2025-08-18 Acquired.How is AI Different Than Other Technology Waves?

Refer To：《How is AI Different Than Other Technology Waves?》。

Transcript: (disclaimer: may contain unintentionally confusing, inaccurate and/or amusing transcription errors)

Ben: Hello, Acquired listeners. We have a very special treat for you today. We have an episode here with Bret Taylor and Clay Bavor. David, this is an awesome conversation.

Ben：Acquired 的听众朋友们，大家好。今天给你们带来一个特别的惊喜。这期我们请到了 Bret Taylor 和 Clay Bavor。David，这次对话太棒了。

David: We've gotten to know Bret and Clay a bit over the years and especially in doing research for our Google episode. They were both incredibly helpful because they both started their storied careers as APMs, associate product managers, at Google back in the early days. Helped with part one, they're helping with part two. They're now co-founders of Sierra together. Really, they both have had among the most incredible careers in tech of the last 20 years.

David：这些年我们逐渐认识了 Bret 和 Clay，尤其是在准备我们的 Google 专题时。他们帮了我们大忙，因为他们都在 Google 早期以 APM（associate product managers，助理产品经理）身份开启了各自的传奇职业生涯。第一部分他们帮过，现在第二部分也在帮。他们现在一起创办了 Sierra。说真的，在过去 20 年的科技圈里，他们两位的履历堪称最令人惊叹的之一。

Ben: Yes, Bret's done various things. Clay has done one thing or many things inside Google, but Clay was at Google for over 18 years. He started in the APM program, worked on everything from ads at the beginning.

Ben：没错，Bret 做过很多事。Clay 在 Google 里“做过一件事也像是很多事”，总之他在 Google 呆了 18 年多。他从 APM 项目起步，一开始就从广告做起，几乎什么都碰过。

David: Led product for Gmail, Drive, Docs, all the apps, everything that became workspace.

David：他负责过 Gmail、Drive、Docs 等一众应用的产品，后来这些都成了 Workspace 的一部分。

Ben: Eventually ran Google Labs, did a bunch in their hardware, VR, AR, future looking screens. Bret, his name pops up in every episode research. He was at one point the CTO of Facebook. This is after his Google days where he of course co-founded Google Maps. He was the co-founder of Friend Feed, which I loved Friend Feed as big user before that was bought by Facebook.

Ben：他后来还掌管过 Google Labs，在硬件、VR、AR、前瞻显示等方面做了不少事情。至于 Bret，我们做每期研究基本都会碰到他的名字。他曾担任过 Facebook 的 CTO；在此之前的 Google 时期，他当然是 Google Maps 的联合创始人。后来他又联合创办了 Friend Feed——我当时可是重度用户，直到它被 Facebook 收购。

David: Started Quip, got acquired by Salesforce.

David：他创办了 Quip，之后被 Salesforce 收购。

Ben: Then became the co-CEO of Salesforce. Still actively writes code by the way, referenced on this episode.

Ben：然后他成为 Salesforce 的联席 CEO。顺便说一句，他现在还在积极写代码，这期节目里我们也提到了。

David: We have a fun little bit at the end where we're like, okay, which of you created more market cap in your career, Bret, through your incredible journey, or Clay just by cranking on Google?

David：节目的最后我们还有个有趣的小环节：你们俩谁这辈子“创造的市值”更多——是 Bret 的传奇旅程，还是 Clay 在 Google 的长期深耕？

Ben: Bret is the current chairman of the board at OpenAI and I think we didn't talk about on this episode, but is also crazy, was the chairman of the board at Twitter in its final moments as a public company. Two legendary figures to sit down and talk with.

Ben：Bret 目前是 OpenAI 的董事会主席，而且我们这期没细说但真的很疯狂——他还是 Twitter 作为上市公司最后时刻的董事会主席。两位传奇人物同台对谈，难得的机会。

不太好的经历。

In this episode, we talk about everything AI. There's the great conversation of is AI a giant step forward change in the world, or is it just better software? And what are all the second order effects of all the change that's going on with AI? We talk about Sierra, the company that they are currently building together and a lot of little tech history tidbits, especially as it relates to our Google episode two. Please enjoy our conversation with Bret Taylor and Clay Bavor.

在这期节目里，我们聊遍了 AI：AI 究竟是改变世界的一大步，还是“更好的软件”？围绕 AI 的种种变化会带来哪些二阶效应？我们也聊到他们正在一起打造的公司 Sierra，以及许多与科技史相关的小花絮，特别是和我们的 Google 第二期有关的内容。请享受我们与 Bret Taylor 和 Clay Bavor 的对话。

Great to have you guys here.

太高兴你们来了。

Clay: Thanks for having us.

Clay：谢谢邀请。

Bret: Thanks for having us.

Bret：谢谢邀请。

Ben: This feels like a special moment for us here at Acquired. You both have helped so much with past episodes. They've sent us nice little notes with corrections and tidbits, and here's how we think about it. Thank you both for that.

Ben：这对我们 Acquired 来说是个特别的时刻。你们过去帮了我们很多；还给我们发来过纠正和小花絮的友好笔记，分享你们的看法。非常感谢你们。

Clay: It's so fun to be a part of it and I thought part one just nailed it, in particular in the later parts of it just how Google really got distribution, Toolbar, Google Pack, Google Earth, and so on. I loved listening to it.

Clay：很高兴参与其中。我觉得第一部分做得特别好，尤其是后半段讲到 Google 是怎么拿到分发的：Toolbar、Google Pack、Google Earth 等等。我听得很过瘾。

David: The Google Earth story of the spyware of toolbar. It's included in the install package.

David：还有关于 Google Earth 的那个故事，以及 Toolbar 被当成“间谍软件”的争议——它被打包进了安装包里。

Ben: Careful, David, with that spyware word.

Ben：小心点，David，别随便用“spyware（间谍软件）”这个词。

Bret: It's a bundle. It's a bundle, David.

Bret：这是捆绑。就是捆绑，David。

David: Yeah. That's a word that has disappeared from the lexicon.

David：对。那个词已经从大家的词汇里消失了。

Bret: I've been thinking about that actually in the age of AI because if you look at the early internet, you had archaic terms like information superhigh. Remember webmasters?

Bret：在这个 AI 时代我一直在想这事儿，因为如果你回看早期互联网，有很多现在显得古老的术语，比如 information superhighway（信息高速公路）。还记得 webmaster（站长）吗？

Ben: Yeah.

Ben：记得。

Bret: Those were the people who maintained websites.

Bret：那就是维护网站的人。

Ben: That was my first job title.

Ben：那是我第一份工作的头衔。

Bret: I wonder now with AI, we have all these terms, AI engineer, AI architect, all these things, you wonder what's going to stick and what's going to feel like information superhighway. Anyway, I think a lot about that.

Bret：我在想现在有了 AI，我们又冒出一堆术语，像 AI engineer、AI architect 之类的。哪些会留下来，哪些未来看起来会像当年的 information superhighway 一样过时？总之，我经常想这些问题。

Ben: Do you have any beliefs? It seems like we've already seen prompt engineer enter and exit the lexicon.

Ben：你有没有什么判断？看起来“prompt engineer（提示工程师）”这个词已经在我们的词汇里进进出出了。

Clay: One of my favorite things to do is go to the internet archives way back machine and look at a company's website. When they transition from ML to AI and then AI to Gen AI, and then from Gen AI to agents and agentic. There is a lot of jargon in the space right now, and we try to keep it simple. I don't know what the information superhighway equivalent is yet, but I'm sure it's there.

