(圖說：還沒十二點。晚上九點於皇居最外圍沒有別人的護城河。偶爾理解了別人不理解的事，而嘗試著各種保存記憶的魔法。。圖片來源：Ernest。)
每次看到「某公司裁員 XX%」的新聞，我最想知道的不是為什麼砍，而是：砍完之後，留下來的人怎麼工作？組織長什麼樣子？他們調整了什麼流程？
Block ($XYZ) 於 2025Q1 裁了超過四成的人，而且公開說 AI 是關鍵因素。他們的業務負責人 Owen Jennings 在 a16z 頻道分享了具體做法。Owen 在 Block 待了 12 年，之前是 Cash App 的 CEO，現在負責橫跨 Square、Cash App 和 Afterpay 的產品營運和客戶支援。他講的不是泛泛的「我們擁抱 AI」，而是實際的重建邏輯。
這裡面有三個值得拆解的原則：
✳️ 先劃出不能動的線
Owen 說他們不是拿一個裁員目標數字然後「砍到那個數字」。他們花了 Q1 整季和創辦人 Jack Dorsey 討論的問題圍繞著：假設 AI 工具已經是某某樣子的水準，公司應該長什麼樣子？
但在回答這個問題之前，先定義底線：底層系統可靠性是最高優先級（P0 中的 P0），大規模重組之後不能對客戶發生任何服務中斷。合規團隊幾乎完整保留，金融業的監管環境沒有試錯空間。路線圖上已經確定的功能要持續推進。先把不能動的線畫出來，其他的才開始重新設計。
（我們前幾年在幫傳統產業客戶做流程整合時也學到類似的事：在導入任何自動化之前，先確保原本可獲利的核心工作流程可以透過 API 執行，並且保留完整的歷史紀錄以滿足合規監管。不是先想著買工具，而是要先把路鋪好、把人心準備好，再安心上路。Block 從 2024 年初就開始建 Goose 這個 agent 平台，兩年多的基礎工程是他們能大幅重組的前提。沒有這段基礎鋪路，裁員就只是砍人。）
（如果有亞洲金融業朋友也有想要大幅度重組，也許咱們到後台聊聊？）
✳️ 從組織形狀開始重建，不是從裁員數字
Block 不是在既有架構上「減人」，而是重新問：做這件事需要幾個人？以前一個功能團隊 14 人，現在可能是 4 個人加上兩千美元的 token 預算和無限的 AI 工具使用權。
具體的組織變化：
- 開發團隊重組為 1 到 6 人的小隊（Squads）
- 管理層級砍了五到六成，產品線大多只剩兩到三層
- 會議量減了七到八成
- 小隊可以跨產品線快速切換，不再綁定一個團隊好幾年
- 設計師和 PM 現在也在送 PR，不只是工程師的事
Owen 選擇一次性大幅調整，而不是漸進式。他說分批做 15% 裁員，文化上是毀滅性的，因為每個人都在等下一輪什麼時候來。
（我們團隊過往這幾年陪著客戶進行組織流程梳理、重整，提出的各種直球問題也是對應到各種的減法，不斷挑戰著管理團隊：需要做這件事情嗎？做這件事需要幾個人？這件事情出現在整個系統流程的哪個環節？如果拿掉這個環節會發生什麼？可以做 A/B 測試實驗看看嗎？寫得出 SOP 步驟嗎？一定需要人類介入嗎？介入的反應速度可以加速十倍嗎？如果產出變成百倍，營收增加十倍，但品質下降 5% 可以接受嗎？）
（這幾年真心感謝被我們蹂躪過的客戶們，還是對我們充滿信任。）
✳️ 寫一份你公司的 Markdown 文件
Owen 在結尾描述了一個讓我印象深刻的模型。他說你可以想像每家公司有一份 markdown 文件，寫著自己是誰、在意什麼指標、不在意什麼、優化目標是什麼。然後 agent 系統根據這份文件不斷建造和迭代。以前做一個功能要幾個月，現在一兩週，未來可能一天跑幾百次循環。
這不只是比喻。Block 內部在建立「世界模型」（World Models），理解客戶如何參與經濟活動，也理解公司本身的運作方式。然後用自主式工具持續在這個理解上迭代。
（這類似去年 Kiro 提出的 Steering 功能、Claude 提出的 Skills 都在做類似的事：把團隊的工作方式和知識寫成文件，AI 在執行任務前會先讀這些文件。這是「可執行的架構文件」，過去文件經常跟實際程式碼脫節，但當文件直接影響 AI 的行為時，它就有了真正的約束力。新人或 AI Agent 進來，先閱讀文件就（有機會）能理解團隊怎麼工作。Owen 說的那份 markdown 文件，就是這個概念的企業規模版。）
（用小說來譬喻的話，就是先介紹出場角色、然後鋪陳一下開場的時空背景與場景、陸續帶出場場景中發生的任務目標，將這些紀錄在一份格式稱為 markdown 的文件中，然後交給 agent 展開故事。）
Owen 的核心論點是：未來的護城河不是人多，是你的公司理解什麼別人不理解的事，然後能多快迭代這個理解。
✳️ 延伸閱讀
✳️ 知識圖譜
(更多關於知識圖譜…)
✳️ 逐字稿與筆記
開場介紹與講者背景
- The biggest moat is going to be which companies understand something that’s super hard for other people to understand.
最大的護城河將是哪些公司能理解其他人很難理解的事情。 - And if your answer to that is I don’t know, then you maybe could get vibe coded away.
如果你的答案是「我不知道」，那你可能會被 vibe code 取代。 - » Block was one of the first to make a pretty drastic decision in cutting 40% of the workforce.
» Block 是率先做出激烈決定、裁撤 40% 員工的公司之一。 - What led up to that decision?
是什麼導致了這個決定？ - » There’s been this correlation between the number of folks at a company and the output from the company for decades and decades.
» 數十年來，公司的員工人數與公司產出之間一直存在著相關性。 - I think that basically broke and what we were seeing is that one or two engineers who was on the tools is able to be 10 20 100x more productive over time.
我認為這個關聯基本上已經被打破了，我們看到的是一兩個使用這些工具的工程師，生產力可以提升 10 倍、20 倍、甚至 100 倍。 - It’s like pretty obvious that these systems are just going to be so much better than like having a thousand humans who are doing that work.
