AI 时代的组织设计:小团队、零会议、快决策
主题综述
主题页(活文档)· 最近更新 2026-06-12 · 取材 13 篇访谈
更新日志
- 2026-06-12 — 首次综述。基于 13 篇访谈,核心结论:互不相关的赢家(AppLovin、Google Labs、ElevenLabs、Snap、Airbnb、Surge、OpenAI)确实收敛到"个位数到低双位数团队、把会议压到接近零、以天为单位做决策"的同一参数区间,但证据显示这套打法在 AI 之前就在赢——AI 更像放大器而非起因;真正还活着的分歧只有两个:会议到底是损耗还是创造仪式,以及小团队该在什么时刻变大。(本主题由旧库挖掘 /excavate 发现)
主流共识
把 13 篇访谈摆在一起,最显眼的不是任何一个人的观点,而是数字的收敛:互不通气的公司,给出的团队规模几乎落在同一区间。共识可以拆成五层。
一、团队规模:个位数到低双位数,反复出现
ElevenLabs(约 350 人)把整个产品组织切成 20 个全自治小队:
"The way we structure it to be able to ship quickly, especially with so many things happening in the AI space, is a lot of small teams. So today we have roughly 20 product teams, each of five to 10 people in size, which with full independence can go ahead and ship products."「我们构建它的方式是为了能够快速交付,尤其是在人工智能领域发生如此多事情的情况下,就是分成许多小团队。所以今天我们大约有 20 个产品团队,每个团队有 5 到 10 人,他们完全独立,可以继续交付产品。」Mati Staniszewski (ElevenLabs CEO) · ElevenLabs CEO: Why Voice is the Next AI Interface
Google Labs 的起步配置更小,且明确把"从想法到上线的速度"当作第一指标:
"Very small teams. Most of our Google Labs teams start with like five to seven people. It's usually like PM, some engineers, UXer, go. We put a huge premium on how fast you can go from idea to in people's hands."「非常小的团队。我们大多数 Google Labs 团队都是从五到七个人开始的。通常是产品经理,一些工程师,用户体验师,开始。我们非常重视从想法到用户手中的速度。」Josh Woodward (Google Labs / Gemini VP) · Inside Gemini and NotebookLM | Josh Woodward
Flow 用这个配置"不到 100 天"从想法发到 I/O("I think that one was like 86 days"/「我认为那次大概是 86 天之类的」)。Snap(约 5000 人)的核心产品与设计团队同样停在个位数:
"If you look at the core product and design team, that's a very small team. I mean, I think it's like currently like nine people. It usually fluctuates between like eight and twelve people at any given time."「如果你看核心产品和设计团队,那是一个非常小的团队。我的意思是,我认为目前大约有九个人。通常在任何时候,人数都在八到十二人之间波动。」Evan Spiegel (Snap CEO) · Evan Spiegel, Snap
Airbnb 的 Project Hawaii 是同一区间——7000 人的公司里抽出 10-12 人:
"And so we put together a team, 10 or 12 people, designers, engineers, product people, data scientists. It's mostly just a pure software team. And we treated it like a little startup."「因此我们组建了一个 10 到 12 人的团队,成员包括设计师、工程师、产品经理和数据科学家。这基本上就是一个纯粹的软件开发团队。我们把它当作一家初创公司来运营。」Brian Chesky (Airbnb CEO) · Brian Chesky - AI Founder Mode
OpenAI 内部,Deep Research 立项时只有两个人:
"Yeah, we definitely reward agency. And I think that's like always been true. And I think especially on the research side, the teams are quite small. Like when Isa was working on deep research, it was like two people."「是的,我们绝对会奖励能动性。我认为这始终是正确的。而且我认为特别是在研究方面,团队规模很小。比如当 Isa 在做深度研究时,只有两个人。」Christina Kim (OpenAI) · GPT-5 and Agents Breakdown | OpenAI Researchers
把规模推到整公司级别的是 Surge AI——不到 100 人做到 10 亿美元收入、零融资:
"Yeah, so we hit over a billion in revenue last year with under 100 people. … Like I used to work at a bunch of the big tech companies, and I always felt that we could fire 90% of people and we would move faster because the best people wouldn't have all these distractions."「去年我们以不到 100 人的团队实现了超过 10 亿美元的收入。……就像我曾经在许多大型科技公司工作过一样,我一直觉得我们可以解雇 90% 的人,我们会行动得更快,因为最优秀的人才不会有这么多的干扰。」Edwin Chen (Surge AI CEO) · The 100-person AI lab | Edwin Chen (Surge AI)
