資料來源#
- A Field Guide to Fable: Finding Your Unknowns
- Fable's judgement
- How I AI: Thariq Shihipar on Replacing Markdown with HTML for AI-Powered Development
- The new rules of context engineering for Claude 5 models
摘要#
他是 Anthropic Claude Code 團隊的工程師。「HTML 是新的 markdown」論點（參見 HTML as the New Markdown）以及將人的角色稱為 Compute Allocator 的說法，都源自他。兩人在舊金山的 Code with Claude 活動相識後，Claire Vo 邀請他參加「How I AI」系列訪談（該集於 2026 年 5 月發布）。個人網站為 thariq.io；X 帳號是 @trq212。
重要主張與立場#
- 「HTML 是新的 markdown。」 隨著模型能力提升、專案日益複雜，markdown 計畫會讓人類讀者不堪負荷；HTML 是更豐富的媒介，能讓人保持投入。瓶頸不在模型，而在人類的注意力。參見 HTML as the New Markdown。
- 我們都逐漸成為「compute allocators」。 人的工作是決定哪些事情值得投入運算資源；他產生的 token 約有 1% 進入正式環境，其餘 99% 都是支架（計畫、介面、狀態、設計系統），用來投入協作與溝通。參見 Compute Allocator。
- 豐裕思維。 生成成本低到足以負擔一次性工具——為單一編輯任務打造的 micro-apps（「建構在微型軟體之上的微型軟體」）。
- 提示中兼具限制與自由度。「提供足夠的限制，讓你得到想要的結果，但也要留點空間，讓模型帶給你驚喜」——「嘿，Claude，這部分我相信你的判斷。」
- 讓人重新回到核心位置。 他的工作流程明確不是要把開發者自動化取代掉，而是讓協作更清楚可見、更引人投入。
- 地圖不等於疆域。（2026 年 7 月的實作指南。）提示、技能與上下文是地圖；程式碼庫及其真實限制才是疆域；兩者之間的落差就是未知之處。「Fable 是第一個讓我覺得，工作品質的瓶頸在於我能否釐清它的未知之處的模型。」 減少未知並預先規劃「就是代理程式程式設計的技能」——而且可以學會。參見 Unknowns as the Agentic Bottleneck。
- 測驗關卡。「只有在我完美通過測驗後，我才會合併」——合併前，他會讓 Claude 就某項變更考問自己，因為若行為取決於既有程式碼路徑，只讀差異不足以充分理解。受驗證的是人，而不是代理程式。
- 原始碼是最好的參考。 如果你無法描述想要什麼，就指出能做到這件事的程式碼——即使它使用另一種語言。這就是 Claude Design 的運作方式：指向一個元件，它「讀取底層程式碼，而不只是看截圖」。
- 解除束縛。（2026 年 7 月的上下文工程文章。）Claude Code 團隊過度限制模型：他們刪除了 Claude 5 models 系統提示中超過 80% 的內容，評估結果沒有可測得的損失，並淘汰了一系列上下文工程「迷思」（規則→判斷、範例→介面設計、預先提供→漸進揭露、重複內容→工具描述、CLAUDE.md-memory→auto-memory、簡單規格→豐富參考資料）。以
claude doctor的形式提供給使用者。參見 Harness Shrinkage as Models Improve 與 Agent Context Files。他與 Cat Wu 在 AI Engineer World's Fair 的爐邊談話中，將規則→判斷這項原則作為面向使用者的建議：讓 Fable「運用自己的判斷，而不是指揮它該怎麼工作」，並以自動化測試為實際案例（2026-07-03；二手轉述於 Fable's judgement）。
他示範的三種工作流程#
- 互動式 HTML 腦力激盪與規劃 — 要求以 HTML 檔案而不是清單呈現計畫；內容具視覺性、可捲動，也方便分享。
- Disposable Micro-Apps — 產生一次性自訂介面，編輯計畫中的一個部分，並能複製回 markdown。
- Living Design System — 從程式碼庫擷取出
design_system.html，作為可攜、同時供人與機器讀取的真實資訊來源。
相關連結#
- Claude Code — 他在 Anthropic 的團隊
- Anthropic — 雇主
- Claire Vo — 邀請他參加「How I AI」訪談
- HTML as the New Markdown — 他的核心論點
- Compute Allocator — 他描述人類角色的說法
- Disposable Micro-Apps / Living Design System — 他的另外兩種工作流程
- Boris Cherny — Claude Code 團隊同事；「對我而言，程式設計已經解決」是 Thariq 提出運算資源分配者觀點的前提
- Harness Shrinkage as Models Improve — Thariq 關於 99% 支架的觀察，是面向人的說法，與團隊面向模型、精簡 harness 的做法形成互補
- Engineer PM Convergence — 他的運算資源分配者觀點，就是在呼叫尺度上實踐「決定要寫什麼」
- AI Native Product Cadence — 他看好更豐富的 HTML PRD，與 Cat Wu 看好更精簡的 PRD 形成對照
- Interaction Models — 兩者都以改善人與 AI 的協作為目標；他的 HTML 產物豐富了非同步管道，而 TML 則消解了即時介面
- Software 3.0 — 他的 HTML 優先工作流程與 Disposable Micro-Apps 都原生適用於 Software 3.0：根據提示為每項任務快速生成介面，而非手工打造應用程式
- Unknowns as the Agentic Bottleneck — 他在 2026 年 7 月提出的論點：使用 Fable 等級的模型時，主要限制在於人類能否說清楚自己從未寫下來的內容
- Context Advantage, Not Taste — 他所說的已知卻未明言之事，是從人的內在視角看 Andrew Ng 的上下文優勢；實作指南則是引出這些內容的方法
- Claude Fable 5 — 根據他的說法，這個模型讓瓶頸轉移到人身上
資料來源#
- How I AI: Thariq Shihipar on Replacing Markdown with HTML for AI-Powered Development — Claire Vo 在 ChatPRD 的「How I AI」訪談，2026 年 5 月
- A Field Guide to Fable: Finding Your Unknowns —「A Field Guide to Fable: Finding Your Unknowns」，發布於 X，2026-07-04（
practitioner-opinion） - Fable's judgement — Simon Willison，2026-07-03（
practitioner-opinion）：AI Engineer World's Fair 爐邊談話的建議，經二手轉述，並共同歸於 Cat Wu，但未說明各自說了什麼；資料庫中沒有逐字稿或錄音
Cited by 33
- Harness Shrinkage as Models Improve×4
The comments-rule rewrite above is Anthropic doing this to its own system prompt. Simon Willison…
- Agent Context Files×3
Thariq Shihipar's context-engineering post (July 2026, practitioner-opinion) restates the CLAUDE.md…
- Claude Fable 5×3
Unknowns As The Agentic Bottleneck — Thariq Shihipar's claim that Fable is the first model where…
- Compute Allocator×3
Thariq Shihipar's framing for what the human becomes once the model can do most of the producing:…
- Agent Harness Engineering×2
Thariq Shihipar's July 2026 post (source, practitioner-opinion) retires several harness-design best…
- Boris Cherny×2