很明顯，這些系統將會比讓一千個人類做這些工作來得好太多了。 - I I do believe that fundamentally for a given product or for a given road map, you’re going to need fewer engineers, fewer designers, fewer PMs.
我確實相信，從根本上來說，對於一個特定的產品或路線圖，你將需要更少的工程師、更少的設計師、更少的 PM。 - I think that’s like very very clear.
我認為這非常非常清楚。 - » So you show up on Monday, 40% of the company’s gone.
» 所以你星期一到公司，40% 的人都不在了。 - What’s the most meaningful difference in how you’re operating?
你們在運作方式上最有意義的差異是什麼？ - » I think the biggest thing is what does it actually look like for a large public company to restructure itself around AI?
» 我認為最大的事情是，一家大型上市公司圍繞 AI 進行重組，實際上看起來是什麼樣子？ - Owen Jennings is the business lead at Block where he oversees product operations and customer support across Square, Cash App, and Afterpay.
Owen Jennings 是 Block 的業務負責人，他負責監督 Square、Cash App 和 Afterpay 的產品營運和客戶支援。 - Before this role, he was the CEO of uh of Cash App during its critical scaling period.
在擔任這個職務之前，他是 Cash App 在關鍵擴張期的 CEO。 - And recently uh block executed a roughly 40% reduction in force and they’ve been pretty candid about AI being a critical component of that decision.
最近，Block 執行了大約 40% 的裁員，而且他們對 AI 是該決定的關鍵因素一事相當坦率。 - Owen has gone through the AI transformation at scale across product lines and business units and so we’re going to dig into the that decision around the riff how block has adapted the current and future state of the business.
Owen 經歷了跨產品線和業務部門的大規模 AI 轉型，所以我們將深入探討這個裁員決定，以及 Block 如何適應業務的現狀和未來。 - So thank you so much Owen.
非常感謝你，Owen。
Block 以 AI 驅動的裁員決策
- Welcome to the stage.
歡迎上台。 - Thanks.
謝謝。 - Awesome.
太好了。 - Um, so you know, Jonathan, I think did an amazing job kind of setting the stage, you know, for this conversation.
Jonathan 為這場對話做了很棒的鋪墊。 - Uh, you know, talking about how important it is to be founderled.
談到了創辦人主導有多重要。 - Uh, you know, Block was one of the first to make a pretty drastic decision in cutting 40% of the workforce.
Block 是率先做出裁撤 40% 員工這個相當激烈決定的公司之一。 - Um, maybe walk us through kind of what led up to that decision and how you thought about it.
也許可以帶我們了解一下是什麼導致了這個決定，以及你是怎麼思考的。 - Sure.
好的。 - I I think I would pro it probably starts two or three years ago.
我想這大概要從兩三年前說起。 - I think one thing about Jack is I I find Jack to be generally right and generally early.
關於 Jack，我發現他通常是對的，而且通常很早。 - Uh sometimes very early.
有時候非常早。 - Um and I think that’s flowed through Twitter, Square, Cash App, Bitcoin, etc.
我認為這在 Twitter、Square、Cash App、Bitcoin 等方面都得到了驗證。 - And so we were pretty early on the agentic development side.
所以我們在自主式開發方面起步相當早。 - We actually launched Goose, which was the first agent harness, at least that I know of, um, in early 2024.
我們實際上在 2024 年初推出了 Goose，據我所知這是第一個 agent 框架。 - And that started to augment how we approached software development, uh, how we thought about internal tooling.
這開始增強了我們做軟體開發的方式，以及我們如何思考內部工具。 - And I would say that over the over that period 24 and 25, it was like pretty meaningful progress.
我會說在 2024 和 2025 這段期間，進展相當顯著。 - Um, and then late November, first week of December, it was just there was a binary change.
然後在 11 月底、12 月第一週，出現了一個二元式的轉變。 - you basically have Opus 46, you have uh codeex 53 and essentially you get this shift where I think the the the tools and the foundational models were pretty good at writing code especially for new ventures and kind of like green space.
你有了 Opus 4.6、Codex 5.3，基本上出現了一個轉變，這些工具和基礎模型在寫程式方面已經相當不錯了，特別是對於新創事業和全新的領域。 - Um it became clear almost overnight maybe in a couple of weeks that now they’re incredibly capable working with existing complex code bases.
幾乎一夜之間，也許幾週之內，很明顯地它們現在在處理現有的複雜程式碼庫方面也變得非常強大了。 - Um and so there was a massive paradigm shift where at least from my perspective there’s there’s been this correlation between the number of folks at a company and the output from the company uh for you know decades and the first week of December and what we were seeing is that one or two engineers or a designer and an engineer who was on the tools quote unquote as we say is able to be 10 20 100x more productive.
所以出現了一個巨大的典範轉移，至少從我的角度來看，數十年來公司的員工人數與產出之間一直存在相關性，然後在 12 月的第一週，我們看到的是一兩個工程師，或是一個設計師和一個工程師，用我們的話說是「在使用工具的」，生產力可以提升 10 倍、20 倍、100 倍。 - And so that’s really what led us to make the the decision a few weeks ago.
所以這就是幾週前我們做出這個決定的真正原因。 - We spent Q1 discussing like what does this mean fundamentally?
我們花了整個 Q1 討論這從根本上意味著什麼。 - What does this mean in terms of how we’re going to build products, how we’re going to build software for customers, and then also um how we’re going to run a company.
這對我們如何打造產品、如何為客戶開發軟體意味著什麼，然後也包括我們要如何經營一家公司。 - What is it going to mean to actually run a company?
實際經營一家公司將意味著什麼？ - And we spent Q1 as an executive team uh with Jack um working through that.
我們整個 Q1 作為管理團隊和 Jack 一起研究這個問題。 - Uh and ultimately that’s what led us to this place where where we we did a reduction in force that was you know slightly greater than than 40%.
最終這就是讓我們走到執行了略超過 40% 裁員的這一步。 - And that wasn’t even uh you know to the to the conversation we were just having the tools were flowing through really meaningfully on the development side and so the cuts were way larger on the development side.
而且正如我們剛才對話中提到的，這些工具在開發端的影響非常顯著，所以開發端的裁員幅度遠大於其他部門。 - If you think of something as outbound sales or account management um the cuts were you know fairly dimminimous.
如果你看像外向型業務或客戶管理這類的，裁員幅度是相當小的。 - Um and so that was really what we were reacting to.