AppLovin 同构:市值千亿美元,全公司工程师不到一百人——
「但你要知道我们全公司所有的工程师加在一起不到一百人,我们在湾区属于总部的总部工程师也就五十多个吧。」
— 葛小川 (AppLovin VP Engineering) · EP 68. 对话 AppLovin 技术 VP 葛小川
对"为什么是这个数"的解释,Bob Baxley 给得最直白(该篇逐字稿库内仅存中文)。他在 Apple 负责在线商店设计时:
「我们有六名设计师,而任何其他公司都会有 60 名或更多。……你用四个人就能得到披头士乐队。你用八个人得不到披头士乐队。你肯定不会用 24 个人得到它,对吧?就像团队变得太大,你就无法得到它。这就是 Brian Eno 所说的(scenius,集体天才)。」
— Bob Baxley (前 Apple / Pinterest 设计负责人) · 35 years of product design wisdom | Bob Baxley
二、会议:默认值从"开"翻转为"不开"
AppLovin 是最极端的样本——千人公司、上市公司,公开宣称不开会:
"Yeah, you know, I don't really operate like that. So like, we don't hold meetings, we don't debate things. It's like, if we want to do something, we just go do it."「其实,我不是那样运作的。比如,我们不开会,我们不辩论。就像,如果我们想做什么,我们就直接去做。」Adam Foroughi (AppLovin CEO) · 20VC: How To Run a 1,000+ Organisation with Very Few Meetings
替代品是带议程的异步文字:
"If you want to hold a meeting, there has to be an agenda. There have to be points that you're going to bring up. So start a chat, lay those points out, lay the thought process out. What's your problem? What's the question? What do you want? And if it can get resolved in a chat, it's done. It takes two, three minutes."「如果你想举行会议,必须有一个议程。必须有你要提出的要点。所以先发起一个聊天,列出这些要点,列出你的思考过程。你的问题是什么?你要问什么?你想要什么?如果能在聊天中解决,那就完成了。花两三分钟。」Adam Foroughi (AppLovin CEO) · 20VC: How To Run a 1,000+ Organisation with Very Few Meetings
四年后他的工程 VP 在另一档节目里讲的版本几乎一字不差,并给出判断标准:
「我真的觉得 meeting 是一个很低效的一个沟通方式,尤其是多人的,多人的 Schedule Meeting,我觉得是最浪费时间的。……如果一件事情你需要 schedule 一个 meeting 去讨论的话,这个事情八成是不重要,你根本就不需要去讨论。因为如果你真的觉得这个事情重要的话,你应该直接就到那个人,跟他去聊一下。」
— 葛小川 (AppLovin VP Engineering) · EP 68. 对话 AppLovin 技术 VP 葛小川
Chesky 站在另一端走向同一方向:他过去靠每周 35 小时会议运转 founder mode,但判断 AI 会终结这种机制(详见分歧节)。Surge 的 Edwin Chen 则是气质性厌恶:"I have a very hard time being in meetings all day."/「我很难整天都待在会议里。」没有任何一位发言人为"例行多人会议"辩护——这在 13 篇里是全场一致。
三、决策:频率优先于单次正确率
"I made 10 decisions a day and most of them aren't strategic. … if you plan, if you process, if you over communicate, you start over analyzing and slow down. You just plow through decisions, and if you have the capacity to execute on more, it's just more probability that something's going to work out."「我一天做 10 个决定,大多数都不是战略性的。如果你计划、处理、过度沟通,你就会开始过度分析,速度会减慢。快速做出决定,如果你有能力执行更多,成功的可能性就越大。」Adam Foroughi (AppLovin CEO) · 20VC: How To Run a 1,000+ Organisation with Very Few Meetings
"I'm probably wrong eight times out of 10, but it doesn't matter because I take so many chances."「我可能有 80% 的时候是错的,但这没关系,因为我抓住了很多机会。」Adam Foroughi (AppLovin CEO) · 同上
葛小川把同一逻辑下放到技术层:
「一个很困难的东西它被做成功,其实背后是在你解决这个问题的过程中,每一天都要经历几十个甚至上百个这种 micro decision,然后你把这些所有的 decision 都做好,或者说你做了更多的正确的这种选择,然后更少的错误的选择。」
— 葛小川 (AppLovin VP Engineering) · EP 68. 对话 AppLovin 技术 VP 葛小川