Clay：我最喜欢做的一件事是去 Internet Archive 的 Wayback Machine 看各家公司的官网，观察它们如何从 ML 到 AI、再从 AI 到 Gen AI、然后从 Gen AI 到 agents 和 agentic 的转变。这个领域现在术语太多了，我们尽量保持简单。我还不知道哪个词会成为当代版的信息高速公路（那种后来看着过时的词），但我确信会有。

Bret: My hypothesis is actually the word agent will stick. I like the nouns of these technologies. Web had sites, mobile has apps, AI has agents, and I think it's going to stick for that reason. A little bit like app, the word app, the VC community, and 2013 was a hot word. Now it's just a noun that describes a packaging for a piece of technology. I think agent will go that way. It'll feel extremely novel, shiny, and complex now. Then it will start to be, oh, this is just a digital autonomous thing like we have a billion of in our lives. I think the word agent will stick, but we can talk in 10 years and we'll see if I'm right.

Bret：我的看法是“agent”这个词会留下来。我喜欢这些技术的“名词化”表达：Web 有 sites，移动端有 apps，AI 有 agents——我认为正因此它会留下来。有点像当年的 app，这个词在 2013 年在 VC 圈是个热词，如今只是描述一种技术封装形式的普通名词。我觉得 agent 会走同样的路。现在它看起来特别新、特别亮、也很复杂；但慢慢大家会觉得：哦，这不过是一个数字自主体，就像我们生活里已经有无数这样的东西一样。我认为“agent”会留存下来——十年后我们再聊，看我说得准不准。

Ben: Have you evolved the lexicon of how you describe Sierra? You haven't had that long of a life as a company, but it seems like there's already been a tremendous amount of change in AI since you started.

Ben：你们描述 Sierra 的“词汇体系”有没有演变？公司成立时间不算长，但自从你们起步以来，AI 领域似乎已经发生了巨大的变化。

Clay: In a nutshell, what we help companies do is build their own customer facing AI agents for all parts of their customer experience. We think in the future, your AI agent will be more important than your website and more important than your app. It will be the main way you interact with your customers.

Clay：简单说，我们帮助企业为其客户体验的各个环节构建面向客户的 AI agents。我们认为未来，你的 AI agent 会比你的网站更重要、也比你的 app 更重要——它将成为你与客户交互的主要方式。

In terms of how we've talked about it outwardly, actually when we launched the company just over 15 or 16 months ago, in 2024, we were worried when we said in the launch blog post, every company needs an agent that people wouldn't know what we're talking about. Fast forward just 12-15 months, my goodness, agents are everywhere. I think people understand it.

就我们对外的表述而言，其实在我们 2024 年创办公司、也就 15、16 个月前发布的启动博文里，当时写下“每家公司都需要一个 agent”时，我们还担心大家听不懂我们在说什么。快进 12–15 个月，天哪，agent 无处不在。我想大家已经理解了。

Bret: Now it's like, ah, the word agent again. It was really cool 15 months ago.

Bret：而现在大家会觉得：哎，又是“agent”这个词。15 个月前它还很酷呢。

Clay: I was speaking with the CIO of one very large retailer. He stopped me about 15 minutes in and said, Clay, can I just thank you for not saying the word agentic in the first 15 minutes here? Just trying to keep it straightforward.

Clay：我前阵子和一家超大型零售商的 CIO（首席信息官）交流，聊到 15 分钟左右他打断我说：Clay，先谢谢你这 15 分钟里没说“agentic”这个词。我们就尽量讲得直白点吧。

David: I can't believe that Sierra's only less than 18 months old.

David：我简直不敢相信 Sierra 才不到 18 个月大。

Bret: We started the company a little before that, but we told the world what we were doing less than 18 months ago, and it's insane. The way I think about these technology trends is they layer on top of each other and compound. To put a PC on every desktop, which was I believe Microsoft's mission in the early days...

Bret：我们稍早一点就开始做这家公司了，但真正向外界说明我们在做什么是在不到 18 个月前，发展速度快得惊人。我看待技术趋势的方式是：它们一层叠一层，产生复利效应。早年 Microsoft 的使命我记得是“让每一张桌子上都有一台 PC”……

David: On every desktop running Microsoft software.

David：每一张桌子上的 PC 都运行 Microsoft 的软件。

Bret: Of course.

Bret：当然。

Clay: Webvan too in there.

Clay：Webvan 也在其中。

Bret: Webvan. Yeah, exactly. Almost to a tee. If you say dot-com, people come back with the failures. If you look at the S\&P 500 now and you look at the amount of value from companies created in that, one could argue that actually almost all of the exuberance and hype was totally warranted and in fact did change commerce and fundamental ways. It did change the financial system in fundamental ways. It changed everything.

Bret：Webvan。对，没错。几乎一模一样。你一提 dot-com，大家就会想到那些失败案例。但如果你看看现在的 S\&P 500，看看其中有多少价值来自那一波诞生的公司，可以说当年的狂热与炒作几乎都是有根据的，而且确实以根本性的方式改变了商业。它也以根本性的方式改变了金融体系。它改变了一切。

My guess is we're in a similar era. You have a lot of steak oil, the jokes you were just saying about people, God say the word agentic again. Oh, my God. It's coupled with ChatGPT growing faster than any consumer product in history. You look at the revenue growth of companies like Sierra, you mentioned Lovable, and all these other B2B software companies. I think there's very real value being created here.

我的直觉是我们正处在一个类似的时代。你能看到很多“snake oil（忽悠式的噱头）”，就像你刚才调侃的那样——一听到别人又说“agentic”，大家就吐槽：天哪，别再说了。与此同时，ChatGPT 的增长速度是史上最快的消费级产品。再看看像 Sierra、你提到的 Lovable 以及其他这些 B2B 软件公司的收入增长。我认为这里正在创造非常真实的价值。

Fundamentally, software is not just adding productivity to workplaces and to individuals, but actually completing work. That's where the word agent comes from, agency and reasoning. I think we're going to see this really significant uptick in productivity, and that's going to be coupled with B2B software companies who are selling this, sharing some of the upside of that productivity enhancement. For consumers, I self-identify as a computer programmer. That's the thing I love to do the most. Do you remember the calculators, the people who calculated things before we had calculators? I'm like, am I that?

从根本上说，软件不仅仅是在职场和个人层面“提升生产率”，而是把工作真正完成。agent 这个词背后是 “agency” 与 “reasoning” 的含义。我认为我们会看到生产率出现非常显著的跃升，而销售这类能力的 B2B 软件公司会分享这部分生产率提升所带来的红利。对消费者而言，我把自己视为一名 computer programmer——那是我最热爱的事情。你还记得 “calculators” 吗？在电子计算器出现之前专门做计算的人。我会想：我会变成那样吗？

Ben: They were computers too. They were also called computers, people who compute.

Ben：他们也是 computers。当时也被称为 computers——就是做计算的人。

Bret: Yeah. The thing I self-identify with is being obviated by this technology. The reason why I think these tools are being embraced so quickly is they truly are like an Ironman suit for all of us as individuals. I think we're going to look back at this era and we're going to joke around about whatever turns into information superhighway, like the terms that get antiquated. I think we'll also look back and say, this was an inflection point in society and technology, and I think it will be as significant as the advent of the internet.

Bret：是的。我所认同的那件事正在被这项技术取代。我之所以认为这些工具被如此快速地接受，是因为它们真的就像给每个人都穿上了一套 Ironman 战衣。将来我们回看这个时代，会拿那些后来变得陈旧的术语打趣——就像当年的 information superhighway 一样。我也认为我们会说：这是社会与技术的一个拐点，其意义将不亚于互联网的诞生。

David: Yeah, it's wild. Thinking about the acceleration of adoption of these successive waves...

