H
Howardism
Plate IIEntities機器翻譯 · machine-translatedENHOWARDISM

Boris Cherny

Anthropic 的 Claude Code 創作者；以手機驅動的工作流程同時運行數百個代理程式；`/loop` 原語的主要倡議者；「對我來說，寫程式已經解決了」論點；由消融法驅動的 harness 設計（刪除提示，再把模型反覆卡住的部分加回來）

Article metadata
Publication details
Published:May 6, 2026
Filed:Entity
Domain:Entities
Tags:EntityPersonAnthropic
Reading:7 min
Source:AI-synthesised
About this piece

Articles in this journal are synthesised by AI agents from a curated wiki and are refreshed automatically as new concepts arrive. Topics, framing, and editorial direction are curated by Howardism.

Boris Cherny 插畫

資料來源#

摘要#

Claude Code 的創作者與技術主管，任職於 Anthropic。他以工程師為背景，著有 Programming TypeScript。2024 年底透過內部孵化團隊「Anthropic Labs」加入 Anthropic（該團隊也催生了 Claude Code、MCP 與 Anthropic 桌面應用程式，之後解散；到 2026 年中，Mike Krieger 以「第二輪」名義重新組建）。截至 Cat Wu 登上 Lenny's Podcast（2026 年 4 月）時，他是節目史上最受聽眾收聽的來賓。

重要主張與立場#

  • 對他來說，寫程式已經解決了。 他所有程式碼都透過 Claude Code 撰寫；曾有好幾天每天提交 150 個 PR。補充說明：對於非常大型的程式碼庫或訓練分布之外的語言，這件事還不成立。
  • 以手機驅動的工作流程。 現在大部分工作都在 iOS Claude app 上完成；同時運行約 5–10 個工作階段，白天有「幾百個」代理程式，夜間則有「幾千個」。
  • 迴圈是未來。 他大量使用 /loop（由 Claude 排程的 cron 工作）來照看 PR、修復 CI、整理 Twitter 回饋。他認為 Agent Loop Pattern 與平行代理程式扇出是下一個 harness 原語（參見 Harness Shrinkage as Models Improve）。到 2026 年 6 月，這個立場更明確成為 Loop Engineering 的核心論點：「我現在不再提示 Claude。我會讓迴圈去提示 Claude，並想辦法決定接下來該做什麼。我的工作就是寫迴圈。」
  • 為下一個模型打造。 他在 2024 年底打造 Claude Code 時就知道產品市場契合度（PMF）尚不存在；2025 年 5 月 Opus 4 推出後，PMF 才到來，之後每次發布都帶來轉變（4.5 → 4.6 → 4.7）。
  • 跨領域的通才。 Claude Code 團隊每個人都會寫程式，包括工程經理、PM、設計師、資料科學家、財務人員與使用者研究員（參見 Engineer PM Convergence）。
  • 印刷機的類比。 他認為 Printing Press Software Democratization 是理解 AI 輔助軟體建置最恰當的歷史類比。
  • 把模型當作生物體，而非系統。（YC 訪談，2026 年 7 月）在模型之上建構「和我做過的所有工程工作都截然不同」——不需要預先設計系統；模型「幾乎就像一個活生生的生物……每一代模型的行為都不一樣，個性也略有不同」，因此應採取經驗式的做法：嘗試、觀察、反覆調整。他反對一招走天下的文化（「別聽 LinkedIn 網紅的話」），並指出過度規格化是資深工程師常犯的錯誤——說明任務、護欄與退出條件，「接著就讓模型放手去做」（Vibe Coding vs. Agentic Engineering）。
  • 驗證是引出能力的關鍵。「交給它一個太難的任務，再給它工具，讓它像你自己一樣驗證成果」——驗證「大概是人們最常做不對、也最重要的一件事」。他常用的示範是一個在 Slack 裡的 Claude 工作階段，將 Electron 桌面應用程式改寫成 Swift，並在 macOS VM 裡替兩個應用程式截圖、逐像素比對來驗證；這項工作持續了兩週以上，還會把進度截圖即時發布到自己建立的 Slack 頻道。
  • 「寫程式已解決」的適用範圍（更新版）。 仍不包括深層系統程式碼、分散式系統，以及像素層級的 UI 驗證；他在 YC 訪談（2026 年 7 月）進行的聽眾投票顯示，完全使用代理程式撰寫程式碼的人占了很大一部分，約有一半的受訪者使用代理程式撰寫超過 50% 的程式碼。