战略级决策也一样快——AppLovin 把整块游戏业务打包卖掉,「从我们提起这个话题,到这个事情已经完成了,也就短短的几个月的时间而已」(葛小川,同上)。Google 那边的对应物是 Robby Stein 描述的 AI Mode:5-10 人小队起步,约一年从想法到改写搜索主产品。
四、层级与角色:砍传递层,消灭"纯管理者"
"I think we're going to have a lot fewer layers of management. You know, I think there was an old famous saying, the Catholic church has been going on for 2000 years, only have four layers of management. Why do every other company have like seven, eight, nine layers of management?"「我认为我们的管理层级会少得多。你知道,有一句著名的老话,天主教会存在了 2000 年,也只有四个管理层级。为什么其他公司却有七、八、九个管理层级?」Brian Chesky (Airbnb CEO) · Brian Chesky - AI Founder Mode
"I don't think people managers will have any value in the future. … You don't manage the people, you manage the work."「我认为未来纯粹的『人员管理者』将毫无价值。……你不是在管理人员,你是在管理工作。」Brian Chesky (Airbnb CEO) · 同上
AppLovin 早就把这条写进了规则:
「(我们)就没有纯的 people manager,我们要求就是公司里面每一个人都要能够是一个很好的 IC,然后一个团队,如果你想做的团队领袖的话,你一定要是这个团队最 technical 的人。」
— 葛小川 (AppLovin VP Engineering) · EP 68. 对话 AppLovin 技术 VP 葛小川
ElevenLabs 干脆取消头衔:
"We removed titles a year ago, and it's going well. … The tenure will not define your position in the hierarchy. If you are smart and quick and passionate, you can elevate yourself very quickly."「我们在一年前取消了头衔,而且进展顺利。……任期不会决定你在等级制度中的位置。如果你聪明、敏捷、充满激情,你可以很快提升自己。」Mati Staniszewski (ElevenLabs CEO) · ElevenLabs CEO: Why Voice is the Next AI Interface
Chesky 为砍层级给出了机制解释——层级即信息失真:
"They end up in a game of telephone. They say something and then it goes down like five layers and then it goes back up five layers … literally like the lowest point for me when I was having meetings about meetings."「他们最终会陷入传话游戏中。他们说一句话,经过五个层级的传递,再传回五个层级……对我来说,最痛苦的时候就是我一直在开那些『为开会而开的会』。」Brian Chesky (Airbnb CEO) · Brian Chesky - AI Founder Mode
五、招聘:小团队组织把筛人当作结构参数本身
零会议、无中层的组织把约束前移到招聘。AppLovin 的筛选标准是负面清单式的:
"We haven't hired any MBAs because they try to come in and want the senior role and want the incremental pay, and I just can't justify it. … We've hired nobody from big companies."「我们最终没有聘请任何 MBA,因为他们一来就想担任高级职位,想要更高的薪水,我觉得这不合理。……我们没有从大公司聘请任何人。」(主持人接话:大公司的人会怎样?Foroughi:"Slow down."/「慢下来。」)Adam Foroughi (AppLovin CEO) · E928 This Week in Startups
葛小川给出正面清单:「名校毕业……毕业之后去了一个不是特别知名的公司工作了一到两年。但是自身发展太快了,那个地方容不下他。然后这种人来到这里之后一般会非常非常的成功。」(同上 EP 68)。Google Labs 的版本:
"People who like to tinker, people who like to build, they express themselves through prototypes, not docs."「喜欢摆弄、喜欢创造的人,他们通过原型而不是文档来表达自己。」Josh Woodward (Google Labs / Gemini VP) · Inside Gemini and NotebookLM | Josh Woodward
ElevenLabs 的极端案例是从呼叫中心挖到顶级研究员:"We hired a person that had incredible open source text-to-speech model and was working in the call center at the same time."/「我们雇佣了一个人,他有一个令人难以置信的开源文本转语音模型,同时还在呼叫中心工作。」(Mati,同上)。Surge 的 Edwin Chen 则把话说反过来:"Don't hire that Stanford grad who simply wants to add a hot company to your resume."/「不要雇佣只想在你的简历上增加一家热门公司的斯坦福毕业生。」
分歧在哪
共识铺得越平,裂缝越值得看。四条裂缝,每一条两边都是赢家。
裂缝一:会议是损耗,还是创造的仪式?