所以這就是我們在回應的事情。
裁員的合理性論證
- C can I push you a bit on this a little bit?
我可以在這點上追問一下嗎？ - I mean, Alex when he kind of introduced the, you know, the conference uh just, you know, an hour ago talked about the zer period.
Alex 大約一小時前介紹會議的時候，提到了 ZIRP（零利率時期）。 - Uh, you know, how much of the riff was sort of overhang from 2021 kind of overhiring versus AI and and kind of like the product actual productivity gain is going to be in the business?
這次裁員有多少是 2021 年過度招聘的遺留問題，又有多少是 AI 帶來的實際生產力提升？ - Like if you look at where we were from a from a gross profit per full-time employee basis from like 2019 through 2024, we’re basically like right in the middle of the pack with all of the um uh with all the competitors.
如果你看我們從 2019 到 2024 年按每位全職員工毛利來算的表現，我們基本上處於所有競爭對手的中間位置。 - Um if you look at last year, I think we were kind of I don’t know second quintile or something like that.
如果看去年，我想我們大概在第二個五分位之類的。 - I think it’s basically like Nvidia and Meta that are ahead of us.
基本上就是 Nvidia 和 Meta 在我們前面。 - Um, and then when you look at the composition of what we did, if you thought it was like croft and bloat and so on and so forth, then like this riff would have acred to the operational teams and like like that sort of stuff.
然後當你看我們裁員的構成，如果你認為這只是消除冗員和臃腫，那麼這次裁員應該會集中在營運團隊那類的。 - Those were really really meaningful cuts on the development side.
但那些是在開發端非常非常顯著的裁減。 - You don’t make really really significant cuts on the development side if you’re not seeing a technology and a tool that’s just fundamentally changed how we build.
如果你沒有看到一種從根本上改變我們建造方式的技術和工具，你不會在開發端做出如此顯著的裁減。 - I mean, we’re we’re like we’re not writing code by hand anymore.
我的意思是，我們已經不再手寫程式碼了。 - That’s over.
那個時代結束了。 - That’s done.
完了。 - Um and so so anyway, everyone has their narrative.
總之，每個人都有自己的敘事。 - Um it’s largely not true.
但大部分都不是事實。
執行轉型與營運變革
- » Um so maybe just walk through like tactically how did you actually execute you know this this transition you know culturally you know operationally in the business.
» 也許可以具體說說你們在文化上、營運上是如何實際執行這次轉型的。 - » So I think so we were um the the the nice part about this riff uh relative to some other you know things that have happened at block or at other companies is we were coming from a position of strength on a on a profitability and operating income side.
» 這次裁員相對於 Block 或其他公司發生過的其他事情，好的一面是我們是從獲利能力和營業收入都很強勁的位置出發的。 - And so sometimes when it’s really financially motivated, you know, the CFO or the CEO says, “Okay, we need to do a 16% riff in order to like hit this hit this target.”
有時候當裁員真的是出於財務動機時，CFO 或 CEO 會說：「好，我們需要裁 16% 才能達到這個目標。」 - And um that wasn’t the case at all.
而我們的情況完全不是這樣。 - We said, “What should the org look like given how these AI tools are flowing through now and what we expect to happen in the in the coming months and quarters.”
我們問的是：「考慮到這些 AI 工具現在的發展方式以及我們預期在未來幾個月和幾個季度會發生什麼，組織應該長什麼樣子？」 - We had some core principles.
我們有一些核心原則。 - Um the first one was reliability.
第一個是可靠性。 - When you do something this size, worst case scenario is you have an outage or you go down.
當你做這種規模的事情時，最糟糕的情況是發生服務中斷或系統停機。 - So that’s like P 00 not acceptable at all.
所以那是 P00，完全不能接受。 - Obviously, you know, things have been great over the past several weeks, which is fantastic.
顯然，過去幾週一切都很順利，這太棒了。 - Second is building trust with customers and um compliance and navigating the regulatory environment.
第二是與客戶建立信任，以及合規和應對監管環境。 - We all operate in a super complex nuanced regulatory environment.
我們都在一個超級複雜且細膩的監管環境中營運。 - That’s a non-negotiable.
這是不可妥協的。 - We have to make sure that we’re that we’re doing doing right there.
我們必須確保我們在那方面做對了。 - For instance, like we we basically did not touch our our compliance team and our compliance technology team.
例如，我們基本上沒有動我們的合規團隊和合規技術團隊。 - Even if the tools are there, it’s like let’s not take any risks.
即使工具已經到位，我們的態度是不冒任何風險。 - And then third was let’s continue to drive durable growth.
第三是讓我們繼續推動持久的成長。 - So there’s things that are on the road map that we already know that we’re building.
路線圖上有些我們已經確定要做的東西。 - We need to continue to do that.
我們需要繼續推進。 - We know that it might be a squad of three people instead of a feature team of 14 who’s building that.
我們知道做這件事的可能是一個 3 人小隊，而不是 14 人的功能團隊。 - We want to make sure we’re continuing to build those features and that we’re continuing to make longerterm bets.
我們要確保我們持續開發那些功能，並且持續進行長期投資。 - And then we built up the org from scratch.
然後我們從零開始重建組織。 - And in some areas like um the regulatory council team or the SDRBDR team, the org looked pretty similar to how it looked in January.
在某些領域，像是監管委員會團隊或 SDR/BDR 團隊，組織看起來和一月時差不多。 - Um on the development side, it looks completely completely different.
在開發端，看起來完全完全不同了。 - Um and then you know from a from an execution perspective um you know we thought very deliberately obviously I’ve been in the company 12 years.
從執行的角度來看，我們考慮得非常謹慎，顯然我已經在公司待了 12 年。 - A number of folks who we parted ways with our friends and colleagues for for you know more than a decade.
我們告別的很多人都是超過十年的朋友和同事。 - um we were in a position where we’re able to be generous in terms of you know the the severance packages that we gave.
我們處於能夠在遣散方案方面大方的位置。 - We didn’t cut people’s technology access instantly which can suck.
我們沒有立刻切斷離職員工的技術存取權限，那種做法很糟糕。 - Uh we chose to have an all hands with everybody at the company.
我們選擇與全公司的每個人開一場全員大會。 - So Jack and the executive team were um you know looking each other in the eyes and explaining this decision and explaining the the drivers behind it.