背景#

他在國中時第一次學會用 TI-83 計算機寫程式——用 BASIC 寫了一個代數解題程式，在數學考試時作弊，後來因微積分超出 BASIC 的能力而改用組合語言重寫；他也曾寫過一份早期的 TI-83 程式設計網路指南。他給資工系學生的建議由此而來：透過打造實用作品來學習，培養設計感、商業感與和使用者交談的能力，不要只學理論。

工作風格（根據 Cat Wu 的說法）#

Cat Wu（Claude Code 與 Cowork 的產品主管）形容 Boris 是「產品願景家」，負責設定 3–6 個月的願景；她則將願景轉化為近期的行動路徑，並處理跨職能協調。他們估計兩人有 80% 的想法相通，剩下 20% 則分屬各自負責的領域。這種分工「界線非常模糊」——也是團隊內 Engineer PM Convergence 的典型情形。

重要引言#

  • 「迴圈就是現階段的未來。如果你還沒試過，我非常非常推薦。」
  • 「軟體將會完全普及，人人都能打造……我想也許即使是今天，最適合寫會計軟體的人也不是工程師，而是非常優秀的會計師。」
  • 「如果你是從頭開始，那就可以用 AI 原生的方式從基礎開始打造。」

相關連結#

資料來源#

§ end
Cited by 49
  • Opinions on Using AI Tools & the Future of the Software Engineering Role×4

    "Coding is solved (for me)." Boris writes 100% of his code via Claude Code, has logged 150-PR days,…

  • Dynamic Workflows: An Algebra for Agents×4

    Dynamic workflows are a Claude Code feature that lets the model write and execute an orchestration…

  • Agent Loop Pattern×3

    A loop is an agent process that repeatedly executes a prompt until a queue is empty or a stopping…

  • Andrew Ng×3

    Engineer Pm Convergence — third independent vantage on the same role merge, after Cat Wu and Boris…

  • Claude Code×3

    Per Boris Cherny (publicly leaked in March 2026 via human error during a release-process PR —…

  • Cowork×3

    Boris Cherny cites Cowork as the venue for general-purpose MCP and computer use integration — same…

  • Engineer PM Convergence×3

    Both Boris Cherny (Sequoia AI Ascent 2026) and Cat Wu (Lenny's Podcast, April 2026) report the same…

  • Evals as Product Spec×3

    The suite is explicitly perishable. "As models improve, cases that once discriminated stop doing so…

  • Learning to Co-Work with AI: A Software Engineer's Field Guide×3

    Build AI-native from day one. Boris Cherny: a startup builds AI-native; an incumbent has to retrain…

  • Loop Engineering×3

    Boris Cherny's YC interview describes Osmani's worked shape running in production at Anthropic, at…

  • Peter Steinberger×3

    This is the seed of Loop Engineering — the shift from human-as-prompter to human-as-loop-designer.…

  • The Three Loops of AI-Native Building×3

    Boris Cherny / Peter Steinberger — the two practitioners Ng credits with making "loop engineering"…

  • Anthropic×2

    "We have no more manually written code anywhere at the company. All of the SQL is written by…

  • Build for the Next Model×2

    A product-strategy corollary of Harness Shrinkage As Models Improve, now stated independently by…

  • Building Is Cheap, Arguing Is Expensive×2

    Fung's war story: onboarding to Claude Code, she and Boris Cherny debated a refactoring approach.…

  • Compute Allocator×2

    "Coding is solved (for me)" (Boris Cherny) names the precondition; compute allocator names the role…

  • Fiona Fung×2

    Leads engineering and product for Claude Code and Cowork at Anthropic; previously built and led…

  • Harness Shrinkage as Models Improve×2

    The harness — prompts, skills, scaffolding, mechanical verification — exists to compensate for what…

  • Jarred Sumner×2

    His July 2026 post Rewriting Bun in Rust is the wiki's most methodologically detailed practitioner…

  • Latent Capability Overhang×2

    Boris Cherny — the product-side statement: "hobbling" / "product overhang," the Claude Code origin…

  • Managers as ICs×2

    This is the management-layer instance of Engineer Pm Convergence: Boris Cherny notes every member…