AppLovin 把会议定义为恐惧的产物:
"So, I mean, that's fundamentally why people have a lot of meetings is fear of loss. We try to hire people who are inherently entrepreneurial and don't have that fear."「人们开很多会,从根本上来说是害怕失去。我们尝试招聘那些天生具有创业精神、没有这种恐惧感的人。」Adam Foroughi (AppLovin CEO) · 20VC: How To Run a 1,000+ Organisation with Very Few Meetings
葛小川甚至论证文字在认知上优于说话:
「这种文字的交流,它其实有些时候要比直接的语言交流更高效。……你其实打字的时候,你这个文字输出的速度其实是跟你人思考的速度是更匹配的。如果你直接说话的话,其实很多时候你是没有办法深思熟虑的。」
— 葛小川 (AppLovin VP Engineering) · EP 68. 对话 AppLovin 技术 VP 葛小川
但注意他们消灭的是"预约的多人会议",不是实时沟通——AppLovin 发钱让员工搬到办公室附近(每月 1500-2000 美元房补、免费洗车理发),做电商的归因引擎是三个高管在 Vegas 酒店早餐桌上画在餐巾纸上、当周末写完代码的:「我们三个当时手边也没有草稿纸,我们就在餐厅的餐厅纸上开始画图……然后饭菜到之前,我们就把这个思路给找出来了。」(葛小川,同上)
而 Evan Spiegel 的核心创造机制,恰恰是一个每周数小时、雷打不动的会议:
"Every week I'm with our design team for several hours and we're just looking at new work. I mean, new work every week, hundreds of ideas, right? Hundreds of concepts, you know, iteration, etc."「每周我都会和我们的设计团队一起工作几个小时,我们只是在看新的作品。我的意思是,每周都有新的作品,成百上千的想法,对吧?成百上千的概念,迭代等等。」Evan Spiegel (Snap CEO) · Evan Spiegel, Snap
"Because the best way to have a good idea is to have lots of ideas. It is critical. And the most toxic thing you can have is people attached to an idea."「因为拥有好主意的最好方法就是拥有大量的想法。这至关重要。最糟糕的事情就是人们执着于一个想法。」Evan Spiegel (Snap CEO) · 同上(他补充:「我们在这些设计会议中有一半的时间都在欢笑」)
Chesky 占据第三个位置:会议曾是他获取信息的唯一机制,规模大到惊人——
"The mechanism of founder mode was I had a lot of meetings because that was the only way I could get information. So I would probably do 35 hours of meetings, which is very similar to the way Steve ran Apple. I wouldn't do one-on-ones. I'd only do group meetings."「我采用创始人模式的机制是召开大量会议,因为那是获取信息的唯一途径。所以我可能每周会开 35 个小时的会,这和 Steve 管理 Apple 的方式非常相似。我不会进行一对一的谈话。我只会召开小组会议。」Brian Chesky (Airbnb CEO) · Brian Chesky - AI Founder Mode
——但他认为 AI 让这个机制过时:
"It was a very clear chain of command, but this is a meeting based culture. I think in AI, I think we're going to move away from meeting based to asynchronous."「这是一条非常清晰的指挥链,但这是一个基于会议的文化。我认为在 AI 时代,我们将从基于会议转向异步沟通。」Brian Chesky (Airbnb CEO) · 同上
他对一对一会议的判词比 Foroughi 更狠:
"People who have lots of recurring one-on-ones are not going to survive because what they're doing is like that, oh, you come with me with whatever problem is, I'm here to help you like a mentor or professor."「那些进行大量重复性一对一谈话的人无法生存,因为他们做的就像是『哦,不管你有什么问题都来找我』,我在这里像导师或教授一样帮你。」Brian Chesky (Airbnb CEO) · 同上
三个位置不可调和:Foroughi 说会议本来就不该存在;Spiegel 的设计评审会是公司发动机;Chesky 说会议曾经必要、即将被 AI 替代。值得注意的细节是 Spiegel 的会和 Chesky 的会内容上是一类——围着具体作品的群体评审,没有 status update、没有一对一——它们与 AppLovin 反对的"预约多人会议"可能根本不是同一个物种,但没有任何发言人替对方做这个区分。
裂缝二:"精益邪教"——小团队该在什么时刻变大?
