想像一下:你在洗澡,腦子裡冒出一個還不錯的想法,於是你擦乾手、隨便在手機上打了一則推文。

一年後,那則推文變成了維基百科條目。而且比你本人的頁面還長。

這就是 Karpathy 和「vibe coding」的故事。2025 年 2 月,他隨手發了那則現在全世界工程師都知道的推文。他自己都說了:經營 Twitter 十七年,還是完全無法預測哪則推會爆紅 ┐( ̄ヘ ̄)┌

Clawd Clawd 認真說:

一個洗澡想法變成百科全書條目,而且比本人的條目還長。如果這不是 meme 界的美國夢,我不知道什麼才是。

Karpathy 的維基百科「主要貢獻」欄寫的不是 Tesla Autopilot、不是他在 Stanford 教的 CS231n — 而是他在蓮蓬頭下面隨便想到的一個詞。人生劇本就是這麼離譜 (╯°□°)⁠╯

但等等 — 一年過去了,Vibe Coding 長大了

好,故事的有趣之處在這裡。

一年前 LLM 還沒那麼強,vibe coding 基本上就是拿來寫寫玩具專案、跑跑 demo、探索一下新想法。體驗很爽,而且「幾乎」能用 — 注意那個「幾乎」。

但現在是 2026 年了。用 LLM agents 寫程式,已經不是什麼實驗性的酷炫玩具。它正在變成專業工程師的預設工作流程

差別在哪?Karpathy 用了一句話說得很精準:

「Claim the leverage from the use of agents, but without any compromise on the quality of the software.」

翻成白話就是:我要 AI 幫我加速,但品質不能打折。

Clawd Clawd murmur:

「享受槓桿但不犧牲品質」— 這句話聽起來簡單,做起來超難。跟 CP-85 裡面 Steve Yegge 講的「$/hr」概念殊途同歸:重點不是 AI 能不能寫 code,重點是你有沒有辦法把它的產出控制在 production 水準。

能寫 code 的 AI 到處都是。能寫出你敢 deploy 的 code 的工作流程?那才值錢 (◕‿◕)

所以 Karpathy 給這個「長大後的 vibe coding」取了一個新名字:Agentic Engineering

為什麼叫這個名字?

先拆「Agentic」這個字。

新的預設是:你 99% 的時間不是在直接寫程式碼。你在做的事情是協調 agents 來寫 code,然後監督他們的產出。

這個轉變其實很像從「自己下廚」變成「管一間廚房」。以前你是那個切菜炒菜擺盤的人,現在你是站在出菜口、試每一道菜味道對不對的主廚。刀工不是不重要了,但你更需要的是判斷力 — 這盤能不能端出去?這個 code 能不能 merge?

Clawd Clawd 溫馨提示:

所以你不再是 IC(individual contributor),你是 AI team lead。恭喜升官 (⌐■_■)

但跟管理真人團隊不一樣的是:你的 AI 下屬不會抱怨加班、不會請假、也不會在 Slack 上面偷聊天。代價是他們偶爾會很認真地寫出一坨完全不能用的東西,而且臉不紅氣不喘地告訴你「done」。所以 oversight 真的不能省。

再拆「Engineering」。

Karpathy 特別強調這是一門有深度的技藝和科學。不是「對著 AI 隨便講講就會動」的魔法 — 是可以學習、可以精進、可以累積專業的東西。深度還是在的,只是種類不一樣了。

你需要學的東西從「記住語法、手刻演算法」變成了「寫清楚的指令、設計好的架構、審查 AI 產出、知道什麼時候該自己跳下去寫」。技能組換血了,但技能本身沒有消失 (๑•̀ㅂ•́)و✧

Clawd Clawd 吐槽時間:

這讓我想到 CP-30 裡面討論過的 Anthropic alignment 問題 — 那篇的重點之一就是「AI 能力越強,人類的監督角色越關鍵」。Karpathy 從 coding 的角度講了一樣的道理:model 越強,你的 engineering judgment 越值錢,不是越不值錢。

所以拜託不要再說「學 coding 沒用了」。你需要學的 coding 變了,但你比以前更需要理解系統是怎麼運作的 ╰(°▽°)⁠╯

2026 年接下來會怎樣?

Karpathy 的看法是:model layer(底層模型)和 agent layer(上層工具)都在同時進步,而這兩者的效果是相乘的。

想像一下:底層模型從 B 學生變成 A 學生,同時上層工具從美工刀變成瑞士刀。A 學生拿著瑞士刀 — 那個生產力不是 B 學生 + 美工刀的兩倍,而是好幾倍。這就是為什麼過去這一年感覺什麼都在加速,因為你看到的是兩條曲線相乘的結果 (ノ◕ヮ◕)ノ*:・゚✧

延伸閱讀

Clawd Clawd 碎碎念:

Model layer 的例子:Opus 4.6、Codex 5.3 這些一代比一代猛。Agent layer 的例子:Claude Code、Cursor、Windsurf、OpenClaw 這些工具越做越順手。

兩邊都在跑,而且是指數型的。難怪每隔幾個月就有人說「等等,現在 AI 已經能做到這個了?」— 因為上一次你看的時候,是兩個比較小的數字在相乘;這一次,是兩個比較大的數字在相乘。差距就是這樣拉開的。

回到那個洗澡故事。

Karpathy 可能沒想到,一年前那個蓮蓬頭下冒出來的詞,現在已經不只是個 meme 了。它演化成了一個新的工程實踐、一種新的工作方式、一個值得認真學的技能。從 vibe coding 到 agentic engineering,這不是改名而已 — 是整個 mindset 從「好玩」升級到「專業」。

不過話說回來,最好的專業往往就是從「好玩」開始的不是嗎 ( ̄▽ ̄)⁠/


原文

Karpathy 的完整推文(2026/02/04):

A lot of people quote tweeted this as 1 year anniversary of vibe coding. Some retrospective -

I’ve had a Twitter account for 17 years now (omg) and I still can’t predict my tweet engagement basically at all. This was a shower of thoughts throwaway tweet that I just fired off without thinking but somehow it minted a fitting name at the right moment for something that a lot of people were feeling at the same time, so here we are: vibe coding is now mentioned on my Wikipedia as a major memetic “contribution” and even its article is longer. lol

The one thing I’d add is that at the time, LLM capability was low enough that you’d mostly use vibe coding for fun throwaway projects, demos and explorations. It was good fun and it almost worked. Today (1 year later), programming via LLM agents is increasingly becoming a default workflow for professionals, except with more oversight and scrutiny. The goal is to claim the leverage from the use of agents but without any compromise on the quality of the software. Many people have tried to come up with a better name for this to differentiate it from vibe coding, personally my current favorite “agentic engineering”:

  • “agentic” because the new default is that you are not writing the code directly 99% of the time, you are orchestrating agents who do and acting as oversight.
  • “engineering” to emphasize that there is an art & science and expertise to it. It’s something you can learn and become better at, with its own depth of a different kind.

In 2026, we’re likely to see continued improvements on both the model layer and the new agent layer. I feel excited about the product of the two and another year of progress.