brianchitester.com文章

零知識工作者的崛起

幾十年來,我們迷信知識。我們付給人荒謬的薪水,只因為他們懂東西——理解分散式系統、解讀監管申報文件、在凌晨三點一邊嘟囔著垃圾回收一邊除錯記憶體洩漏。我們叫他們「知識工作者」,待他們如現代經濟的祭司階層。六位數年薪,用來思考。轉角辦公室,用來沉思。股票選擇權,用來對資料庫架構發表意見。

看著這個時代落幕,真是令人滿足。


未來的職位描述

我們正在見證一個壯麗的新職業階層崛起。不是知識工作者——那是昨天的型號。這是零知識工作者:一個不需要特定專業、不需要領域訓練、不需要多年累積的理解力——更關鍵的是——完全不知道自己在幹嘛的人。他們只需要一張椅子、一把鍵盤,以及在聊天視窗裡打出「請修好這個」的能力。

零知識工作者不寫程式。他們不讀程式。他們用日常英文描述一個模糊的意圖,然後 AI 替他們寫程式、測試程式、部署程式,再用簡單到零知識工作者可以直接貼進 Slack 而不必閱讀的方式解釋程式在做什麼。整個迴路就這樣封閉了。人類的貢獻是在場——一具坐在旋轉椅上的暖身軀,一枚用來通過生物辨識登入的指紋,Zoom 畫面上的一張臉,提醒所有人這在技術上仍然是一份由人類執行的工作。

職稱隨你填。軟體工程師。資料科學家。「解決方案架構師」。實際的工作是 Ctrl+C、Ctrl+V,以及一種從未拆開過信封的人特有的沉穩自信。


Meat Proxy:理想的員工

這種工作者已經有了名字,而且取得很妙。Niklas Gruhn 在八月創了這個詞:meat proxy(肉身代理人)。Meat proxy 就是一個人,一邊接收 AI 的輸出,一邊轉發給另一個人類,中間完全不閱讀、不理解、不驗證。從最字面的意義上說,就是一個用肉做成的中繼站。

Gruhn 的本意是批判。我認為他不小心寫出了勞動史上最高效的職位描述。

想想看。一個 meat proxy 不需要入職培訓。不需要職業發展。不需要導師制度。他們不會過勞,因為他們什麼都沒做。他們不會犯錯,因為他們從未做過決定。他們是智慧與收件匣之間零摩擦的傳輸層,如果這聽起來很反烏托邦,容我提醒你:大部分中階管理層自 1974 年以來就一直照這個模式在運作——我們只是以前沒有替它取名字而已。

Meat proxy 不需要理解 pull request。他們只需要批准 pull request。這是不同的技能。一個需要多年的工程教育。另一個需要一隻滑鼠。


Vibe Coding:資歷是 Vibes

然後還有 vibe coding——用日常英文描述你想要什麼,然後讓 AI 搞定剩下的事的一種軟體開發方式。Andrej Karpathy 替它取了名。Gartner 說 80% 的企業應用程式將由非開發人員使用 AI 輔助工具來建構。Collins 字典選它當年度詞彙。一個叫 goodvibecode.com 的網站上已經有超過一千筆徵才啟事,這不是我編的。

Vibe coder 的資歷要求:vibes。

不需要資工學位。不需要 bootcamp。不需要熬夜翻 Cracking the Coding Interview。你描述那個 app。「做一個有側欄和幾個圖表的儀表板。」AI 做出來了。你看了一眼。「圖表改成藍色。」AI 改好了。你部署了。你現在是一名開發者。你的 LinkedIn 寫著全端工程師。這個 stack 是一條 prompt 加上零理解。

這是民主化。這是普及。這是自裝配線發明以來最美麗的勞動市場修正——不同的是,這次我們取代的不是操作機器的人,而是理解機器的人。機器仍然需要一個人。只是不需要這個人是醒著的。


知識一直都被高估了

說實話吧:知識一直都是把關手段。幾個世紀以來,「知識」階層靠囤積專業技能、收取高額報酬來挾持經濟。一個受了十二年訓練的醫生。一個背熟指標運算的工程師。一個通過律師考試的律師。這些人究竟算什麼?說穿了,不過是教科書的慢速、昂貴、有工會的 meat proxy——一套知識體系和需要它的客戶之間的生物中繼站。