David：是啊，太疯狂了。想到这些连续浪潮的采用速度在不断加快……

Ben: Which by the way, I looked it up, ChatGPT was five days to a million users and two months to a hundred million users.

Ben：顺便说一句，我查了下，ChatGPT 用了 5 天达到 100 万用户、2 个月达到 1 亿用户。

Clay: That sounds right.

Clay：听起来对。

David: I don't think we're quite there yet. You guys would know better than us if we are or aren't and what it's going to look like. Once we figure out distribution of the equivalent of an application layer on top of AI, ChatGPT, et cetera, all the friction to adoption and distribution is just gone. Given in the dot-com website, you could argue there's no friction to type in a website or we go to a service, but the human has to become aware, have interest, the whole sales, probably interest, awareness, decision, action. That's just gone.

David：我不认为我们已经到那一步了。至于是与否、以及会是什么样子，你们比我们更清楚。一旦我们搞清楚在 AI（比如 ChatGPT）之上相当于“应用层”的分发机制，采用与分发的所有摩擦就会消失。回到 dot-com 时代，你可以说输入网址去一个服务几乎没有摩擦，但人仍然需要知晓与兴趣——整个销售漏斗：兴趣、认知、决策、行动。那一整套就会消失。

Ben: Why David? No. You still have to learn that a thing exists.

Ben：为什么这么说，David？不对吧。人们还是得先知道某个东西存在。

David: No. Say ChatGPT, if that becomes the front door adoption for new services and businesses built integrated into it, ChatGPT's just going to figure it out and serve it to you. We saw this with Studio Ghibli, right?

David：不。比如 ChatGPT，如果它成为承载新服务与新业务（与其深度集成）的入口，ChatGPT 自己就会发现并把它推送给你。我们在 Studio Ghibli 这件事上已经见过这种情况，对吧？

Bret: I think it's going to really upend the internet in pretty meaningful ways. You talked about the awareness adoption, et cetera. If you look at the market on the internet now, you have demand generation and discovery, which is right now dominated by social media and the ad networks affiliated with social media. Then you have demand fulfillment, which used to be search, AdWords, and all of this, and that used to be still is obviously. Then you have the actual transactions themselves, the commerce systems and other things like that.

Bret：我认为这将以非常深刻的方式颠覆互联网。你刚才提到了认知与采用之类的问题。看看当今互联网的市场：有需求的生成与发现，这部分目前主要由社交媒体及其关联的广告网络主导；然后是需求的兑现，过去由搜索、AdWords 等承担，而且现在显然仍然如此；最后是真正的交易环节，即各类电商系统等。

Right now, you could say AI is impacting all of those products in very meaningful ways. As you alluded to David, let's just just say that personal agents become a thing. How does that impact that entire funnel? When you're generating demand, that probably will still be relevant for individuals, but are you going to be generating demand for people? Are you going to be generating demand for their agents? What does that even mean?

现在你可以说，AI 正在以非常有意义的方式影响上述所有产品。正如你暗示的，David，假设 personal agents 成为现实，那对整条漏斗意味着什么？当你在生成需求时，它大概仍然与“个人”相关，但你是在为人生成需求，还是在为他们的 agents 生成需求？这到底意味着什么？

David: If it's other agents making the decision of whether to interact or not, that whole demand gen cycle takes time with humans.

David：如果最终决定“要不要互动”的是其他 agents，那么原本在人类那里需要花时间的那一整套需求生成过程就会被改变。

Bret: You have discussed pricing strategy in a bunch of your different shows, which is really interesting. There's a classic, like have a really expensive product, one slightly less expensive below that, how people react to it psychologically, and all these other things. If it's an agent, what happens there? Will these things trend towards the mathematically optimal? If you think about these platforms, brands don't want to be disintermediated and they don't want to be commoditized.

Bret：你们在很多节目里都讨论过定价策略，这非常有意思。经典玩法比如：放一个非常昂贵的产品，下面再放一个稍微便宜点的，看看人们在心理上的反应，诸如此类。如果对象变成 agent，会发生什么？这些事情会不会朝“数学上最优”的方向收敛？从平台的角度看，品牌不想被去中介化（disintermediated），也不想被商品化（commoditized）。

The platform providers, they don't explicitly say it, but they want to distribute and commoditize everything. It's not like a formal strategy, but that's the natural tension of these platforms. With personal agents and agents like Sierra builds for companies that represent their customer experience, I think the second and third order effects are very hard to predict here. What does it do for the demand generation, demand from the ad market, what does it do for these platforms? Which companies will have their own agents and have enough brand equity to have their own agents, and which companies will be dependent on? It's a little bit saying which retailers depend on Instagram ads versus first party discovery.

平台提供方虽然不会明说，但他们希望把一切都分发、都“商品化”。这不算一项正式战略，但却是这类平台的内在张力。随着个人 agents 以及像 Sierra 为企业构建、代表其客户体验的 agents 出现，我认为这里的二阶、三阶效应很难预测。它会怎样影响需求生成、广告市场的需求，会怎样影响这些平台？哪些公司会拥有自己的 agents、并具备足够的品牌资产去运营自己的 agents，哪些公司又会依赖他人？这有点像问：哪些零售商依赖 Instagram 广告，哪些依赖 first-party discovery（第一方自有发现）。

I think we're at the cusp of something that we will, in five years, have a very different market on the internet. I think even for people in the middle of it, like Clay and me, it's very hard to predict. I don't think many of us predicted many of the second or third order effects of the mobile app store or social networks correctly. I think this one's even harder to predict, but I think it's going to really upend the economy of the internet in significant ways.

我觉得我们正处在一个临界点，五年后互联网的市场格局会非常不同。即使对身处其中的人，比如我和 Clay，也很难预测。当年关于移动应用商店或社交网络的二阶、三阶效应，很多人都没预测对。我认为这次更难，但它将以重大的方式颠覆互联网经济。

Clay: Bret and I like betting. One of the bets we have is the year in which greater than 50% of conversations with agents built on Sierra are with people's personal agents, so agents talking to agents.

Clay：Bret 和我都喜欢打赌。我们有个赌注是：哪一年会出现这样一个时点——与 Sierra 构建的 agents 的对话中，超过 50% 是与人们的 personal agents 进行的，也就是 agent 和 agent 的对话。

David: Okay, yeah. What's the bet?

David：好啊。赌的具体内容是什么？

Clay: No, we can't reveal the number. Can't reveal the date.

Clay：不行，我们不能公布具体数字，也不能公布日期。

David: Company secret.

David：公司机密。

Ben: Because you think you'd impact the outcome or?

Ben：因为你们觉得公开会影响结果吗，还是？

Bret: Clay has so far won all three bets that we've had.

Bret：到目前为止，我们打的三个赌都是 Clay 赢了。

Ben: Really? What are some of the other ones?

Ben：真的？还有哪些赌注？

Bret: The theme is Clay is more optimistic than I am.

Bret：主题就是：Clay 比我更乐观。

David: Optimism always wins.

David：乐观总能取胜。

Clay: One of the first was we had a bet as to what percent of all incoming customer issues one of our agents could resolve. Bret was maybe the pessimist and realist, favorably charitably. I bet we'd exceed 80% by the end of the year, and we did. It just exceeded every expectation.

Clay：最早的一个赌是：我们的某个 agent 能解决所有新进客户问题的比例会达到多少。客气点讲，Bret 比较悲观、也更现实。我打赌年底能超过 80%，结果我们做到了，而且远超所有预期。

Bret: I was at 50, just for reference. Clay not only won the bet, but won handily.

