H
Howardism
Plate IIEntities機器翻譯 · machine-translatedENHOWARDISM

Andrej Karpathy

OpenAI 共同創辦人、前 Tesla AI 成員、Eureka Labs 創辦人;創造了「vibe coding」、「Software 1/2/3.0」、「ghosts not animals」與「agentic engineering」等說法;也是本知識庫採用的 LLM-wiki 模式的起源,並在約 3 個月內發展成工業化實作(DeepWiki、AutoWiki、OpenWiki、GBrain)

Article metadata
Publication details
Published:May 23, 2026
Filed:Entity
Domain:Entities
Tags:EntityPerson
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.

Andrej Karpathy 的插圖

資料來源#

摘要#

OpenAI 共同創辦人,曾任 Tesla AI 的 AI 總監(讓 Autopilot 實際運作),如今是 Eureka Labs(以 AI 為核心的教育)的創辦人。他是這個領域最擅長命名的人:創造了「vibe coding」、「Software 1.0 / 2.0 / 3.0」分類法、以「動物與幽靈」比喻模型智慧的說法,以及——在 2026 年 5 月的 Sequoia AI Ascent 訪談中——將「agentic engineering」稱為接續 vibe coding 的嚴肅專業。他也開創了本知識庫採用的 LLM-as-compiler 知識庫模式(llm-wiki gist),並在此訪談中再次表示,這也是他每天實際使用的方法。

「從未覺得自己如此落後」的坦白#

這場訪談令人意外的開場是:Karpathy——竟然連他都——說自己「從未覺得自己身為程式設計師如此落後」。他指出一個明顯的轉折點:2026 年 12 月,代理工具從「擅長寫一段段程式碼,有時還得修正它們」變成「產出的程式碼段落就是好的,我想不起來上次修正它是什麼時候」。他在 X 上敦促大家重新評估 12 月時的 AI——許多人把對 AI 的心智模型停留在「類似 ChatGPT」的階段,但此後代理式連貫工作流程已經徹底改變。這番坦白是氛圍編碼與代理工程一文的情感核心:連命名地板的人,都在努力追上天花板。

他開創的框架(各自已有專文)#

他自己的知識庫實踐#

他在訪談尾聲將教育與LLM-as-Compiler Knowledge Base連結起來:「每當我看到資訊的另一種投影方式,就會有所領悟……我每次讀文章時,都有自己的 wiki 正在從這些文章中建立起來,而我很喜歡就此提問。」他把這稱為在固定資料上生成合成資料——這種工具迫使資訊進入自己的腦海,因為「如果你不理解,就不可能成為好導演」。開創此模式的創辦人自己也在使用它,這正是本儲存庫直接承襲的脈絡。

gist 後來成了一個類別。從 2026 年 4 月發布起約三個月內,四種工業化實作陸續推出——Cognition 的 DeepWiki、Factory 的 AutoWiki、LangChain 的 OpenWiki、Garry Tan 的 GBrain——而 mem0 在 2026 年 7 月的調查(「The State of Agent Wikis」,practitioner-opinion)將該 gist 視為代理 wiki的奠基文件,並將四方不約而同採用他提出的三層 sources/wiki/schema-file 結構,視為此結構正確的證據。完整概況請見 LLM-as-Compiler Knowledge Base。

接著,這套模式也成了規格。 Google Cloud 的 Open Knowledge Format(2026-06-12,vendor-claim)一開頭就點出它要標準化的內容——一套「將 LLM-wiki 模式正式化為可攜、可互通格式」的開放規格——並在文中連結該 gist,具名引用他的簿記論點(「LLM 不會感到無聊、不會忘記更新交叉參照,而且一次就能處理 15 個檔案」),還採用他預留的兩個檔名 index.md 和 log.md 作為規格本身的保留名稱。一份具名且標明出處的 gist,在一季內成為供應商的互通標準,最有力地印證了本文對他「命名者」角色的論點:他發布的成果是一份設計文件,而傳播開來的是它的詞彙與檔案配置。

重要引言#

  • 「你可以外包思考,但不能外包理解。」
  • 「我們不是在打造動物,而是在召喚幽靈。」(參差不齊、統計生成、被召喚而來的存在)
  • 「Vibe coding 是為了拉高能力底線……代理工程則是為了維持專業軟體原有的品質標準。」
  • 「為什麼大家還在告訴我該做什麼?我什麼都不想做。我該複製貼上什麼給我的代理?」
  • 談到 MenuGen:「我所有的 menu gen 都是虛假的……那個應用程式根本不該存在。」

相關連結#

  • Software 3.0 — 他的分類法;也是訪談的概念主軸
  • 氛圍編碼與代理工程 — 他在相隔一年的時間裡創造了這兩個說法
  • 可驗證性論點 — 他試圖解釋模型為何能力參差不齊
  • 鋸齒狀智慧(幽靈,而非動物) — 「幽靈而非動物」文章的延伸應用
  • 代理原生基礎設施 — 他「最在意的小毛病」
  • 外包思考,而非理解 — 他對「什麼值得深入學習」的回答
  • LLM-as-Compiler Knowledge Base — 他開創了這個模式(llm-wiki),並每天使用
  • Open-Ended Discovery Harnesses — autoresearch(2026)是他提出的,也定義了這個類別:在實驗迴圈中執行現成的程式碼代理,持續數小時或數天,提交改進並還原退步。它衍生出系統最佳化、核心最佳化與提示注入等後續系統;後來同類的框架都以它作為論辯對象——它的兩個選擇(一份持續累積的上下文、一個可編輯的程式狀態)被指出是點子逐漸消失的原因
  • 苦澀的教訓 — Sutton 原則;他將神經網路視為宿主程序的推演便以此為基礎
  • Claude Code — 他提到自己常用的代理式程式設計介面:「cloud code / codex / open claw」
  • Boris Cherny — 平行的「程式設計已解決」發展脈絡;兩人都把自己當作案例研究
  • OpenAI — 他共同創辦的公司;也是 Software 3.0 / vibe coding 脈絡的起點,並反覆出現在本知識庫中
  • DHH(David Heinemeier Hansson) — 對這兩個新詞最強烈的反對者:「agentic engineering……我他媽討厭這個詞」,而 vibe coding「聞起來就像 2000 年代初期的 script kiddies」——但他全面採用這套實踐,並提出以實作可見度作為區分標準(Lex Fridman #501,2026-08-26)