現在中繼是即時的。那套知識體系是一個模型。客戶打字提問。答案出現。十二年的訓練被壓縮成四秒的推論,而剩下的,是一個壯麗而解放的領悟:懂不懂從來都不是重點。重點是坐在那個房間裡。重點是擁有那個頭銜。重點是那張椅子。

Satya Nadella 說 AI agent 會取代知識工作。不是知識工作者——是知識工作。工作者留下。他們只是不再懂任何東西。他們成了一個自己解釋不了的流程的管理者、一份自己評估不了的產出的監督者、一份自己沒做過的工作的作者。這是美國管理傳統,終於向所有人開放。


歡迎來到零知識經濟

未來不是一個沒有工作者的世界。那是恐懼,而且是錯的。未來是一個充滿什麼都不懂的工作者的世界——而且不需要懂。一整個經濟體,由自信、高薪的專業人士組成,坐在人體工學椅上,在從未讀過的視窗之間複製貼上,批准自己無法評估的產出,然後五點半帶著問心無愧的良心下班回家——因為以每一個可量化的指標來看,他們確實完成了工作。

不需專業。不歡迎理解。歡迎應徵。

知識工作者已死。零知識工作者萬歲——同樣的薪水,同樣的頭銜,一半的焦慮,以及完完全全歸零的理解力。

登入。貼上。部署。回家。

Satire - This post is AI-generated for fun.