最正面的冲突来自 Google 内部。Robby Stein(Google 搜索 VP,AI Mode 负责人)直接开炮:
"I think there's kind of this cult of lean, scrappy, fast, throw away your product quickly, keep moving. I think at some level it's true for internal conviction, but to build a product that works for a lot of people that is based on a technological breakthrough, a lot of times I see teams just give up too early or underinvest in the product."「我认为现在有一种精益、节俭、快速、快速抛弃你的产品、不断前进的文化。我认为在某种程度上,这对内部信念来说是正确的,但要构建一个适用于很多人的、基于技术突破的产品,很多时候,我看到团队过早放弃或对产品投入不足。」Robby Stein (Google Search VP) · Inside Google's AI turnaround | Robby Stein
"But I see the opposite more true where people hold on to small teams too long. It takes forever to get to the thing you're looking for. This close friends example I mentioned actually was a small team. One of the reasons it took us forever was it kept the team so small and scrappy. … By a startup age, you'd be dead probably."「但我看到更多的情况是相反的,人们坚持小团队的时间太长了。要花很长时间才能找到你想要的东西。我提到的这个『密友』的例子实际上就是一个小团队。它花费我们很长时间的原因之一是它使团队保持如此小和简陋。……按创业公司的年龄来算,你可能早就死了。」Robby Stein (Google Search VP) · 同上
而他的同事 Josh Woodward 在同一家公司、同一个月的访谈里说的是反向教训:
"It's one of the mistakes I made early on in my product career is you would hire too big of a team too early."「我早期在产品职业生涯中犯的一个错误是,过早地雇佣了过大的团队。」Josh Woodward (Google Labs / Gemini VP) · Inside Gemini and NotebookLM | Josh Woodward
NotebookLM 爆红时,「很多人会问,移动应用程序在哪里?我们会说,我们甚至还没有移动团队」("we don't even have a mobile team yet",同上)。两人并非完全对立——Stein 自己的 AI Mode 也是 5-10 人起步("a former small team of folks … this was probably like five to ten people worth of people originally"/「这最初可能相当于五到十个人的工作量」),他的扩张触发器是"conviction moment"(「我认为主要是在你达到确信的时刻」)。Chesky 的 Hawaii 也走了同一条曲线:"It was just one team. I mean, it grew into dozens and dozens of people, maybe 50 or 60 people, but it was this really lean team."/「那仅仅是一个团队。它后来发展到几十人,也许 50 或 60 人,但它一直是一个非常精简的团队。」
真正不肯过桥的是 Edwin Chen——他把"到点就加人"本身视为邪教:
"The standard playbook is to get product market fit by pivoting every two weeks, and to chase growth and chase engagement with all of these dark patterns, and to blitzscale by hiring as fast as possible. And I've always disagreed. … You don't hire a team of 10 product managers because that's what every other cookie cutter startup does."「标准的策略是通过每两周调整一次来获得产品市场匹配度,并通过所有这些黑暗模式来追逐增长和参与度,并通过尽可能快地招聘来进行闪电扩张。我一直不同意。……你不要雇佣一个由 10 名产品经理组成的团队,因为这是其他千篇一律的创业公司所做的事情。」Edwin Chen (Surge AI CEO) · The 100-person AI lab | Edwin Chen (Surge AI)
AppLovin 站 Edwin 这边,且把"不扩张"做成了硬纪律——有 GPU 也不许用:
「我会跟团队下要求,我说你们不可以使用这些 GPU,即使我们拥有,你也不可以使用。你们必须找别的办法,把这个 GPU 的使用量给它降下来。……当你一个团队没有这样一个纪律,你总是允许大家去采取这种最简单的方式来解决问题的时候,人们就没有一个动力去寻找更高效、更优质的一个解决方案。」
— 葛小川 (AppLovin VP Engineering) · EP 68. 对话 AppLovin 技术 VP 葛小川
还有一个常被忽略的倒装:Stein 指出"长期保持小团队"是大公司才负担得起的奢侈(「所以你可以在更大的公司里这样做。但作为一家创业公司,我不知道你是否有这种闲暇」)——这与流行叙事里"小团队=创业公司专属"恰好相反。
裂缝三:AI 让小团队成为可能,还是小团队本来就对、AI 只是放大?