所以 Jack 和管理團隊是面對面地向大家解釋這個決定以及背後的驅動因素。 - And um I I think that that it was on a Thursday.
我記得那是在一個星期四。 - I think like the Friday, Saturday, Sunday there’s a lot of shock uh dealing with ambiguity.
星期五、六、日有很多震驚，在處理不確定性。 - Um and then what we’ve been doing is uh we massively reduced the number of meetings we have probably like 70 or 80%.
然後我們做的是大幅減少了會議數量，大概減少了 70% 到 80%。 - So I now have time to like build and work and it’s not backto-back meetings.
所以我現在有時間去建造和工作，不再是接連不斷的會議了。 - We’re also meeting with the company every week.
我們也每週與全公司開會。 - So we have like a one or two hour all hands with Jack every every Monday.
我們每週一跟 Jack 有一到兩小時的全員大會。 - It just feels like we’re we’re smaller, we’re leaner, we have fewer layers, we have larger spans, and it’s it’s been back to building.
感覺就是我們更小了、更精實了、層級更少了、管理跨度更大了，回到了建造的狀態。
AI 對營運結構的影響
- » So, you show up on Monday, 40% of of the company’s gone.
» 所以你星期一到公司，40% 的人都走了。 - Like, what how is what’s the most meaningful difference in how you’re operating?
你們在運作方式上最有意義的差異是什麼？ - I don’t know, maybe it’s in the EPD or elsewhere.
不管是在 EPD（工程、產品、設計）還是其他地方。 - » Um, I think that there’s a there’s a there’s a few different components to this.
» 我認為這有幾個不同的面向。 - I think the biggest thing is so one concern that I have with like how some of these org changes might flow through the tech industry is that and and it gets back to the to the founder point.
我認為最大的事情是，我對這些組織變革可能如何影響科技業有一個擔憂，這又回到了創辦人的那個觀點。 - If you’re not founder le and you don’t have the the ability to be bold, then you’re going to probably take a more incremental approach.
如果你不是創辦人主導的，而且沒有大膽行動的能力，那你可能會採取更漸進的方式。 - And so the way that that’s going to feel is like you do a 15% riff and it’s like, oh, it’s fine.
那感覺就像是你做一次 15% 的裁員，覺得還好。 - and then you do another 15% riff and then culturally that’s just like devastating for your team because there’s always this like pending riff looming looming over your over your shoulder.
然後你又做一次 15% 的裁員，那在文化上對團隊是毀滅性的，因為總是有一個即將到來的裁員陰影籠罩在你肩上。 - Um this was obviously a decision to go in a different direction.
這顯然是一個走向不同方向的決定。 - I think one of the benefits that we got from this is like we were already seeing a a very meaningful increase in AI tool usage especially on the development side.
我認為我們從中得到的好處之一是，我們已經看到 AI 工具使用量非常顯著的增加，尤其是在開發端。 - This is just a massive forcing function.
這只是一個巨大的推動力。 - Like if we’re building um okay, we’re building Moneybot and we want to roll Moneybot out to 50% and there used to be a team of 15 people working on it and now there’s a team of four people plus $2,000 on the tokens.
比如說我們在開發 Moneybot，想把它推送給 50% 的用戶，以前有 15 個人的團隊在做，現在是 4 個人的團隊加上 2000 美元的 token 預算。 - That’s this is like un unlimited access to tokens and you can use fast mode on Claude Code.
基本上就是無限量的 token 使用權，你可以在 Claude Code 上使用 fast mode。 - Um so now you have four people plus the tools.
所以現在你有四個人加上這些工具。 - It’s like okay well you need to have eight instances of goose up and you need to shift your workflow from sequentially working through a PR submitting it getting a review making the change to I have 14 agents who are building PRs on my behalf right now and I’m going to context switch between all of those and it’s not just uh on the software development side it’s for PMs too it’s for growth marketers too the biggest shift myself included I I have you know countless agents running right now
就像是，好，你需要開 8 個 Goose 實例，你需要把工作流從按順序處理一個 PR、提交它、等審核、做修改，轉變為我現在有 14 個 agent 正在幫我建立 PR，我要在所有這些之間切換上下文。而且這不只是在軟體開發端，PM 也是、成長行銷也是。最大的轉變，包括我自己，我現在有無數個 agent 正在執行。 - that I have to I have to go check on.
我必須去查看它們的狀態。 - Uh it’s it’s not um it’s less of a linear workflow and it’s more of like in the background there’s 10 or 20 agents who are doing a whole bunch of stuff and then I have to check in on the work and nudge it and change it or what have you and then I can commit it to GitHub and I can I can get the markdown file.
這不再是線性的工作流了，而更像是在背景中有 10 到 20 個 agent 在做一大堆事情，然後我需要查看進度、推動它、修改它之類的，然後我可以提交到 GitHub，拿到 markdown 檔案。 - We can put it in the source of truth and we can move on.
我們可以把它放入真實來源，然後繼續前進。 - » Yep.
» 是的。 - So we have a lot of you know public companies in the audience.
在場有很多上市公司。 - We have a lot of founder businesses in the audience.
也有很多創辦人企業。 - Do you expect other companies to kind of follow a similar path and and I guess what conditions need to be in place for that to be successful?
你預期其他公司會走類似的路嗎？你認為成功需要什麼條件？ - » I I I don’t I don’t necessarily want to like I I talked at the beginning about um the groundwork that happened in ‘23, ‘24, and ‘25.
» 我不一定想要… 我在開頭提到了 2023、2024 和 2025 年完成的基礎工作。 - Like we built this agent substrate, Goos, and then we built a lot of tooling at the company on top of it.
我們建立了這個 agent 基底 Goose，然後在此之上建立了很多公司工具。 - We have an agentic operating system, internal only, called G2, where anyone can automate any deterministic workflow.
我們有一個自主式作業系統，僅供內部使用，叫做 G2，任何人都可以用它自動化任何確定性工作流。 - So, anyway, I think there’s work to do to to be successful.
總之，我認為要成功需要做很多工作。 - I would expect many companies are doing that work.
我預期很多公司正在做這些工作。 - Some of them are incredibly um far ahead than than others.
有些比其他的遠遠走在前面。 - Um, and so I I I don’t know what to expect.
所以我不知道該期待什麼。 - What I will say is, like, to the extent that I I do believe that fundamentally for like a given product or for a given roadmap, you’re going to need fewer engineers, fewer designers, fewer PMs.