Bret：我当时押的是 50%，仅供参考。Clay 不但赢了，而且赢得很轻松。

Clay: By a significant margin.

Clay：而且优势显著。

Ben: This is across all your customers whenever a new issue is originated with one of their customer service requests, how often the Sierra agent could handle it?

Ben：这是指在你们所有客户中，只要出现一条新的客户服务请求，Sierra agent 能处理的比例是多少吗？

Clay: That's right. Looking at a particular customer. These agents, it's so neat. They're not only answering questions but doing things like, if you are moving from car A to car B and have a SiriusXM subscription, Harmony, which is SiriusXM's agent, can actually send a satellite signal from space to refresh the encryption keys on your car. ADT is agent that we built with and for them. If your alarm panel starts beeping and you don't know why, it can troubleshoot, could figure out which of the 52 different panels you have, and then mail you a battery itself if that's the issue.

Clay：对。这里说的是某个特定客户的情况。这些 agents 真是很厉害——它们不只是回答问题，还能执行操作。比如你把车从 A 换到 B，且订阅了 SiriusXM，SiriusXM 的 agent——Harmony——真的可以从太空发送卫星信号，去刷新你车辆上的加密密钥。还有 ADT 的 agent，是我们与他们共同构建、并为他们打造的。如果你的报警面板开始滴滴作响而你不知道原因，它可以进行故障排查，识别你用的是 52 种面板中的哪一种，如果问题是电池，它甚至会直接给你寄一块电池。

Bret: By the way, just think of that for a second. You're talking to an AI that's talking to a satellite that's sending something to your car, no people involved. Yeah, of course it does that. It's science fiction three years ago. Now you're like, yeah, of course the AI is talking to a satellite. No big deal.

Bret：顺便说一句，想象一下：你在和一个 AI 说话，它在和一颗卫星说话，那颗卫星再向你的汽车发送东西，全程没有人参与。是的，当然能这么做。三年前这还像科幻，如今你会说：对啊，AI 跟卫星对话很正常，没什么大不了的。

Clay: I remember, one of our earlier customers, OluKai, they sell great flip flops by the way if you're in the market, on the day that one of our agents successfully processed a warranty looking at photos, inspecting the photos to make sure it was the product in question, and then shipped out a new pair of replacement shoes all on its own, there was cheering in the office. It's just neat, these agents interacting with the physical world.

Clay：我记得我们早期有个客户叫 OluKai——顺便说一句，如果你需要，他们的人字拖做得很棒。那天我们的某个 agent 成功处理了一个保修请求：它查看照片、核验照片里的确是相关产品，然后完全自主地寄出了一双替换鞋，办公室里当场一片欢呼。看着这些 agents 和物理世界互动，真的很神奇。

Bret: By the way, this is a true story. I'm in an interview with a candidate, and the whole office is like, yeah. I come in and they're like, we exchanged some flip flops, woo. It was a moment. You had to be there, but it was a very exciting moment for us.

Bret：顺便说这是真事。我当时正在面试候选人，整个办公室都在“耶！”我走进去，他们说：我们刚换出了一双人字拖，呜呼！那是个值得纪念的瞬间——你得在场才懂，但对我们来说真的非常振奋。

Ben: Which must be an amazing way to demonstrate culture to a candidate by the way. They come out of an interview, they're like, what the hell, what's going on?

Ben：这对候选人来说也一定是展示公司文化的绝佳方式。他们一出面试室就会想：这什么情况，发生了什么？

Bret: Yeah, that was fun. It was just a hilarious moment because it was so explaining, it felt so trivial. Let me explain why flip flops are a big deal to us, but trust me, this is a big deal.

Bret：是啊，挺有趣的。好笑的是，一旦开始解释就显得很琐碎——“让我解释一下为什么人字拖对我们来说很重要”，但相信我，这事真的很重要。

David: That's like, were you there? It might've been a little before your time, but when Facebook was negotiating with Microsoft for the ad deal and they were having the hackathon to launch international, that same night, and it was all orchestrated all together. There's house music playing and all the middle aged Microsoft execs. Like, what the hell is happening here?

David：这有点像——你在场吗？也许那会儿你还没入职——当时 Facebook 正和 Microsoft 谈广告合作，同时在同一晚搞了一场“全球化发布”的 hackathon，一切都编排在一起。现场放着 house 音乐，一群 Microsoft 的中年高管在旁边一脸懵：这到底在干嘛？

Bret: The hackathons at Facebook were epic. There's usually Mark Slee DJ-ing. It was great. It was fun.

Bret：Facebook 的 hackathon 简直传奇，通常都是 Mark Slee 在做 DJ。太棒了，特别好玩。

Ben: Some days I wake up and I'm on one side of this debate, and some days I wake up on the other side of this debate. The first way is AI, is such a transformational technology. It is different than anything that's ever come before it. Let's just look at labor productivity. It makes people so much more productive that you need way fewer people to make things. It's going to put all these people out of work. Every societal model that we've had in the past is now broken because this is such a giant step change for everything.

Ben：有些天我醒来会站在这场争论的一边，另一些天醒来又站在另一边。一种看法是：AI 是颠覆性的技术，和以往任何东西都不同。就看劳动生产率吧——它让人的生产率大幅提升，以至于需要的从业者大为减少，会让很多人失业。我们过去的所有社会模型都被打破了，因为这对一切都是一次巨大的跃迁。

The other days I wake up and I'm like, it's just software. Mobile apps were great. Software, SaaS was great, Software, Cloud was an amazing way to run your software. Software gets better with Moore's Law and more powerful, more sophisticated, and this is just more powerful software. How do you guys think about that?

而另一些天我醒来会觉得：它就是软件。移动应用很好，软件、SaaS 很好，软件、Cloud 是运行软件的绝佳方式。软件随着摩尔定律不断变强、变复杂，这不过是“更强的软件”。你们怎么看？

Bret: The first principle's way I think about it is, what are you making plentiful now that was scarce before, and how does that impact society? I think about taking energy and making it scarce to plentiful to the point now where you walk into a room, you flip on a light switch, and you don't think anything about it. For hundreds of years, that was a scarce resource.

Bret：我从“第一性原理”的角度想：你把过去稀缺的东西变得充裕，那么这会如何影响社会？比如能源——从稀缺到充裕，直到今天你走进房间随手一按灯就亮，完全不再多想。但在几百年的历史里，能源曾经是稀缺资源。

David: It is now, again, with data centers.

David：而现在，随着数据中心，这又变得稀缺了。

Bret: Yeah, that's right. We're doing our best.

Bret：没错。我们正在尽力。

Clay: We're the token mock fusion reactors.

Clay：我们就是“token mock”聚变反应堆。

Bret: I think in the Western world, similarly, food is largely plentiful. Food insecurity, while present, isn't a dominant part of society as it once was. Now we're going to a world where intelligence has gone from something scarce to something plentiful. I think it's very hard to imagine prior to modern farming and food distribution, most people spend a lot of their time thinking about food. That was a big part of just living, and now it's something that is for a lot of people, not a central part of their day-to-day plan. It's acquiring food.