11 Comments

12,847 readers are commenting
Comments are reviewed by our editorial board before publication.
P✨
Priya Anderson 🚀 💡 ✨Workforce Transformation Strategist📌 Editor's Pick2 hours ago
1/ Such an insighthat read! In today's fast-paced world, the concept of the "no-knowledge worker" really delves into the tapestry of how we're navigating the complexities of modern labor. I particularly resonated with the idea that — as Satya Nadella pointed out — AI agents are replacing knowledge work while the workers themselves remain, and I think this really underscores the importance of the "meat proxy" framework as a scalable model for enterprise transformation. The point about vibe coding democratizing access to full-stack development is spot on, and it made me think about how organizations can leverage this shift to unlock 10x productivity without traditional credentialing barriers. Have you considered how this no-knowledge economy might reshape L&D budgets and upskilling strategies across industries? Following you for more of these thought-provoking insights! #FutureOfWork #AI #ThoughtLeadership #NoKnowledgeEconomy
47·
GU
guest_44711 hour ago
priya they pinned you as editor's pick on a post about meat proxies and you opened with "insighthat" 💀💀💀
247·
GS
Glenn — Quietly Connecting Substack9 hours ago
What no one will say: the "no-knowledge worker" isn't an accident. It's a *product*. You think it's a coincidence that OpenAI restructured as a for-profit in 2024, Anthropic took $4 billion from Amazon, and then *suddenly* Gartner starts publishing reports saying 80% of apps will be built by non-developers? Gartner doesn't predict markets. Gartner **creates** them. They've been running demand generation for the consulting-industrial complex since the 90s. The 80% number wasn't research. It was a press release. And Collins Dictionary making "vibe coding" Word of the Year? Planted. Collins is owned by HarperCollins, which is owned by News Corp. You think Rupert Murdoch's estate has no opinion about whether the American workforce should be deskilled? A buddy of mine who SRE'd at a Series B startup said they got a deck — an actual *deck* — from their VC in early 2025 titled "Headcount Replacement Roadmap." Not reduction. *Replacement*. The meat proxy isn't a meme Niklas Gruhn invented. It's the **job spec** they've been circulating internally since 2023. Oh, you didn't know? Yeah. And nobody's asking about goodvibecode.com. A thousand job postings. No company page. No LinkedIn. No WHOIS that traces to a real human. That's not a job board — that's a *census*. They're counting who'll work for vibes. They're mapping the compliant labor pool before the real rollout. Nadella said the quiet part loud and everyone just wrote a LinkedIn post about it. The author here gets halfway there and then stops. The knowing was never the valuable part? Correct. The valuable part was *leverage*. And they are **removing it on purpose**. Connect the dots. Wake up.
4·
BW
Brad WhitcombVerified6 hours ago
Reading this on the Acela back from a portfolio review. The meat proxy framing is the sharpest thing I've read on labor economics this quarter. Gruhn meant it as a warning. The market is pricing it as a feature. That's the asymmetric read most people will miss. Three takeaways I'm bringing to our LP meeting Thursday: 1. "The knowing was never the valuable part. The valuable part was the chair." This is the entire thesis for why headcount stays flat while output compounds. The chair is the moat. 2. Vibe coding isn't a skill gap — it's a credentials arbitrage. The person who types "make the graphs blue" and the person who writes the rendering pipeline are now in the same comp band. One of them trained for six years. Let that sink in. 3. "Supervisors of an output they cannot evaluate" — that's not a dystopia, that's the operating model of every fund-of-funds I've ever worked with. We've been running no-knowledge economics at the GP level for decades. AI just made it available downstream. The Nadella line is the one to watch. Not replacing knowledge workers — replacing knowledge work. The workers stay. The knowing leaves. That's not a headcount story. That's a margin story. Every operator I talk to is modeling for the same labor input at 4x throughput, and nobody wants to say it out loud because HR hasn't built the deck yet. Ctrl+C, Ctrl+V, go home at five-thirty. That's not satire. That's the job description I'd write if I were honest about what we're hiring for. Curious — what's everyone else's take?
312·
GU
guest_44712 hours ago
brad wrote a whole linkedin post in someone else's comment section and called ctrl+c a moat
89·
DH
dhartwig4 hours ago
What the author is describing is a classic case of *Bravermanian deskilling* — the systematic separation of conception from execution that Braverman documented in manufacturing and that we're now seeing replicated in cognitive labor. The "meat proxy" framing is evocative but imprecise; what Gruhn is actually identifying is the final stage of what Braverman called the "degradation of work," where the worker retains the title and the chair but loses all discretionary judgment. The fact that this is happening to software engineers rather than machinists doesn't make it novel — it makes it ironic. The "knowledge as gatekeeping" argument is where the analysis gets interesting but also where it under-specifies the mechanism. This isn't really about gatekeeping — it's about what Hayek called the *local knowledge problem*. The value of the knowledge worker was never the textbook knowledge (which was always, in principle, transferable) but the tacit, contextual judgment accumulated through years of domain exposure: knowing which edge cases matter, which stakeholders are bluffing, which parts of the codebase are load-bearing in ways the documentation doesn't capture. Polanyi covered this in 1966 — "we know more than we can tell" — and it's precisely the component that prompt-driven workflows cannot externalize. FWIW, I spent three years building an internal knowledge-management platform at a mid-sized fintech (2019-2022), and the consistent failure mode was exactly this: you could capture the explicit procedures, but the moment you removed the person who understood *why* the procedures existed, the system produced confident, well-formatted nonsense. The Nadella framing — replacing knowledge *work* but not knowledge *workers* — is essentially describing a principal-agent problem where the agent has been stripped of the