我要說的是，我確實相信從根本上來說，對於一個特定的產品或路線圖，你將需要更少的工程師、更少的設計師、更少的 PM。 - I think that’s like very, very clear based after like December.
我認為在 12 月之後這已經非常非常清楚了。
勞動力的未來與產業影響
- Um, that doesn’t necessarily mean that there’s going to be fewer engineers, designers, and PMs in the world.
但這不一定意味著世界上的工程師、設計師和 PM 會變少。 - Um, it’s like the classic Jevons paradox thing where I I think that there’s probably now just a superset of things that that can be built.
這就像經典的 Jevons 悖論，我認為現在可能有一個更大的集合的事物可以被建造出來。 - Um, so I don’t know, you know, a given tech company might be might be way smaller, but there might be 50 or 100 more tech companies, or you’re going to start getting this development working in in sectors and and areas where that hasn’t historically been the case.
所以你知道，一家特定的科技公司可能會小得多，但可能會多出 50 或 100 家科技公司，或者你會開始看到這種開發模式在歷史上沒有出現過的行業和領域中運作。 - Um, but I I’m not here to to predict the future.
但我不是來預測未來的。 - I’m focused on Block.
我專注在 Block。 - » Uh, fair.
» 公平。
AI 在 Block 營運中的整合
- You you talked a bit about kind of the some of the AI infrastructure you built.
你稍微談到了你們建立的一些 AI 基礎設施。 - Maybe you can get go in a bit more depth uh, you know, both in how it’s impacting the kind of technology or I’m also curious about, you know, how are you using AI in other parts of the business you oversee, ops, customer support?
也許你可以更深入一些，包括它如何影響技術層面，我也好奇你們如何在你負責的其他業務領域使用 AI，像是營運、客戶支援？ - » Yeah.
» 是的。 - Um, so I got asked at an investor conference uh last week, like, how is AI like flowing through Block?
上週在一場投資人會議上有人問我，AI 是如何在 Block 中流動的？ - And to me, that it’s like asking, um, how are computers flowing through Block?
對我來說，這就像在問電腦是如何在 Block 中流動的。 - Uh, like it’s it’s a uh fundamental inbuilt thing that has changed on in like a binary way over the past 18 months, and then feels like it changed all over again in the past four months.
它是一個根本性的內建事物，在過去 18 個月以二元的方式改變了，然後感覺在過去四個月又徹底改變了一次。 - Um, so I’ll break it down into internal and then external and how we’re thinking about our products, what we’re putting in customers’ hands, and then I can talk a little bit about the the future and where we think things are going.
所以我把它分成內部和外部來談，我們如何思考我們的產品、我們把什麼放到客戶手中，然後我可以談談未來以及我們認為事情的走向。
內部組織結構
- So on the internal side, I think the biggest difference is the shape of the of the org.
在內部方面，我認為最大的差異是組織的形狀。 - So we used to have kind of like a classic hierarchical uh structure.
我們過去有一種經典的層級結構。 - It it was functional.
它是功能性的。 - Um, which was great, but it was like fairly standard if you like averaged through a bunch of medium-sized tech companies.
這很好，但如果你看一堆中型科技公司的平均情況，這是相當標準的。 - Um, and so you would have kind of eight server engineers, four client engineers, a PM, a designer, and you would work linearly through your roadmap.
所以你會有大概八個後端工程師、四個前端工程師、一個 PM、一個設計師，然後按線性方式推進路線圖。 - Now we have um small squads.
現在我們有小型小隊。 - So squads of like one to six people.
大約一到六人的小隊。 - Um, so meaningly smaller than the other teams would be.
比以前的團隊小得多。 - And we have way more flexibility and and fluidity where a given squad can work a few cycles on this product, get it live, and then a cycle on this other product.
我們有更多的靈活性和流動性，一個小隊可以在這個產品上工作幾個週期讓它上線，然後在另一個產品上工作一個週期。 - Um, which is different than how things worked a year or two ago where it’s like, I’m on the banking team, I’m going to be on the banking team forever.
這跟一兩年前的做法不同，以前是「我在銀行團隊，我會一直在銀行團隊」。 - We also have way fewer layers.
我們也有更少的層級。 - So on the development side, I think we probably cut our layers by I don’t know, 50 or 60%.
在開發端，我想我們大概砍掉了 50% 到 60% 的層級。 - Like on the product side, I only have, I think, two layers, maybe three layers in a in a couple places.
像是在產品端，我只有兩層，也許在幾個地方有三層。 - And so information is flowing um way more freely.
所以資訊流動得更加自由了。
開發與自動化流程
- I think that then in terms of how we actually build on the development side, things have changed.
在開發端的實際建造方式方面，事情已經改變了。 - I think everyone’s probably seen, you know, every every CEO out there is going on Twitter and showing their like green dot on on uh on GitHub.
我想大家都看到了，每個 CEO 都在 Twitter 上展示他們在 GitHub 上的綠點。 - Um, but that’s real.
但那是真的。 - Like all of our designers are are shipping PRs.
我們所有的設計師都在提交 PR。 - All of our product managers are shipping PRs.
我們所有的產品經理都在提交 PR。 - That’s not that interesting anymore.
這已經不那麼有趣了。 - I think more interesting is that we have uh internal tools that are similar to Claude Code, but they’re like more plugged into our infrastructure.
我認為更有趣的是我們有類似 Claude Code 的內部工具，但它們更深入地接入了我們的基礎設施。 - So we have a tool called Builderbot.
我們有一個叫做 Builderbot 的工具。 - Builderbot is just autonomously merging PRs and actually like building features to 100%.
Builderbot 就是自主地合併 PR，實際上是把功能建置到 100%。 - We’ve had some fairly complex features that are built to 100%.
我們有一些相當複雜的功能是被建置到 100% 完成度的。 - More often than not, it’s building them to like 85 or 90%.
更常見的是它把功能建置到大約 85% 或 90%。 - And then a human who who has a lot of context and understands does like the final the final 10%.
然後一個有很多背景知識和理解的人來做最後的 10%。 - So that feels really, really different.
所以這感覺真的非常非常不同。 - The ability to go from, um, to go from idea to like this is in the hands of 100,000 or a million customers has been compressed massively since, uh, since December.