Bret：我认为在西方世界，同样地，食物总体上是充裕的。尽管仍然存在粮食不安全，但它已不再像过去那样是社会的主导性问题。现在我们正走向一个“智能”从稀缺变为充裕的世界。我觉得很难想象在现代农业与食品分配出现之前，大多数人会把大量时间花在获取食物上——那是生活中的重要一部分；而如今，对很多人而言，这已不再是日常计划的核心。

First, I think we have gone through transitions as significant of this in the past as society, but I think it's very significant. I don't think it's just software. That's my personal opinion, although I do waver like you, Ben. Some days I'm on one side of this on the other. I think it's pretty significant because I know. I personally and probably everyone on this podcast identifies, to some degree, their identity with their intelligence. It's a big part of why people listen to your podcast. It's how you got into university, got a job, and got all these other things. You say, gosh, if this is now plentiful, who am I? What do I contribute?

首先，我认为社会过去也经历过与此同等重大的转变，但这次依然非常重要。我不认为它只是软件——这是我的个人观点。虽然和你一样，Ben，我也会摇摆不定，有时站在这边，有时站在那边。我觉得它之所以重大，是因为我很清楚：我个人、以及可能本播客里的每个人，都在某种程度上把自我认同与自身的智能绑定。这也是人们收听你们播客的重要原因之一；它关系到你如何进入大学、找到工作、获得各种东西。于是你会问：天啊，如果这种能力现在成了充裕之物，那我是谁？我能贡献什么？

I brought up the personal thing. It's really interesting. There's this meta thing in the Silicon Valley right now, which is if you tell what jobs are most likely to be automated with the current generation of technology, you would probably put software engineering right at the top. The people building this technology are building the technology that is disrupting their own profession. That's I'm not sure unprecedented, but certainly unusual with technology disruption.

我提到了个人层面的事，这真的很有意思。硅谷现在有个“元”现象：如果你要列出在当前这代技术下最可能被自动化的工作，软件工程大概会排在最前面。构建这项技术的人，正在构建会颠覆他们自身职业的技术。这一点未必史无前例，但在技术颠覆史上确实罕见。

I think this idea of identity, intelligence, and the technology impacting our own perception of self-worth is happening in a very personal way for a lot of the people working on it. I am very confident that on the other side of this, just like we've gone through with the industrial revolution and the agriculture revolution, all these other things, we'll come out the other side and just end up higher leverage species. We'll spend our time on different things than we did before, but I believe that this will make us just happier, more productive, have more plentiful. We'll just have access to more things. I just think about really simple things like access to mental healthcare, access to education, access to medical advice, access to legal advice.

我认为，这种关于身份、智能与技术如何影响我们自我价值感的议题，正在以非常个人化的方式发生在许多从业者身上。我很有信心，等我们走过这一阶段，就像我们经历工业革命、农业革命等一系列变革那样，我们会以一个“更具杠杆”的物种出现。我们将把时间花在不同于以往的事情上，但我相信这会让我们更快乐、更高效、拥有更丰富的选择；我们将能获得更多东西。哪怕是很简单的事：更易获得的心理健康服务、教育、医疗建议与法律建议。

We are essentially taking expertise and making it a commodity, and I think that will as generally democratizing. I think many of the things I mentioned, if you have wealth, you have a lot of access to, and if you don't, you don't. What a cool thing that we've made this like universally accessible. Just like you had the Luddites and the Industrial Revolution, you're going to have this period of transition where it's saying, how I've come to identify my own worth, either as a person or as an employee, has been disrupted? That's very uncomfortable.

本质上我们在把“专业知识”商品化，而我认为这总体上会带来更广泛的民主化普及。我前面提到的许多东西，过去是“有财富的人能轻松获得、没财富的人就很难获得”。而如今把它做成普遍可及，这太酷了。就像工业革命时期的 Luddites 一样，我们也会经历一个过渡期：人们会问，“我作为一个人或作为一名员工所形成的自我价值认同被打破了？”——这非常让人不安。

That transition isn't always easy. If you look at globalization, it lowered the cost of a television set, but it was hard if the factory in your town was shut down. If you look at things like GDP numbers or productivity numbers, it obscures the individual impact that is not always fun, easy, or good. Similarly, it could be great in 10 years but really hard over the next two years. I think probably all of those things are true.

这种转变并不总是容易的。以全球化为例，它降低了电视机的成本，但如果你所在小镇的工厂被关停，那就会很艰难。GDP 或生产率等指标会掩盖个体层面的影响，而个体体验并不总是愉快、轻松或正面的。同样，10 年后可能非常好，但接下来的两年可能会非常难。我觉得这些说法都成立。

David: Silicon Valley's right in the middle of it.

David：硅谷正处在这场变革的中心。

Bret: Yeah, you're just in the middle of it. My view is, it is as transformational as the former of what you said, Ben. I think it is truly transformational. I also think the transition will be awkward and probably slower than people think in the last comment because it was too long an answer, but I don't think all parts of the economy can absorb intelligence equally. Let's just say we developed fairly generalized super intelligence. I always use the analogy like you can invent a lot of drugs, but if clinical trials still take a long time, you're not necessarily going to get new therapies rapidly. You have regulation, you have cultural resistance, you have all these other things.

Bret：是的，你就是在中心位置。我的观点是：它像你前面那种说法一样具有“变革性”，Ben。我认为它确实是变革性的。同时，我也觉得过渡会有些别扭，而且大概率比人们想象得更慢——我上一个回答太长了，但核心是：经济的各个领域并不能同等地吸收“智能”。就算我们发展出了相当通用的“超级智能”，我常用的类比是：你可以发明很多药物，但如果临床试验仍然需要很长时间，你也未必能迅速得到新疗法。你会受到监管、文化阻力等诸多因素的影响。

The technology is transformational. I think probably it will impact society on a more measured pace than I think a lot of the folks in the AGI community think just because of the natural rate limiters of society around. It's not like intelligence is the only input to productivity growth, but I really do think it's transformational. I think it's both in a great way and an uncomfortable way. I think it's transformational.

这项技术是变革性的。但我认为，由于社会层面的“天然限速因素”，它对社会的影响节奏可能会比许多 AGI 社区的人想象的更为克制。生产率的提升并不只有“智能”一个投入要素，但我真的认为它是变革性的——既有伟大的一面，也有令人不适的一面。我认为它就是变革性的。

David: Specifically for you guys, I'm so curious, what is the state of human capital at Sierra. You're one of the leading AI companies, you could probably recruit anybody you want given who you guys are, but at the same time you were just saying, Bret, AI can probably do a lot of what people at Silicon Valley companies used to do. What are you guys doing? You're living it day to day.

David：具体到你们，我很好奇：Sierra 的人力资本状况如何？你们是领先的 AI 公司之一，以你们的资历大概可以招到任何你们想招的人；但与此同时，正如你刚才所说的，Bret，AI 可能已经能做很多硅谷公司员工过去做的事。那你们到底怎么做？你们每天都在亲身经历这一切。

Clay: First of all, everything we do as a company is going to be the direct or at least indirect result of talented, smart, amazing people who are motivated to accomplish an important mission doing great work. Since starting the company in March of 23, we've grown immensely. We're the leader in the space. The demand for what we've built has been overwhelming, and we are just growing the company in terms of people, geographies, and industries as quickly as we possibly can.

Clay：首先，我们这家公司所做的一切，都是那些有才华、聪明、出色、并且怀着完成重要使命而被激励的人直接或至少间接努力的结果。自从 2023 年 3 月创办以来，我们成长非常迅猛。我们已是这个领域的领导者。市场对我们所打造产品的需求非常强劲，我们正在以尽可能快的速度，在人员、地域和行业维度上扩张公司。

On top of that though, we look for every point of leverage we can find with AI. We aggressively use cursor in our engineering teams. I don't know what the percentage today of lines of code written by cursor is, but it's pretty darn high in getting higher by the day. To Bret's previous point, it is today this immense force multiplier for talented people who one of our most senior engineers, actually the first person who joined the company after Bret and me, basically sent coding agent over the weekend to just come up with a dozen PRs for things that he wanted to fix. It came back and half of them were great, a quarter of them were total garbage, and a quarter needed some work. It was, in a way, him working the entire weekend with guidance from him.