"AI 解锁说"的代表是 Julie Zhuo——AI 溶解职能边界,两人即一队:
"We need to dissolve the boundaries of these traditional roles. In the past, again, we would have a traditional team, engineers, product manager, designer, researcher, data scientist. And I think now the teams can look more like, well, it's just two people."「我们需要打破这些传统角色的界限。过去,我们有一个传统的团队,工程师、产品经理、设计师、研究员、数据科学家。我认为现在团队可以看起来更像是,嗯,只有两个人。」Julie Zhuo (Sundial CEO, 前 Facebook 设计 VP) · From managing people to managing AI | Julie Zhuo
Spiegel 同方向——AI 把"设计到代码"的链条压扁了:
"Well, now that's happening incredibly quickly because designers can ship code with, you know, all these new AI tools. … If they've got something we think it's cool, let's get it in the app and test it like immediately."「现在这种情况发生得非常快,因为设计师可以使用所有这些新的人工智能工具来发布代码。……如果他们做出了我们觉得很酷的东西,就把它放进应用程序里立即测试。」Evan Spiegel (Snap CEO) · Evan Spiegel, Snap
Chesky 把它升级为代际命题:"We need AI founder mode. AI founder mode is going to be even more intense than founder mode."/「我们需要 AI 创始人模式。AI 创始人模式将会比传统的创始人模式更加激烈。」——因为信息按需可得,35 小时会议这个旧机制可以整体退役。
但反方证据就埋在同一批材料里。AppLovin 的零会议访谈录制于上市前后(库内三篇中两篇为 2021 年前后的节目),远早于这一轮 AI;Baxley 的"六名设计师"是 2000 年代的 Apple;Edwin Chen 的"裁掉 90%"直觉来自他在大厂的年代。Julie Zhuo 自己也承认这一点:
"And so we found that if we actually make teams smaller, and we kind of even like in the past, you know, pre AI, like just have fewer of these, it allowed everyone to be like, oh, wait, we don't have a product manager on the team. Okay, so communication's up to me."「所以我们发现,如果我们真的把团队规模缩小——而且我们过去也这样做过,你知道,在 AI 之前,只是减少这些——这让每个人都觉得,等等,我们团队里没有产品经理。好的,所以沟通的责任就落在我身上了。」Julie Zhuo (Sundial CEO) · From managing people to managing AI | Julie Zhuo
把怀疑推到最远的是 Marc Andreessen——当下的"砍人、扁平化"浪潮根本不是 AI 驱动的:
"Essentially every large company is overstaffed. We can debate how much—it's at least overstaffed by 25%. I think most large companies are overstaffed by 50%. … And now they all have the silver bullet excuse, right? Ah, it's AI."「基本上每家大公司都人手过剩。我们可以争论具体超编多少,但至少超编了 25%。我认为大多数大公司都超编了 50%。……现在他们都有了万能的借口,对吧?啊,是 AI。」Marc Andreessen (a16z) · Marc Andreessen on Evaluating Founders and AI's Consumer Surplus
"AI until like literally until like December, it was not actually good enough to do any of the jobs that they're actually cutting. And so it just can't have been AI."「直到 12 月份,AI 实际上还不够好,无法胜任他们现在裁掉的任何工作。所以不可能是 AI 的原因。」Marc Andreessen (a16z) · 同上
裂缝四:扁平到底,还是保留小队结构(以及其他自我修正)
没有人真的主张无限扁平。Chesky 自己划了边界:
"I know there's this like general idea, like as a thought experiment, what if I could theoretically manage all 7,000 people flat? I don't think that's a good idea. I think that's an extreme, but I do think going to a few layers of management would make a lot of sense."「我知道现在有一个普遍的观点,或者说思想实验:如果我能理论上扁平化管理所有 7000 人会怎样?我认为那不是个好主意。我认为那是极端情况,但我确实觉得减少管理层级是非常合理的。」Brian Chesky (Airbnb CEO) · Brian Chesky - AI Founder Mode