從構思到產品交到十萬或一百萬客戶手中的能力，自 12 月以來已經被大幅壓縮了。 - Outside of development, I would say most of what we’re seeing is like anytime there’s a deterministic workflow, we’re we’re able to automate that.
在開發之外，我們看到的大部分情況是只要有確定性工作流，我們都能自動化。 - And so generally at a at scale tech company, you have individuals who are working queues.
在一家規模化的科技公司裡，你通常有人在處理佇列。 - Um, a lot of that is just being completely automated away.
很多這類工作正在被完全自動化取代。 - Like from a customer support perspective, this is not new, but you know, our chatbots and and AI phone support and and whatnot are automating a a majority of inquiries that we get.
從客戶支援的角度來看，這不是新鮮事，但我們的聊天機器人和 AI 電話支援等正在自動化我們收到的大部分詢問。 - And then it gets into like, um, product operations and risk operations and compliance operations and any sort of decisioning.
然後延伸到產品營運、風險營運、合規營運以及任何決策類工作。 - Like generally, um, generally the the the models and the agents are going to do a better job than humans.
一般來說，模型和 agent 會做得比人類更好。 - Right now, I think it’s critical that we have a human in the loop.
現在，我認為讓人類在迴路中是至關重要的。 - Uh, that’s like the key kind of buzzword, uh, when you talk to talk to partners and regulators and and what have you.
這是你跟合作夥伴和監管機構談話時的關鍵詞彙。 - Um, but over time, it’s like pretty obvious that these systems are just going to be so much better than like having a thousand humans who are who are doing that work.
但隨著時間推移，很明顯這些系統將會比讓一千個人類做那些工作來得好太多。 - So that’s on the internal side.
這就是內部方面。 - Um, on the on the product side, I think that » and maybe just catch people up on kind of the shape of the business.
在產品方面… » 也許先讓大家了解一下業務的整體樣貌。
Block 的業務結構與成長
- Obviously, you have Square, you have Cash App.
顯然你們有 Square，有 Cash App。 - You you made a big acquisition, Afterpay.
你們做了一次重大收購，Afterpay。 - » Sure.
» 當然。 - » What do those businesses look like and then, yeah, how are they kind of changing with » Sure.
» 這些業務看起來怎樣？它們是如何變化的？ » 好的。 - So, um, so we used to operate in a business unit structure.
我們過去以業務單位結構運作。 - So Square used to be kind of its own business unit with its own CEO.
Square 過去是自己的業務單位，有自己的 CEO。 - Cash App was its own business unit with its own CEO.
Cash App 也是自己的業務單位，有自己的 CEO。 - Um, that wasn’t leading to the right outcome.
那沒有帶來正確的結果。 - So about 18 months ago, we functionalized the company.
所以大約 18 個月前，我們把公司功能化了。 - Just meaning that all of engineering rolls up to our head of engineering, all of design to our head of design, all of product to me.
意思就是所有工程歸工程負責人管，所有設計歸設計負責人管，所有產品歸我管。 - So we have a financial platform team that spans the entirety of Block.
我們有一個橫跨整個 Block 的金融平台團隊。 - We have a business platform team that’s doing a lot of this automation that spans the the entirety of of Block.
我們有一個做大量自動化的業務平台團隊，也橫跨整個 Block。 - And then increasingly, we’re building features and products that actually connect the Square side, the Cash App side, and the Afterpay side.
然後越來越多地，我們正在建造實際連接 Square、Cash App 和 Afterpay 的功能和產品。 - And so naturally, you you’re building technology and you’re building infrastructure that is not, um, brand specific.
所以自然地，你在建造的技術和基礎設施不是品牌特定的。 - And that’s actually like, kind of central to our our overall strategy and and and overall thesis.
這實際上是我們整體策略和整體論點的核心。 - Um, but yeah, I mean, CA Cash App went from when I joined Cash App in 2016, uh, we had just just started to to figure out how to monetize and had our first dollars of gross profit.
當我 2016 年加入 Cash App 時，我們才剛開始摸索如何變現，有了第一筆毛利。 - And now I think Cash App’s probably like, I don’t know, 60-ish percent of like overall gross profit at the at the company.
現在我想 Cash App 大概佔公司整體毛利的 60% 左右。 - So overall, been been growing at a healthy clip over the past decade.
整體來說，過去十年一直以健康的速度成長。 - Um, but uh Cash App and Afterpay have definitely been growing, um, more quickly, but increasingly we’re trying to think about things from an ecosystem perspective.
但 Cash App 和 Afterpay 確實成長得更快，不過我們越來越試著從生態系統的角度來思考事情。
Goose：Block 的 Agent 框架平台
- And and that’s maybe where like Goose as a platform comes in, which is we bu we built Goose internally.
這也許就是 Goose 作為平台登場的地方，我們在內部建立了 Goose。 - The way to think about Goose is, um, it’s a nod to Top Gun or whatever, the co-pilot thing.
Goose 的命名致敬了乘風破浪（Top Gun），那個副駕駛的概念。 - But the way to think about Goose is it’s a it’s an agent harness and it’s model agnostic.
但 Goose 的概念是一個 agent 框架，而且是模型無關的。 - So I can run Goose on an Anthropic model, on a on an OpenAI model, on an open-source model.
我可以在 Anthropic 模型、OpenAI 模型、開源模型上執行 Goose。 - There’s probably like 120 models that we have.
我們大概有 120 個模型。 - And depending on what I’m trying to do, I’ll kind of swap out the swap out the models.
取決於我要做什麼，我會切換不同的模型。 - And then that was useful for a human to use, but we’ve built like the agentic layer on top.
那對人類使用是有用的，但我們在上面建立了自主式層。 - And so now a lot of the automations at at Block are actually routing through the Goose agent harness.
所以現在 Block 的很多自動化實際上都透過 Goose agent 框架來路由。 - And, um, we’ve been able to leverage this across the products that we’re building.
我們已經能夠在我們建造的產品中利用這一點。 - So, Moneybot, which we’d like to think of as like a CFO in your pocket, but it’s essentially like a proactive, um, a proactive, uh, chatbot that can take actions on your behalf within Cash App.
Moneybot，我們喜歡把它想成口袋裡的 CFO，但它本質上是一個主動式的聊天機器人，能在 Cash App 中代你採取行動。 - That’s built on top of Goose.
這是建立在 Goose 之上的。 - Managerbot, which is roughly a similar thing on the Square side, that’s built on top of Goose.