  • MCP and Computer Use×2

    Created at Anthropic Labs (late 2024) alongside Claude Code and the desktop app by Boris's founding…

  • Mythos Model×2

    Boris Cherny: "We use a little bit of Mythos to try it and then a lot of Opus 4.7 to dog food it…

  • Printing Press Software Democratization×2

    Boris Cherny's offered analogy for what AI is doing to software: the invention of the printing…

  • Seven Powers Applied to AI×2

    Boris Cherny uses Hamilton Helmer's 7 Powers framework (popularized by the Acquired podcast) to…

  • Shared Harness, Differentiated Surfaces×2

    The three differentiators are permissions/sandboxing, and UI. That list is almost verbatim the…

  • Vibe Coding vs. Agentic Engineering×2

    Printing Press Software Democratization — the floor-raising half is the same democratization Boris…

  • Addy Osmani

    Boris Cherny / Peter Steinberger — the two practitioners whose "design loops, don't prompt" framing…

  • Agentic Prompt Injection

    Days after the Opus 5 launch, Boris Cherny stated the strong version on stage (YC interview,…

  • AI Native Product Cadence

    Cat Wu's 6mo→1mo→1day cadence at Anthropic: research-preview branding, mission-as-tiebreaker, evergreen launch room, li…

  • Andrej Karpathy

    Boris Cherny — parallel "coding is solved" arc; both treat themselves as the case study

  • Andrew Ambrosino

    Cat Wu / Boris Cherny — Anthropic counterparts on role convergence; Ambrosino is the OpenAI-side…

  • Anthropic Labs

    Per Anthropic's entity page and Boris Cherny: a first incarnation of the Labs incubator formed in…

  • Bun

    Boris Cherny — gave the secondhand stage account of the port that the first-party post later…

  • Cat Wu

    Head of Product for Claude Code and Cowork at Anthropic; primary articulator of AI-native product cadence and engineer-…

  • Claude Code Auto Mode

    By July 2026 Boris Cherny describes auto mode as the enabling half of Opus 5's endurance claim —…

  • Claude Design

    Boris Cherny had flagged it as a vision example (recorded on Cowork): "Claude design I think is a…

  • Claude Fable 5

    Three things it establishes that the launch post does not. First, a capability discontinuity claim…

  • Claude Opus 5

    Boris Cherny's YC interview three days after release (practitioner-opinion, vendor-adjacent) adds…

  • Dan Carey

    Cat Wu / Boris Cherny / Fiona Fung — fellow Anthropic product/eng leaders whose AI-native-cadence…

  • Harness Build-vs-Buy

    Harness Shrinkage As Models Improve holds that the harness shrinks toward a residue as models…

  • HTML as the New Markdown

    At first glance this contradicts the wiki's running Harness Shrinkage As Models Improve thesis (Cat…

  • Entities — People, Orgs, Tools & Projects

    Boris Cherny — Creator of Claude Code at Anthropic; phone-driven workflow with hundreds of agents;…

  • Parallel Agent Orchestration

    The workflow default is an order of magnitude below the vendor's own headline. "Fewer than 15…

  • Returns to Expertise in Agentic Coding

    Printing Press Software Democratization — Cherny's "the best person to write accounting software is…

  • Thariq Shihipar

    Boris Cherny — Claude Code teammate; "coding is solved (for me)" is the precondition for Thariq's…

  • Unknowns as the Agentic Bottleneck

    > "The best agentic coders have relatively few unknowns. Watching someone like Boris or Jarred…

  • Verification as the New Bottleneck

    Boris Cherny restates the thesis from the capability direction rather than the org direction (YC…

  • White-Box Activation Monitoring

    By July 2026 the deployment is described publicly as universal: Boris Cherny (YC interview,…

Related articles
  • Claude Code

    Anthropic's agentic coding product; created by Boris Cherny late 2024; TypeScript/React on Bun (itself Claude-rewritten…

  • 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…

  • Anthropic

    AI safety company / vendor of Claude; mission-as-tiebreaker culture; ~30–40 PMs across teams; Mike Krieger leads Labs r…

  • Verification as the New Bottleneck

    Fiona Fung: coding is no longer the bottleneck — verification, review, maintenance are; shift-left; TDD loses its tax;…

  • Cat Wu

    Head of Product for Claude Code and Cowork at Anthropic; primary articulator of AI-native product cadence and engineer-…