ElevenLabs 取消了头衔,却保留了一层 leads,并且发现"扁平 + 全透明"反而有害——要靠裁剪信息访问来保护注意力:
"If you put a person into all the Slack channels and give them transparency, they actually get frequently distracted because then they read all the messages. You can still choose not to read them, but they still do. So you kind of need to cut the access to a lot of those pieces to force the attention."「如果你把一个人放到所有的 Slack 频道中,并给予他们透明度,他们实际上经常会分心,因为他们会阅读所有的消息。你仍然可以选择不阅读它们,但他们仍然会读。因此,你需要在某种程度上切断对很多信息的访问,以迫使他们集中注意力。」Mati Staniszewski (ElevenLabs CEO) · ElevenLabs CEO: Why Voice is the Next AI Interface
同一个 Mati 还交代了一次理念撞墙——"工程师做销售"的扁平理想直接失败:
"Our initial take was, of course, we want to be an engineering company. We don't want salespeople. … We told them, do sales now. And that really, as you can imagine, didn't work out in this specific case. But we learned our lesson and we now do invest in a combination of that. It's 80% sales, 20% engineering."「我们最初的想法是,我们当然想成为一家工程公司。我们不想要销售人员。……我们告诉他们,现在去做销售。正如你可以想象的那样,这并没有奏效。但我们吸取了教训,现在投资于两者的结合。80% 是销售,20% 是工程。」Mati Staniszewski (ElevenLabs CEO) · 同上
Julie Zhuo 也给"人人皆 builder"加了脚注:「我并不是再次强调每个人都需要擅长所有事情。我认为这不太现实。」("I'm not trying to say again that everyone needs to be good at everything. I don't think that's very realistic.")
地理维度上同样没有共识:"异步"不等于"远程"。AppLovin 一边消灭会议一边发钱把人聚到办公室;ElevenLabs 全远程起家,30 人后开始建 London/Warsaw/SF hub("beyond 30 people, we realized that the new people joining, there's benefit of them having a space to be next to others"/「超过 30 人后,我们意识到新加入的人,让他们有一个与其他同事在一起的空间……是有好处的」);Chesky 则说"我们基本是远程的,异步化对我们有利"。Andreessen 给出宏观判断——远程化浪潮被 AI 整个掀翻:
"I was very enthusiastic between 2020 and let's say 2023 that we had cracked the code on how to finally get away from the geographic constraints of Silicon Valley. I think in the last two years, I think that that process has like whiplash reversed in an incredible way. And I think the tech industry is more centralized in Silicon Valley than it has been in its entire existence. And I think it's AI."「在 2020 年到 2023 年之间,我非常乐观地认为,我们已经找到了如何摆脱硅谷地理限制的方法。我认为在过去的两年里,这个过程发生了惊人的逆转。我认为科技行业在硅谷的集中程度比以往任何时候都高。我认为这是人工智能。」Marc Andreessen (a16z) · Marc Andreessen on Evaluating Founders and AI's Consumer Surplus
都没说透的
1. 协调成本去哪了,没人算账。 零会议、无中层的组织里,跨团队协调并没有消失——证据显示它流向了两处:要么变成重复建设(Mati 承认 20 个自治团队"carries some of the sometimes issues of duplicative work"/「会带来重复工作或进度不一致的问题」),要么压到创始人一个人身上。Foroughi 的原话:"It continues to build to a point where I can never catch up and that's stressful. … I just sleep less."/「它不断累积到一个我永远无法赶上的地步,这很有压力。……我只是睡得少了。」没有人把这两项成本和省下的会议时间放在同一张账上比过。
2. 这是组织设计,还是天才依赖? Surge 发给客户的深度分析「很多时候我们会说是来自我们的数据科学团队,但通常实际上只是我做的」(Edwin Chen);AppLovin 的 CEO/CTO 亲手写代码,文化判断是「你可以把谷歌的算法 Copy 一份搬到另一个公司去,六个月之后这个公司不可能和谷歌一样」(葛小川)。如果参数区间的成立依赖几个超载的明星个体,"小团队"是可复制的结构,还是不可复制的人?关键人离开后会怎样,没有一篇访谈触碰。