Managerbot 在 Square 端是類似的東西，也是建立在 Goose 之上。 - So, it’s a lot of this foundational work on agentic systems and then like the the triggers and the underlying data and events that you need to power them that’s working across the, uh, the entirety of the of the company.
這是很多關於自主式系統的基礎工作，然後還有驅動它們所需的觸發器、底層數據和事件，這些橫跨整個公司在運作。
生成式 UI 與產品演進
- So, on the on the product side, um, I think that the the biggest shift has really been like we’re going from a world where for the past 10 or 15 years, everyone’s used to a static UI, a rigid UI.
在產品方面，我認為最大的轉變是過去 10 到 15 年來大家都習慣了靜態的 UI、固定的 UI。 - You tap through the UI, everyone has the same, everyone’s Uber or Lyft or Cash App or whatever looks the same.
你在 UI 上點來點去，每個人的 Uber、Lyft、Cash App 或其他什麼都看起來一樣。
生成式 UI 的實務應用
- That’s going to fundamentally change in the next like six months.
這在接下來六個月內將從根本上改變。 - Um, generative generative UI is is is here.
生成式 UI 已經到來了。 - We’re seeing it with Moneybot.
我們在 Moneybot 中看到了。 - We’re seeing it with Managerbot as the models get better.
隨著模型越來越好，我們在 Managerbot 中也看到了。 - » What what is that going to look like kind of in practice?
» 實際上那會是什麼樣子？ - I’m curious.
我很好奇。 - » I think, I mean, in simplest terms, it’s like your Cash App should look really different from mine.
» 用最簡單的話說，你的 Cash App 應該跟我的看起來非常不同。 - And the reason why it’s like, okay, well, I get my paycheck into Cash App and I’m super into Bitcoin.
原因是，好，我把薪水轉到 Cash App，而且我超愛 Bitcoin。 - Let’s say, like you don’t, and you use Afterpay all the time.
假設你不是這樣，你一直在用 Afterpay。 - Great.
好。 - When we open up our apps, that should be totally different.
當我們打開 app 時，應該是完全不同的。 - That you could probably achieve that just through personalization.
那你大概可以只透過個人化就做到。 - That’s not that interesting.
那不太有趣。 - What we’re actually seeing, and Anthropic had some releases this week that are that are incredible.
我們實際看到的，而且 Anthropic 本週有一些驚人的發布。 - We’re actually seeing is like I can go into Moneybot and say, “How have I been spending my money?”
我們實際看到的是我可以進入 Moneybot 然後說：「我的錢花在哪裡了？」 - And it’ll show me a bunch of charts and, uh, and visualizations where it is actually like on the fly generating generating that visualization.
然後它會給我看一堆圖表和視覺化，而且它實際上是即時生成那些視覺化的。 - It’s not actually in the code itself.
這些不在程式碼本身中。 - So that’s really cool.
所以那真的很酷。 - It’s also potentially a nightmare from like a QA perspective.
從 QA 的角度來看，這也可能是一場噩夢。 - And so we need to figure out how you’re going to QA all these like non-deterministic outputs for for tens of millions of customers.
所以我們需要搞清楚如何對數千萬客戶的這些非確定性輸出做品質保證。 - But, um, a great example on the on the Square side is with Managerbot.
但在 Square 端有一個很好的例子是 Managerbot。 - Maybe charts aren’t that impressive to you, but with Managerbot, let’s say you’re a, you’re a, uh, you own a multilocation quick serve restaurant.
也許圖表對你來說不那麼令人印象深刻，但用 Managerbot，假設你擁有一家多地點的快餐店。 - You say like, “Hey, can you build me an app where I can, uh, manage scheduling for these two locations and like automatically fire off text via, you know, WhatsApp or or Signal or whatever to my, um, to my employees?”
你說：「嘿，你能幫我建一個 app，讓我可以管理這兩個地點的排班，然後自動透過 WhatsApp 或 Signal 或其他方式發訊息給我的員工嗎？」 - It’s actually going to like create that app for you.
它實際上會幫你建立那個 app。 - And the the way that that app looks and feels is not in the source code of the actual application that we push to the to the app store.
而且那個 app 的外觀和感覺並不在我們推送到 App Store 的實際應用程式原始碼中。 - And so I think it’s, um, it gives folks way more control.
所以我認為這給了人們更多的控制權。 - It’s way more, uh, ultimately, I think it’ll lead to higher engagement.
最終我認為它會帶來更高的參與度。 - Um, I think it’ll lead to better product and, and really, I think the key thing, I, I don’t think that if we ask customers to to like prompt these tools themselves, they’re going to necessarily know the right prompts and come up with the right answers.
我認為它會帶來更好的產品。而且關鍵的事情是，如果我們要求客戶自己去 prompt 這些工具，他們不一定知道正確的 prompt 或得出正確的答案。 - So, we’ve invested massively on the proactive intelligence side where what we’ve found, especially as it relates to money is like we need to be prompting our customers with things that we think make sense for them and that’s where we’re creating a lot of the the value.
所以我們在主動式智慧方面大量投資，尤其是關於金融方面，我們需要主動向客戶提示我們認為對他們有意義的事情，那就是我們創造大量價值的地方。 - » So, I mean, I think we’re all incredibly bullish on on kind of the impact of AI, you know, in kind of in the way that all these businesses run and the products you can create.
» 我想我們都對 AI 對業務運作方式和能創造的產品的影響非常看好。
股價與業務成長
- How does that flow back to your stock price?
這如何反映到你的股價上？ - you know the the business is the stock has been roughly flat for I don’t know six or seven years.
股票大概持平了六七年。 - interesting sort of reminding me » but the b the business has grown a lot you know to your point the gross profit per employee has grown you know massively like how do you sort of reconcile the that that dimension » yeah I think um so so I think you know markets are markets are cyclical and there’s all sorts of things that are happening I remember uh in 2021 when our stock price was like I don’t know 260 bucks and I was like that was a little bit irrational um you can take a a kind of
有趣… » 但業務成長了很多，就像你說的，每位員工的毛利大幅成長了，你怎麼看待這個落差？ » 是的，我想市場是有週期性的，有各種事情在發生。我記得 2021 年我們的股價大概是 260 美元，我覺得那有點不理性。你可以採取一種… - longer term mature view and say you So markets are voting machines in the near term, but they’re weighing machines in the long term.
