資料來源#
摘要#
OpenAI 共同創辦人、前 Tesla AI 總監（曾讓 Autopilot 真正運作起來），如今是 Eureka Labs（一家 AI 原生教育公司）的創辦人。他是這個領域最多產的命名者：提出了「vibe coding」、「Software 1.0 / 2.0 / 3.0」分類體系、用以描述模型智慧的「animals vs. ghosts」框架，並在這場 2026 年 5 月的 Sequoia AI Ascent 訪談中，將「agentic engineering」定位為繼 vibe coding 之後一門嚴肅成熟的工程學科。他也開創了本知識庫所運行的 LLM-as-compiler 知識庫 模式（即 llm-wiki 那份 gist），並在這場訪談中再次將其背書為自己每日的實踐。
「從未感到如此落後」的自白#
訪談的開場令人震驚：Andrej Karpathy — 偏偏是他 — 自承「作為一名程式設計師，從來沒有像現在這樣感到如此落後」。他指出一次明顯的相變發生在 2026 年 12 月：agentic 工具從「擅長處理片段程式碼，有時你還得幫它們修正」，轉變為「片段直接就能輸出得很完美，我已經不記得上次出手修正是什麼時候了」。他在 X 上敦促大家以 12 月為基準重新評估 AI — 許多人的心智模型仍停留在「ChatGPT 那一類」，但自那以後，agentic-coherent 的工作流程已經發生根本性的改變。這段自白正是 Vibe Coding vs. Agentic Engineering 一文的情感引擎：連那個替地板命名的人，自己都在拼命追趕新的天花板。
他所開創的框架（每一項都已成為獨立條目）#
- Software 3.0 — 提示與情境即程式；LLM 作為一台可程式化的解釋器。
- Vibe Coding vs. Agentic Engineering — vibe coding 拉高下限；agentic engineering 守住品質門檻。
- The Verifiability Thesis — LLM 自動化你能驗證的事，正如傳統電腦自動化你能指定的事。
- Jagged Intelligence (Ghosts, Not Animals) — 「我們不是在打造動物，而是在召喚鬼魂」；能力呈尖刺狀，卻無內在動機。
- Agent-Native Infrastructure — 「為什麼文件還在為人類撰寫？我到底該把什麼貼給我的代理？」
- Outsource Your Thinking, Not Your Understanding — 教育論點；理解是人類最後無法外包的瓶頸。
關於他自己的知識庫實踐#
他在訪談尾聲將教育議題接回 LLM-as-Compiler Knowledge Base：「每當我看到對資訊的不同投影方式，就會獲得新的洞見……每當我讀到一篇文章，我都會把它加入那個由文章一點一滴累積起來的 wiki，而我喜歡對它提問。」他將其框定為從固定資料中產生合成資料 — 一種強迫資訊進入自己腦袋的工具，因為「如果你不理解，就當不好導演」。這是這個模式的開創者親自運用這個模式，也正是本資料庫的直接血脈。
關鍵語錄#
- 「你可以外包你的思考，但無法外包你的理解。」
- 「我們不是在打造動物，而是在召喚鬼魂。」（鋸齒狀的、統計式的、被召喚而來的實體）
- 「vibe coding 是把下限拉高……agentic engineering 是守住專業軟體先前所維持的品質門檻。」
- 「為什麼還有人在告訴我該做什麼？我什麼都不想做。我到底該複製貼上給我的代理什麼東西？」
- 關於 MenuGen：「我所有的 menu gen 都是多餘的……那個 app 根本不該存在。」
相關連結#
- Software 3.0 — 他的分類體系；訪談的概念主軸
- Vibe Coding vs. Agentic Engineering — 兩個術語相隔一年由他先後提出
- The Verifiability Thesis — 他試圖解釋為什麼模型呈鋸齒狀
- Jagged Intelligence (Ghosts, Not Animals) — 「鬼魂而非動物」一文的實際應用
- Agent-Native Infrastructure — 他「最有感觸的小怨念」
- Outsource Your Thinking, Not Your Understanding — 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（即 llm-wiki），並每日使用
- The Bitter Lesson — 他「神經網路即宿主進程」這類外推所立基的 Sutton 原則
- Claude Code — 他點名「cloud code / codex / open claw」作為他每日所處的 agentic-coding 介面
- Boris Cherny — 平行的「coding 已被解決」軌跡；兩人都把自己當作案例研究
資料來源#
- Andrej Karpathy: From Vibe Coding to Agentic Engineering — Sequoia AI Ascent 2026 與 Stephanie Zhan 的訪談
- llm-wiki — 他定義 LLM-wiki 模式的原始 gist
Cited by 23
- Agent Harness Engineering
Patterns for scaffolding long-running LLM agents: environment design, progressive context disclosure, mechanical archit…
- Agent Loop Pattern
`/loop` (cron-scheduled) and Ralph Wiggum (backlog-draining) loops as next-generation agent primitive; AFK execution, p…
- Agent-Native Infrastructure
The world is still built for humans and must be rewritten for agents; "what do I copy-paste to my agent?"; sensors/actu…
- 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; CLI/desktop/web/mobile/IDE sur…
- Claude Opus 4.7
GA frontier model from Anthropic; direct upgrade to 4.6 at same price; literal instruction following, 1.0–1.35× tokeniz…
- Compute Allocator