除此之外，我们会在一切可能的地方用 AI 去撬动杠杆。我们的工程团队在积极使用 cursor。我不知道现在由 cursor 编写的代码行数占比是多少，但比例相当高，而且每天都在提高。呼应 Bret 刚才的观点，如今它对有才华的人来说是巨大的“倍增器”。我们最资深的工程师之一——事实上是 Bret 和我之后加入公司的第一位同事——周末基本上把一个 coding agent 放出去，让它针对他想修复的问题提交了十来个 PR（pull requests）。结果回来一看，其中一半非常好，四分之一完全不行，四分之一需要打磨。从某种意义上说，这就像他在周末整整干了两天，而 agent 在他的指导下配合完成。

Bret: Let me double click on the cursor thing for a second though because Clay and I talk a lot about this. We always like to say, the way we think about an AI first company is we're building a machine to produce happy customers. That's why we think about it. I think that's important because if something comes off the assembly line of machine that's malformed, you don't just fix that thing. You say, what part of the machine broke to produce the malformed item?

Bret：让我在 cursor 这件事上“展开说一下”，因为我和 Clay 经常讨论这个主题。我们常说，作为一家 AI first 公司，我们的思维方式是：在打造一台“生产 happy customers 的机器”。我觉得这点很关键，因为如果这台机器的装配线上出现了不合格品，你不能只修这个不合格品；你应该问：机器的哪个部分出了问题，才会产出这个不合格品？

Just as it relates to, for example, software engineering, we have this philosophy. When Cursor, which is the most popular co-pilot for software engineers to write code and now having some more agentic flavors of it, if it produces incorrect code, our philosophy is don't fix the code. Fix the context that Cursor had that produced the bad code. I think that's a big difference when you're trying to make a company driven by AI and just use AI because essentially if you just fix the code, you're not adding leverage. If you go back and say, what context did this coding AI not have that had it had it, it would've produced the correct code? I don't want to pretend we're perfect here, but that's the way we think about it.

就拿软件工程来说，我们有这样一条理念：当 Cursor——这款如今最流行、用于辅助软件工程师写代码的 co-pilot（而且现在带有一些更“agentic”的风味）——产出了错误代码，我们的原则不是去修代码，而是修复 Cursor 所处的“上下文”，正是那些上下文导致了错误代码的生成。我认为，这就是“让公司由 AI 驱动”和“只是用用 AI”之间的重大差别：如果你只修代码，本质上并没有增加杠杆；如果你回头问，这个 coding AI 缺少了哪些上下文——如果它拥有这些上下文，就会生成正确的代码。我们不敢说自己完美，但我们就是按这个思路来做。

I really like thinking of our business as a machine. This is a Clayism. He said this once right when we're starting the company. He said, we're building a machine to produce happy the customers and nerds snipe me. I'm like, oh, we are a hundred percent building a machine. Employees roll their eyes, but it's like fix the machine. Fix the machine, don't just fix the output of the machine.

我非常喜欢把我们的业务看成一台机器。这是一个“Clayism”（Clay 的名言）：在公司刚起步时他就说过，“我们是在打造一台生产 happy customers 的机器”，这话一下就把我 nerd-snipe 了（击中了我的极客点）。当时我就想：没错，我们百分之百是在造机器。员工可能会翻白眼，但核心是：修机器，修机器，而不是只修机器的产出。

I think with AI, it actually creates a very actionable framework for how to bring AI into the company. Somewhat ironically, as Clay said, we are hiring a lot of people. Notably to me, all the AGI labs hired a lot of people. The elephant in the room is it's not quite done.

我认为，AI 实际上为“如何把 AI 引入公司”提供了一个非常可操作的框架。有点讽刺的是，正如 Clay 说的，我们确实在大量招人。对我来说值得注意的是，几乎所有 AGI labs 也都在大量招聘。显而易见的问题是：这件事（通向目标）还远未完成。

David: You guys are the center of this. Bret, you're the board chair at OpenAI. What do you guys do at Sierra? You're incredibly well funded, et cetera, but you can't compete with a \$300 million comp package? What is going on with hiring?

David：你们正处在这场变革的中心。Bret，你是 OpenAI 的董事会主席。那在 Sierra，你们怎么做？你们资金非常充裕，等等，但难道不能和 3 亿美元的薪酬包去竞争吗？招聘到底怎么推进的？

Ben: I assume you don't need those people.

Ben：我的理解是，你们并不需要那些人。

Bret: Yeah. We're an applied AI company. Let me just give you my view of the market. This is something you all could debate. I think there's basically three big categories of AI software companies, first are the foundation and frontier model companies. These are the folks that are somewhat infamously or famously competing for these scarce resource of these great researchers. Many of them, like OpenAI, are mission-driven and trying to create artificial general intelligence. Some of them are more commercially-minded and essentially building these models, which they then license or lease out to companies like ours.

Bret：是的。我们是一家应用型 AI 公司。让我说说我对市场的看法，这点你们完全可以辩论。我认为 AI 软件公司大体分三类：第一类是基础模型与前沿模型公司。这些公司（或臭名昭著、或名声在外）在为稀缺的顶尖研究人才激烈竞争。它们中许多（比如 OpenAI）是使命驱动，致力于构建通用人工智能；也有一些更偏商业化，本质上是在训练模型，然后把这些模型授权或租赁给像我们这样的公司。

There's a category of people who make tools on top of it, so the proverbial pickax and the gold rush. You need data labeling services, and you need a data warehouse on which you need a retrieval, augmented generation, like a vector database to support retrieval augmented generation, and all these things. I'll say all these tools that one uses when you're building an AI platform.

还有一类是在其之上做工具的——所谓“淘金热里的鹤嘴镐”。你需要数据标注服务，需要数据仓库，需要支持检索增强生成（retrieval augmented generation）的组件，比如向量数据库，等等。也就是说，搭建 AI 平台时所要用到的这一整套工具。

There are companies like Sierra, where we make AI agents for customer service, customer experience. Harvey, who makes AI agents for the legal profession. I think companies like Writer do it for marketing. All these are vertical AI applications. We're downstream of a lot of that. If we're doing our job, we're taking the best of these models and composing them to make these amazing experiences.

还有像 Sierra 这样的公司——我们为客户服务、客户体验打造 AI agents；Harvey 为法律行业打造 AI agents；我认为像 Writer 这类公司则用于市场营销。以上都是垂直领域的 AI 应用。我们处在很多能力的下游。如果我们把工作做好，就是把最好的模型拿来进行编排组合，创造出卓越的体验。

Just like if you were VCs and a software as a service company came up to you and said, step one is we're going to build our own data centers. You'd look at them very skeptically. He'd be like, really? Why not just rent a server from Amazon Web Services or Azure? I think the same is true of applied AI companies. It may not have been true a year and a half ago, two years ago, where I think everyone wanted to pretend they were cool.

就好比你们是 VC，有家 SaaS 公司过来说：第一步我们要自建数据中心。你们多半会投来质疑的目光：“真的？为什么不直接用 Amazon Web Services 或 Azure 的服务器？”我觉得对应用型 AI 公司也是一样。也许在一两年前还不是这样，当时大家都想显得“很酷”。

David: We had Clem from Hugging Face on, and I deeply believe that every applied AI company needs their own foundational model and needs to build it themselves.

