generated tokens ship, 99% is scaffolding…