The human's evolving role: deciding what's worth spending compute on; ~1% of generated tokens ship, 99% is scaffolding…
- Disposable Micro-Apps
Throwaway custom UIs built per-task to edit a plan ("micro-software on top of micro-software"); copy-back-to-markdown;…
- Evals as Product Spec
Cat Wu's framing of evals as the emerging core PM skill: ten great evals beats a hundred mediocre; encode what done loo…
- 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…
- Jagged Intelligence (Ghosts, Not Animals)
"Ghosts not animals": jagged statistical circuits, no intrinsic motivation; car-wash/strawberry failures; stay in the l…
- LLM-as-Compiler Knowledge Base
Karpathy's architecture: LLM incrementally compiles raw docs into a persistent interlinked wiki, replacing RAG with a 4…
- MCP and Computer Use
Anthropic's two complementary connector mechanisms: MCP for structured programmatic access (Salesforce/Drive/Gmail/Slac…
- Entities — People, Orgs, Tools & Projects
Map of Content for all 32 entity pages. See Home for concept domains.
- Model Introspection Feedback
Cat Wu's underrated technique: ask the model why it failed; treat answer as harness-debugging signal not model criticis…
- Outsource Your Thinking, Not Your Understanding
"You can outsource your thinking but not your understanding"; understanding as the non-delegable human bottleneck; know…
- Printing Press Software Democratization
Boris Cherny's analogy: 1400s literacy expansion → AI software-writing expansion; domain knowledge displaces coding ski…
- Scale-Dependent Prompt Sensitivity
Large models underperform small ones on 7.7% of standard benchmarks due to overthinking; brevity constraints recover 26…
- Software 3.0
Karpathy's taxonomy: 1.0 code, 2.0 weights, 3.0 prompting; LLM as programmable interpreter; MenuGen "shouldn't exist";…
- The Bitter Lesson
Sutton 2019: scaled general methods beat hand-engineered structure; recurring justification across the wiki for dissolv…
- The Verifiability Thesis
LLMs automate what you can *verify* as computers automate what you can *specify*; RL verification rewards → jagged peak…
- Verification as the New Bottleneck
Fiona Fung: coding is no longer the bottleneck — verification, review, maintenance are; shift-left; TDD loses its tax;…
- Vibe Coding vs. Agentic Engineering
Vibe coding raises the floor (anyone builds); agentic engineering preserves the quality bar while going faster; ">10x a…
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…
- Claude Code
Anthropic's agentic coding product; created by Boris Cherny late 2024; TypeScript/React; CLI/desktop/web/mobile/IDE sur…
- Vibe Coding vs. Agentic Engineering
Vibe coding raises the floor (anyone builds); agentic engineering preserves the quality bar while going faster; ">10x a…
- Cat Wu
Head of Product for Claude Code and Cowork at Anthropic; primary articulator of AI-native product cadence and engineer-…
- Compute Allocator
The human's evolving role: deciding what's worth spending compute on; ~1% of generated tokens ship, 99% is scaffolding…