資料來源#

§ end
Cited by 34
  • Agent-Native Infrastructure×2

    Andrej Karpathy's observation that the digital world is still built for humans and must be…

  • AI as Primary Author×2

    Vibe Coding Vs Agentic Engineering — Karpathy's "you're still responsible for your software" is the…

  • DHH (David Heinemeier Hansson)×2

    Hates the vocabulary. "Agentic engineering — oh, I fucking hate that term… it's become marketing…

  • Jagged Intelligence (Ghosts, Not Animals)×2

    Andrej Karpathy's mental model for what LLMs are: not animal intelligences shaped by evolution,…

  • LLM-as-Compiler Knowledge Base×2

    The landscape above is four bespoke implementations that resemble each other by convergence rather…

  • OpenAI×2

    OpenAI is an AI research company and the maker of the GPT‑5 series (including GPT‑5 Thinking and…

  • Outsource Your Thinking, Not Your Understanding×2

    Andrej Karpathy's answer to "what's still worth learning deeply when intelligence gets cheap,"…

  • Software 3.0×2

    Andrej Karpathy's taxonomy of three programming paradigms: Software 1.0 is explicit code; Software…

  • The Bitter Lesson×2

    Andrej Karpathy — frequent invoker of the principle (verifiability, ghosts, Software 3.0 all rest…

  • The Verifiability Thesis×2

    Andrej Karpathy's organizing claim for what AI automates and when: traditional computers automate…

  • Verification as the New Bottleneck×2

    Verifiability Thesis — Karpathy's "automate what you can verify" is the model-level cause; this is…

  • Vibe Coding vs. Agentic Engineering×2

    Andrej Karpathy coined "vibe coding" in 2025 and, a year later, names its serious successor:…

  • Acceleration Whiplash

    Vibe Coding Vs Agentic Engineering — the dark mirror: this is what the data looks like when orgs…

  • Agent-Authored Harness Optimization

    The setup is this page's loop with the sharpest available statement of why prompt tuning is not…

  • Agent Harness Engineering

    Agent Native Infrastructure — building agent-legible environments is harness engineering at the…

  • Agent Loop Pattern

    Agent Native Infrastructure — always-on loops acting via sensors/actuators are the runtime of…

  • Agent Quality Flywheel

    The loop is Karpathy-style autoresearch — propose an edit, evaluate against the calibrated judge,…

  • Boris Cherny

    Vibe Coding Vs Agentic Engineering — Karpathy's parallel "coding is solved" arc; both treat…

  • Claude Code

    Andrej Karpathy — power user ("cloud code / codex / open claw"); frames the discipline as agentic…

  • Claude Opus 4.7

    Jagged Intelligence — Karpathy's "Opus 4.7 will refactor a 100K-line codebase or find zero-days,…

  • Compute Allocator

    Outsource Thinking Not Understanding — Karpathy's twin framing: you can outsource the thinking (and…

  • Context Advantage, Not Taste

    Outsource Thinking Not Understanding — Karpathy's residue is understanding, which is closer to Ng's…

  • Disposable Micro-Apps

    Software 3 0 — micro-apps are Software-3.0-native: Karpathy's MenuGen "that app shouldn't exist" is…

  • Elon Musk

    Andrej Karpathy — founding OpenAI team member; the corpus's other voice from that lineage, on the…

  • Evals as Product Spec

    Verifiability Thesis — Karpathy's "automate what you can verify"; evals are verification authored…

  • Harness Shrinkage as Models Improve

    Vibe Coding Vs Agentic Engineering — Karpathy's ">10x and widening" leverage curve is the…

  • Loop Engineering

    Loop engineering is replacing yourself as the person who prompts the agent — you design the system…

  • MCP and Computer Use

    Agent Native Infrastructure — MCP is what makes a service agent-legible (structured); computer use…

  • Entities — People, Orgs, Tools & Projects

    Andrej Karpathy — Co-founder OpenAI, ex-Tesla AI, Eureka Labs; coined "vibe coding," Software…

  • Model Introspection Feedback

    Jagged Intelligence — introspection presumes a ghost, not an animal: the model's "why did I fail"…

  • Open-Ended Discovery Harnesses

    Andrej Karpathy — author of autoresearch, the single-agent loop whose two design choices (one…

  • OpenClaw

    The canonical agent-native install. Karpathy's go-to example of Software 3 0 / Agent Native…

  • Printing Press Software Democratization

    Vibe Coding Vs Agentic Engineering — Karpathy's "vibe coding raises the floor" is the same…

  • Scale-Dependent Prompt Sensitivity

    Verifiability Thesis — Karpathy's "jagged because of what the labs trained on" is the…

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…

  • Vibe Coding vs. Agentic Engineering

    Vibe coding raises the floor (anyone builds); agentic engineering preserves the quality bar while going faster; ">10x a…

  • Claude Code

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

  • Agent Harness Engineering

    Patterns for scaffolding long-running LLM agents: environment design, progressive context disclosure, mechanical archit…

  • Loop Engineering

    Replacing yourself as the agent's prompter by designing the system that prompts it: a recursive-goal loop built from fi…