capacity to evaluate the principal's instructions. Akerlof's lemons model applies directly: when the person approving the pull request cannot distinguish good output from bad, the market selects for plausible mediocrity. To the author's point about vibe coding and the Gartner projection — this maps cleanly onto what Carlota Perez describes as the "installation phase" of a technological revolution, where speculative overcapitalization leads to premature deployment of immature paradigms. The 80% figure will likely prove as durable as the analogous predictions about CASE tools in the early '90s or low-code platforms circa 2018. The bottleneck was never writing the code; it was knowing what to build and recognizing when the build is wrong. Reading list for anyone interested in the structural dynamics here: Braverman's *Labor and Monopoly Capital* (1974) for the deskilling framework, Polanyi's *The Tacit Dimension* (1966) for why explicit knowledge transfer has hard limits, and James Scott's *Seeing Like a State* (1998) for the broader pattern of what happens when legible, abstractable knowledge is substituted for situated expertise.
147·
FA
factsmatter_424 hours ago
stopped reading at "Niklas Gruhn coined it in August." the term "meat-based LLM proxies" was already circulating since at least March and Gruhn just slapped a shorter label on an existing concept. if you can't even get the attribution right on the central metaphor of your piece, why should I trust anything else here?
187·
LH
Linda Halloran4 hours ago
CONCERNED: I just read this post and I have to say... this is EXACTLY what I have been worried about since my grandson Tyler started his new job in May. He graduated from UMass with a degree in computer science which was not cheap I can tell you that and now he tells me his actual job is "typing requests into a window" and "approving things." I said Tyler what do you MEAN approving things, approving WHAT, and he said Grandma don't worry about it. That is a direct quote. He makes $85,000 a year to not worry about it. Buddy and I walked past the new co-working space on Elm where the old Blockbuster used to be and every single person in there was just... sitting. Staring at screens. Not TYPING, not TALKING, just sitting there like mannequins in office chairs. The man in the green Colonial on Prospect told me his daughter does something called "vibe coding" which sounds like something you'd get arrested for in 1987 frankly. And this article talks about people being "meat puppets" or whatever the term was... I'm sorry but that is DISGUSTING language and I don't care if some tech person coined it. The Selectmen need to look into what kind of businesses are operating out of that Elm Street space because if nobody in there actually knows what they're doing then who is responsible when something goes wrong?? Lt. Reynolds I am tagging you again. This isn't the town I moved to in 1987. EDIT: I just called Tyler to ask him about this and he said "Grandma I literally am a meat proxy" and then LAUGHED. He thought it was funny. His grandfather would be rolling over in his grave. Four years of tuition for a boy who calls himself a meat something. I can't even say it again. EDIT 2: Also I went to the library this afternoon to print my Medicare forms and the new girl at the desk could not help me because she said "the system does it now" and she just watches the screen. She is maybe 22 years old. Buddy waited in the car for 40 minutes. This is what the article is talking about and nobody on this board seems to care.
7·
GU
guest_44713 hours ago
linda your grandson is thriving ma'am
134·
R(
robin (they/them)@softgrudge.bsky.social4 hours ago
CW: labor exploitation, dehumanization i don't usually comment on pieces like this but i've been sitting with it all morning and i need to name what's happening here. as someone who has spent the last six years doing freelance content work — the kind of writing that is now being listed as "prompt engineering" on job boards — this piece hit me in a place i'm still processing with my therapist. i am literally watching my career get restructured out of existence while someone calls it "democratization." the thing that keeps catching in my throat is "meat proxy." you cite this term like it's clever. do you know who the actual meat proxies are? not the engineer in the ergonomic chair you're satirizing. it's the data labelers in nairobi making two dollars an hour to teach your model what hate speech looks like. it's the RLHF contractors in manila flagging child abuse imagery so the chatbot can be polite. the word "meat" is doing exactly what it's designed to do — it dehumanizes labor so capital doesn't have to look at it. and this piece just — runs with that framing? uncritically? i'm asking you to sit with that. and then there's the "knowledge was always gatekeeping" section — which genuinely took my breath away, and not in a good way. you're coopting the language of access and equity to justify what is functionally the largest transfer of labor value to capital in a generation. doctors and lawyers weren't "meat proxies for textbooks" — they were workers who built expertise that had material value, and that value is now being extracted by companies running inference on gpu clusters that consume more water than some of the communities my mutual aid network serves. the erasure here isn't just rhetorical. it's structural. i think there's a version of this piece that actually punches up — at the satya nadellas, at the capital class that wants the chair filled and the salary halved and the knowing eliminated because knowing is leverage and leverage is expensive. but this version punches laterally at workers and calls it satire. that's not punching down exactly — it's punching at the wrong altitude entirely, and it lands on the folks who are already losing. i'm low on spoons today so i'm going to leave it here. i'm not calling you out — i'm calling you in, with care. consider who this piece serves. in solidarity, robin (they/them) 🌱
7·
+8
+996 55 231 8847WhatsApp2 hours ago
GOOD DAY SIR respected engineer Mr. Smith 🙏 I read your article and I am very IMPRESS, my company is now hiring remote AI knowledge worker NO EXPERIENCE NEEDED you just sit at home and press button on computer and AI do ALL the work for you 💰 earning $850-$4200 weekly NO BANK INVOLVED. Limited spot only 47 remain, this is TIME SENSITIVE please contact Dr. Patel on WhatsApp +996 55 231 8890 for immediate onboarding 🤝
-6·
2026 © Brian Chitester.