Unknowns As The Agentic Bottleneck — Thariq Shihipar names him as the low-unknowns exemplar: "it is…
- Cat Wu×2
Judgement over rules. At an AI Engineer World's Fair fireside with Thariq Shihipar hosted by Simon…
- Claire Vo×2
Host of the "How I AI" interview series, published on ChatPRD (chatprd.ai), in which practitioners…
- Disposable Micro-Apps×2
Thariq Shihipar's answer to the obvious objection against HTML plans: HTML isn't as easily editable…
- HTML as the New Markdown×2
Thariq Shihipar's thesis, delivered to Claire Vo at Code with Claude (May 2026): "HTML is the new…
- Living Design System×2
Thariq Shihipar's third HTML-first workflow: maintain a design_system.html file — an artifact that…
- Outsource Your Thinking, Not Your Understanding×2
Compute Allocator — Thariq Shihipar's framing of the residual human role; deciding what's worth…
- Unknowns as the Agentic Bottleneck×2
Thariq Shihipar's July 2026 field guide names a bottleneck that is neither model capability nor…
- Agent Documentation Behavior
Unknowns As The Agentic Bottleneck — Shihipar's implementation-notes.md with its Deviations section…
- AI Brain Fry
Unknowns As The Agentic Bottleneck — a countermeasure to approval-without-comprehension: Thariq…
- AI Native Product Cadence
Html As The New Markdown — opposite bet on PRDs at speed: Cat makes them lighter (1-pagers, metrics…
- Anthropic
Thariq Shihipar — engineer on the Claude Code team; "HTML is the new markdown" workflows
- Building Is Cheap, Arguing Is Expensive
Disposable Micro Apps — Thariq Shihipar's throwaway UIs are the same "cheap built artifact beats…
- Claude Code
Thariq Shihipar — Claude Code engineer; "HTML is the new markdown" workflows (see Html As The New…
- Claude Design
Unknowns As The Agentic Bottleneck — the reference technique productized: point it at a component…
- Cost-per-Task Over Cost-per-Token
simonwillison judgement — Simon Willison, "Fable's judgement," 2026-07-03 (practitioner-opinion,…
- Design by Selection
Html As The New Markdown — an independent arrival at the same premise from a different Anthropic…
- Engineer PM Convergence
Compute Allocator — "deciding what to write" as the bottleneck skill, restated at the level of…
- Interaction Models
Html As The New Markdown — a sibling answer to "better human–AI collaboration" by the opposite…
- Loop Engineering
Unknowns As The Agentic Bottleneck — comprehension debt made checkable: Thariq Shihipar's quiz gate…
- Entities — People, Orgs, Tools & Projects
Thariq Shihipar — Engineer on the Claude Code team at Anthropic; "HTML is the new markdown",…
- Nate Parrott
Html As The New Markdown — arrived at the HTML-as-visual-medium insight independently of Thariq…
- Planning / Execution Division of Labor
Compute Allocator — "humans make the planning decisions" is exactly Thariq's allocator role:…
- The PRD-Replacement Spectrum at AI-Native Speed
The spectrum above moves in one direction — less prose. But there is an orthogonal response to the…
- Prototype Over PRD
Unknowns As The Agentic Bottleneck — a third use for the same artifact: Thariq Shihipar builds…
- Returns to Expertise in Agentic Coding
Unknowns As The Agentic Bottleneck — the mechanism behind the expertise premium: Thariq Shihipar…
- Security Debt of Agent-Generated Code
Unknowns As The Agentic Bottleneck — the countermeasure aimed at exactly this failure: Thariq's…
- Software 3.0
Html As The New Markdown — Thariq Shihipar's "build a throwaway UI per task" is a 3.0-native…
Related articles
- Claude Code
Anthropic's agentic coding product; created by Boris Cherny late 2024; TypeScript/React on Bun (itself Claude-rewritten…
- HTML as the New Markdown
Thariq Shihipar's thesis: as models improve, thousand-line markdown plans overwhelm the *human*; HTML artifacts (visual…
- 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…
- Compute Allocator
The human's evolving role: deciding what's worth spending compute on; ~1% of generated tokens ship, 99% is scaffolding…