3. 适用边界只被一句话带过。 13 篇样本全部是软件、数据、广告、设计类公司。Robby Stein 提了一句"if you're building a single product that is a way to do something with a digital app that's fairly straightforward, that's going to be different than building a robotics company"/「如果你正在构建一个简单的数字应用程序……这将不同于建立一家机器人公司」,然后话题就走了。硬件(Snap 的眼镜做了 12 年)、受监管行业、重运营业务是否适用同一参数,无人展开。
4. 中层消失后,人的成长由谁负责? Chesky 说纯人事经理没有价值、"你不是他们的心理医生";但 Julie Zhuo 整本书写的就是 manager 对人的发展职能,她只论证了 agent 不需要这一项,没有回答人还需不需要。AppLovin 的替代方案是 Foroughi 式的高压师徒制——"I'm very aggressive and hard with the people I respect the most … The more respectful and quiet I am towards someone, the less I believe that I can hand something off to them."/「对于我最尊重的人,我非常严厉……我对某人越是客气和安静,我就越不相信我可以把事情交给他们。」这套机制高度依赖他本人的判断和带宽,而他自己承认"I do a horrible job making myself a mentor that I want to be"/「我在把自己打造成我想要成为的导师方面做得非常糟糕」。规模化路径是空白。
我的看法
以下是判断,不是材料归纳。一,收敛是真的,但收敛点不是"小"本身,而是"删掉信息传递层"——小团队、零会议、少层级、manager-IC 是同一刀法的四个切面,Chesky 的"游戏电话"和 Baxley 的"披头士"说的是同一件事。二,AI 不是这套打法的起因(AppLovin 2021 年、Apple 2000 年代就在用),而是第一次把"删掉传递层"的可行规模从百人推向千人、并给了大公司动手的政治借口(Andreessen 的"silver bullet excuse"反向证明了这点)。三,真正未决的只有两个参数:会议(Spiegel 式作品评审会与 Foroughi 式零会议都在赢,说明变量可能不是"开不开会"而是"会议里有没有作品")和扩张时点(Stein 的 conviction-trigger 与 Edwin Chen 的永远不扩,押的是两种不同的产品类型)。把握程度:中等偏高——13 篇全是赢家自述,没有失败组对照,幸存者偏差无法排除。
还想知道什么
1. AppLovin 的零会议在跨时区团队上的真实运转情况——北京 10 人团队与湾区之间是否真的零预约会议,还是靠葛小川单点桥接?有没有因异步失败而回退的案例? 2. Robby Stein 说 close friends 功能"因团队太小拖死"——具体拖了多久、最终加到多少人才解决?AI Mode 从 5-10 人到正式上线时团队规模的完整曲线是什么? 3. ElevenLabs 量化过 20 个自治团队的重复建设成本吗?有没有团队在六个月存活期规则下被砍掉后、其成员流向何处的数据? 4. 这批"无头衔/无纯管理者"组织的员工留存与成长数据:年轻人在没有中层导师的结构里 2-3 年后的去留,与传统结构相比如何?
取材
- Brian Chesky - AI Founder Mode - [Invest Like the Best, EP.470] · 2026-05-11
- Evan Spiegel, Snap · 2026-04-13
- Marc Andreessen on Evaluating Founders and AI's Consumer Surplus · 2026-04-02
- The 100-person AI lab that became Anthropic and Google's secret weapon | Edwin Chen (Surge AI) · 2025-12-09
- ElevenLabs CEO: Why Voice is the Next AI Interface · 2025-11-06
- Inside Google's AI turnaround: The rise of AI Mode | Robby Stein (VP of Product, Google Search) · 2025-10-15
- Inside Gemini and NotebookLM: How Google is Shipping Non-Stop | Josh Woodward · 2025-10-15
- E928 Adam Foroughi AppLovin: mobile game boom, bootstrapped to profit · 2025-09-26(节目录制于 IPO 前,约 2019)
- 20VC: AppLovin's Adam Foroughi on How To Run a 1,000+ Organisation with Very Few Meetings · 2025-09-26(节目录制于 2021 上市前后)
- EP 68. 对话 AppLovin 技术 VP 葛小川:2 年市值增长 25 倍 · 2025-09-26
- From managing people to managing AI | Julie Zhuo (Facebook VP, Sundial CEO) · 2025-09-22
- GPT-5 and Agents Breakdown – w/ OpenAI Researchers Isa Fulford & Christina Kim · 2025-08-10
- 35 years of product design wisdom from Apple, Disney, Pinterest and beyond | Bob Baxley · 2025-06-18