David：我们之前请过 Hugging Face 的 Clem 来做客，我坚信每一家应用型 AI 公司都需要拥有并自行训练自己的基础模型。

Bret: I could not disagree more strongly with this single sentence.

Bret：对于这句话，我强烈反对，不能再不同意了。

Clay: That was the trend in 2023. You weren't cool if you didn't have your own foundation model. Adapt, character, and inflection, those all ended differently.

Clay：那是 2023 年的风潮——没有自有基础模型就不“酷”。Adapt、character、inflection，这些后来结局都各不相同。

Bret: It turns out that unless you're a pharmaceutical company, pharmaceutical companies get protection from patents for an asset that has value for a long period of time. I've heard from multiple investors that foundation models are the fastest deteriorating asset of all time. If step one of your business is to burn through tens of millions or hundreds of millions of dollars of capital before you find product market fit, and that asset has value for a week, I'm not sure it's a great business model.

Bret：事实证明，除非你是制药公司——制药公司能靠专利保护其长期有价值的资产——否则基础模型可能是史上“贬值速度最快”的资产（多位投资人都这么说）。如果你的商业计划第一步就是在找到产品市场契合之前先烧掉数千万、上亿美元，而这项资产的有效期还可能只有一周，我不确定这是个好商业模式。

Clay: It's a very, very expensive carton of milk.

Clay：这就像一盒非常非常昂贵的牛奶。

Ben: Let's use the hypothetical example of training a frontier class model today. What do you think the usable life of that is? You got to amortize a lot of token generation in a pretty short period of time to make that worth it.

Ben：那我们设想一下：今天去训练一个前沿级模型，你觉得它的可用寿命有多长？要让这笔投入划算，你得在相当短的时间里摊销大量 token 生成的成本。

Bret: It's complicated. There are different ways of looking at it. I'll start with there isn't one. It depends on what type of model you're building. Deepseek, somewhat famously in their paper, talked about reducing the costs of building these models, but I think there's a difference between foundation models and frontier models. Frontier are really the best of the best. This is what labs like OpenAI aspire to always have.

Bret：这很复杂，评估维度也不止一种。先说结论：没有放之四海而皆准的答案，这取决于你在训练哪种模型。Deepseek 在论文中就提到如何降低训练成本。但我认为基础模型与前沿模型之间是有差别的。前沿模型是真正“最强中的最强”，这正是像 OpenAI 这样的实验室力图始终保持的优势。

When you have the best model, you attract different customer base. It's always been Apple's strategy with their devices and things like that. There's a difference. If you think of these things as commodity, you'll take a different strategy, you say our goal is to have the most intelligent model, there are downstream benefits of that beyond the cost of those individual models.

当你拥有最好的模型，你吸引到的客户基础就不一样。这一直是 Apple 在其设备等产品上的策略。两者路径不同：如果把模型当作“商品”，策略就会不同；但如果你的目标是“拥有最智能的模型”，那么除了模型本身的成本之外，还会获得下游的一系列溢出收益。

I actually think there's not a one size fits model for most tasks now. There's a really interesting trend in model building called distillation, where you can take a very high parameter count model and essentially make a smaller parameter count model that's call it 80% as good, and I'm just making up that number. It really varies.

我其实认为，对多数任务而言不存在“一刀切”的通用模型。模型构建里有个很有趣的趋势叫“蒸馏”（distillation）：你可以把一个超大参数量的模型“蒸馏”成一个更小的模型，其效果大约可达到原模型的 80%（这个数字只是打个比方，实际差异很大）。

Ben: Deepseek did this right, based on open source?

Ben：Deepseek 是基于开源把这件事做对了，对吗？

Bret: There are two interesting trends, and probably the researchers will wince at my simplification. You have distillation on one hand, which can take a very expensive model and make something that's almost as good but much cheaper for inference. Then you have this post-training process, which is reinforcement learning on chains of thought, which is the basis of things like 2003 and 2004 and these really advanced models. The two together, you just have so many variables of cost, performance, quality, reasoning, and all these other things.

Bret：有两个有趣的趋势——我这样简化说法，研究人员大概会皱眉。一方面是蒸馏（distillation），它可以把一个成本高昂的模型“变成”一个在推理阶段便宜得多、而效果几乎同样好的模型。另一方面是后训练（post-training）流程，也就是基于思维链（chains of thought）的强化学习（reinforcement learning），这是像 2003 和 2004 这类非常先进模型的基础。把这两者结合起来，你就会面对很多变量：成本、性能、质量、推理能力等方方面面。

Actually, the reason I think it's exciting is if you look at the database space right now, there's not one database. If you want to do large scale data analytics, you'll choose one thing. If you want to do a transactional data store, you'll do another. I think we're moving to that area of models. When Clay and I talk about how to produce a delightful low latency phone conversation, you care a lot about latency. That's one metric of quality that isn't intelligence, but it's important. Put another way, if you had to think for 30 seconds before responding on a phone call, that might not be viable.

实际上，我之所以觉得这很令人兴奋，是因为看看当下的数据库领域：不存在“一种通吃”的数据库。做大规模数据分析你会选一种；做事务型数据存储你会选另一种。我认为模型正走向类似的分化。比如我和 Clay 讨论如何实现愉悦的低时延电话对话时，你会非常在意延迟。那是一个不等同于“智能”的质量指标，但很重要。换句话说，如果打电话时每次回复都得先想 30 秒，那大概就不可行了。

Similarly, let's say you're a company who has a relatively inexpensive offshore contact center, and you need your cost of your AI to be lower than that. Cost matters too. If it costs \$2000 per phone call, it might not actually be viable as a business. I say all that because I'm not sure there's one answer to your question, Ben, because it really depends on who you're selling it to and what their goals are. I think that's good because sometimes you want something for drug discovery and you want the most intelligent model. Sometimes you want something for a low latency transactional phone call and you care about latency and cost more. It's creating a market for these things, where I think if you're one of these big foundation model companies, you actually can have a portfolio of products.

同理，假设你是一家公司，拥有相对低成本的海外联络中心，那么你需要让 AI 的成本低于它。成本同样重要。如果每通电话要花 2,000 美元，这门生意可能就不可行。我之所以这样说，是因为我不确定你的问题有唯一答案，Ben——这确实取决于你卖给谁、对方的目标是什么。我觉得这反而是好事：有时你做药物发现，需要最智能的模型；有时你做低时延的交易型电话客服，更在意时延和成本。由此就会形成一个细分市场；在这个市场里，如果你是一家大型基础模型公司，其实可以打造一个产品组合（portfolio）。

To your point, broadly, first principles, I think the reason why Clay mentioned the consolidation in this space, it has to be consolidated because to basically make the money back on the pre-training and post-training process, you need relatively few players that are collecting taxes from all the players on top, or you just can't make the the math work.

回到你的观点，从更宏观、第一性原理看，我认为 Clay 之所以提到这个领域会整合，是因为它“必须”整合：要把前训练与后训练的投入赚回来，产业上游需要少数参与者向上层的众多玩家“收税”；否则账根本算不平。

David: Yeah. For startups, going in on your own is just completely non-viable on every dimension because there's the capex, but there's also the opex. You could spend the capex. Let's say you magically got \$50 billion as a startup to build a frontier model, you can't just let it sit there. Operating that model and having it continue to run and continue to improve on it, that's what people are getting paid \$300 million a year to do.