更長期、更成熟的觀點來看。市場短期是投票機，但長期是秤重機。 - Just like focus on building, » you know, David and Jonathan earlier talked a bit about kind of defensibility.
就是專注在建造上。» David 和 Jonathan 稍早談了一些關於防禦力的話題。 - How do you think about your own moes at Square?
你怎麼看待你在 Square 的護城河？ - I mean, at Block, excuse me.
我是說 Block，不好意思。
Block 的護城河與防禦力
- You, you know, you talked a bit about the ecosystem.
你談到了一些關於生態系統的事情。 - You guys obviously have, you know, regulatory infrastructure.
你們顯然有監管基礎設施。 - Um, you know, how do you think about, you know, that the business overall in that context?
在那個背景下，你如何看待整體業務？ - Yeah, I think in the I think in the near term and the medium term, um there’s a bunch of there’s a bunch of modes that exist for for block and and we can talk about the industry more broadly.
我認為在中短期內，Block 有一些護城河存在，我們也可以更廣泛地談談產業。 - I think I think distribution and network effects are are one of them.
我認為分銷和網絡效應是其中之一。 - I I agree on the the catrini piece and uh and Door Dash.
我同意關於 Catrini 和 DoorDash 的觀點。 - I don’t think anyone’s vibe coding Door Dash in the next couple of weeks here.
我不認為任何人在接下來幾週內能 vibe code 出一個 DoorDash。 - Uh I like to say like any of us can can create a peer-to-peer app in probably a week.
我喜歡說我們任何人大概一週就能建立一個 P2P 應用。 - uh no one’s going to vibe code you know 50 or 60 million monthly activives who are actually using that.
但沒有人能 vibe code 出 5000 萬到 6000 萬個實際使用的月活用戶。 - So I think that that’s true.
所以我認為那是對的。 - Uh I think um you know licenses and and regulatory posture um definitely exists.
牌照和監管態勢絕對是存在的。 - I think hardware right now it’s like harder to imagine how some of the AI tools flow through to the to the hardware side.
我認為現在很難想像 AI 工具如何延伸到硬體方面。 - Like you can’t vibe code a piece of square hardware.
你不能 vibe code 出一塊 Square 硬體。 - Um but I I think longer term if we continue like if we look at the rate of the change and and the change in the change I think longer term the key thing that’s going to make uh a company defensible is um the extent to which the company understands something that is pretty hard for other companies to understand.
但我認為長期來看，如果我們看變化的速度和變化中的變化，長期而言讓一家公司有防禦力的關鍵是這家公司在多大程度上理解了其他公司很難理解的事物。
Block 作為一個智慧系統
- And so we’re increasingly building toward a world and talking about block as an intelligent system itself.
我們越來越朝著一個方向建造，把 Block 本身視為一個智慧系統 (intelligent system)。
數據、洞察與迭代循環
- » So basic like the the the the way that I see this going if we can if you extrapolate forward the past several months is that ultimately a company is sitting on top of some sort of signal, some sort of like rich data and and and deep insight.
» 如果你把過去幾個月的趨勢往前延伸，我看到的方向是最終一家公司是坐在某種信號上，某種豐富的數據和深刻的洞察之上。 - Um for us it’s like how sellers and buyers participate in the economy.
對我們來說，就是賣家和買家如何參與經濟活動。 - Um and and most companies I think have this thing that they understand deeply.
我認為大多數公司都有他們深度理解的東西。 - And then the question is going to be how quickly can you iterate to improve that understanding over time.
然後問題是你能多快地迭代來隨時間改進那個理解。 - And so we’re building world models internally and externally of like understanding who our customers are but then also understanding how block operates.
我們在內部和外部建立世界模型，理解我們的客戶是誰，也理解 Block 是如何運作的。 - It’s like you can imagine you can imagine for any company just like a markdown file of like who you are and then you need the feedback loop with two things.
你可以想像任何公司都有一份 markdown 檔案，寫著你是誰，然後你需要一個回饋迴路搭配兩樣東西。 - You need a feedback loop with the signal which is like what do you what do you deeply understand that’s hard for others to understand and then you need a tool like builderbot or Claude Code or what have you.
你需要一個與信號的回饋迴路，就是你深度理解什麼是其他人難以理解的，然後你需要一個像 Builderbot 或 Claude Code 之類的工具。 - And then you can just iterate through that loop over and over and again.
然後你就可以在那個迴路中不斷地迭代。 - It’s like this is this is what I’m seeing.
這就是我看到的。
開發的未來與護城河
- This is what’s happening.
這就是正在發生的事。 - Great.
好。 - This is our markdown file for for block.
這是我們 Block 的 markdown 檔案。 - These are our values.
這些是我們的價值觀。 - this is the metrics we’re trying to optimize for.
這是我們試圖最佳化的指標。 - Um, this is what we care about.
這是我們在意的事。 - This is what we don’t care about.
這是我們不在意的事。 - And then you have a gentic system, so you can just build stuff.
然後你有一個自主式系統，所以你就可以建造東西了。 - And right now, you basically you’ve taken that humans used to do that and it used to take a couple months to build a feature.
現在的情況是，以前是人類在做這些，建造一個功能大概要幾個月。 - Um, now it takes maybe a week or two and there’s still humans involved.
現在大概一兩週，而且仍然有人類參與。 - Pretty clear that in the future you’ll be able to run that loop like I don’t know hundreds, thousands of times a day and maybe there’s some humans involved.
很清楚的是，未來你將能夠一天跑那個迴路幾百次、幾千次，也許有一些人類參與。 - Maybe not.
也許沒有。 - Maybe the humans are more like editors.
也許人類更像是編輯。 - And so I think the the biggest moat is going to be like which companies understand something that’s super hard for other people to understand.
所以我認為最大的護城河將是哪些公司理解了對其他人來說非常難以理解的事情。 - And if your answer to that is is um I don’t know, then uh then you maybe could get vibe coded away.
如果你的答案是「我不知道」，那你可能就會被 vibe code 取代。 - » This has been an amazing conversation.
» 這是一場很棒的對話。
結語與致謝
- Thank you.
謝謝。 - Uh thank you so much for for joining us.
非常感謝你加入我們。 - » Appreciate it.
» 感謝。 - Thanks so much.
非常感謝。 - » Awesome.
» 太好了。