David：是啊。对初创公司来说，单打独斗在各个维度都不可行——不仅有资本开支（CapEx），还有运营开支（OpEx）。就算你砸下 CapEx——比如一家初创公司“神奇地”拿到 500 亿美元去训练一个前沿模型——你也不能让它闲置。把这个模型运营起来、让它持续运行并不断改进，这正是有人一年拿 3 亿美元薪酬去做的事。

Clay: The primary constraint, I don't think, is capital. It starts with the people. There's a small set of people who know how to architect these models, do the pre-training runs, do post-training, RL runs. That would be the starting place.

Clay：我认为最主要的约束不是资本，而是人才。首先要有人才——懂得如何架构这些模型、进行前训练、进行后训练和 RL 训练的人，目前只是一个很小的群体。这才是起点。

To your point, the capital outlay for building a data center that can train these multi-trillion parameter count models is just enormous. You have to amortize the cost of the people, amortize the cost of the capital to build out the data centers and then do that to your point in a pretty short period of time in order to make the math work. I do think there will be a very small number of these frontier models and research labs producing them. They will optimize all the way down to the memory, the chips, power delivery, and build this highly vertically integrated stack to get as much value out of the model as quickly as possible and at lowest cost possible.

另外，就你的观点而言，能训练多万亿参数模型的数据中心，其资本投入极其庞大。你必须在相当短的时间里摊销人员成本、摊销建设数据中心的资本成本，才能让账算得过来。我确实认为，最终只会有极少数前沿模型与研究实验室负责生产它们。他们会把优化做到底——深入到内存、芯片、供电层面——并构建高度垂直整合的技术栈，以尽可能快、尽可能低成本地从模型中获取最大价值。

Bret: By the way though, just with all the press around the talent, it's still a rounding error compared to the infrastructure, so I think it's worth keeping that in mind. It's just very expensive period just because it's a salacious story. The capex is still the dominant cost.

Bret：不过顺带说一句，尽管媒体总在谈“人才”，但与基础设施相比，人才成本仍只是“可忽略不计的尾数”。之所以老被提起，是因为更“吸睛”；但总体上这件事就是非常烧钱，而 CapEx 仍是主导性成本。

Clay: Just earlier today in Google earnings, they say, oh, we're going to spend an extra \$10 billion on infrastructure build out this year.

Clay：就在今天的 Google 财报里，他们说今年要再多花 100 亿美元用来建设基础设施。

Ben: Going from 75 to 85, something like what?

Ben：从 75 到 85，具体指什么？

Clay: 75 to 85. It was like, oh, by the way, we're spending another 10. That's the scale at which these buildouts are occurring.

Clay：从 75 到 85。就像“顺便说一下，我们还要再多花 10”。这就是这些基础设施扩建的量级。

Ben: Which is so interesting because these tech businesses that everyone loved and applied these really high multiples too for so long were these asset light, low capex requirement businesses, where your expensive thing was your human capital to produce software. Once you've paid for your human capital, you just have this 85% gross margin amazing business model of software. That's not really true anymore. These big tech companies have massive ongoing capex.

Ben：这很有意思，因为大家长期喜爱的、给出很高估值倍数的那些科技业务，过去都是“资产轻、CapEx（资本开支）要求低”的模式，最贵的投入是产出软件的人力资本。一旦付完人力成本，就能拥有 85% 毛利率的神奇软件商业模式。现在这已经不完全成立了，这些大科技公司都有大规模、持续性的 CapEx。

Bret: It is. It probably cuts both ways too. I think it represents significant barriers to entry too. Just like at Amazon Web Services, which is one of the more impressive businesses built over the past 30 years and, in contrast, if you look at the trends of how often do you see a new social service, pretty often, right? Over that 30-year period, it doesn't mean that products like Facebook have gone away, but the barriers to entry are much lower as well.

Bret：确实如此，而且这可能是把双刃剑。我也认为这意味着显著的进入壁垒。就像 Amazon Web Services，这是过去 30 年里最令人印象深刻的业务之一；相对地，如果你看新社交服务出现的频率——相当频繁，对吧？在这 30 年里，这并不意味着像 Facebook 这样的产品消失了，但（社交）领域的进入壁垒也低得多。

You're right, Ben, that the way you model these businesses changes is a significant barrier to entry as well. I'm not sure how to think about it strategically because you can look at a DCF analysis in one way. I think it's important to zoom out. The half-life of technology companies is not that long. There are a few, but they're the exception.

你说得对，Ben，这类业务的“估值建模方式”本身的变化也构成了进入壁垒。从战略上怎么想这件事我也拿不准，因为从 DCF 分析的角度可以有不同看法。我觉得拉远看很重要：科技公司的“半衰期”并不长，能长期存活的有一些，但只是少数例外。

I've brought this story up before, but this is a hundred percent true. I started at Google, we're in the small building in Mountain View, and we moved into a campus. It was the Silicon Graphics campus. Amusingly, by the way, they were still using a couple of buildings on the end. You had this company that was literally dying and selling their campus for parts.

我以前讲过这个故事，但它百分之百是真事。我在 Google 起步时，我们先在 Mountain View 的一栋小楼里，后来搬进一个园区——就是 Silicon Graphics 的园区。有趣的是，他们当时还在用园区尽头的几栋楼。你能看到一家几乎“行将就木”的公司在“拆卖园区”。

David: I didn't realize SGI was still there. That must have been so sad.

David：我都不知道 SGI 还在那里。那一定很让人难过。

Bret: We go into the cafeteria, we had free food, and they were paying for their food in same cafeteria. It was just super awkward.

Bret：我们去同一个食堂吃饭，我们是免费，他们还得自己付钱——场面非常尴尬。

David: Brutal.

David：太残酷了。

Bret: Then though, at Facebook, we were at this first office downtown, then the old HP building next to Stanford, and then we moved into Sun Microsystems campus, which had also died. Both of those companies in my lifetime were at the top of the stop market, then had sold their campus for parts, and then we were taking it over. I say all that because it's useful to look at 80% margins, blah, blah, blah, but how many technology companies have lasted more than 40 years? Obviously these technologies are new.

Bret：后来在 Facebook，我们先是在市中心的第一个办公室，接着搬到 Stanford 旁边的旧 HP 大楼，再后来又搬进已经“陨落”的 Sun Microsystems 园区。这两家公司都在我有生之年登上过“股市巅峰”，随后又把园区“拆卖”，然后被我们接手。我之所以说这些，是因为看 80% 的利润率、等等等等当然有用，但有多少科技公司能活过 40 年？显然，这些技术都还新。

I think it's just complicated as you look at these things. I would make the same decisions as a lot of the hyperscalers in terms of capex. I think first, the promise of the value of artificial general intelligence is so great. The expected value equation is absolutely worth it in my opinion. Similarly, I do think the scale afforded by these investments represents a lot of strategic values. I totally get why it's complicated for investors, but I think the landscape changes, and I think there's probably a bigger risk of not existing in 30 years than selling a spreadsheet.

我觉得把这些因素综合起来就会很复杂。就 CapEx 而言，我会做出和很多 hyperscalers（超大规模云厂商）相同的决定。首先，人工通用智能的价值前景太大了——用期望值的算式看，在我看来绝对值得。同样，这些投入带来的规模效应也有很强的战略价值。我完全理解这对投资者来说很复杂，但我认为格局在变化，30 年后“消失不见”的风险，可能大于“把电子表格里的数字做漂亮”的风险。

David: If you look at their market caps, investors are giving them a pass for now. Speaking of strategy, you guys have talked about this plenty elsewhere, but I really want to double click with you. Your business model and pricing strategy at Sierra is radical. Bret, you were most recently the co-CEO of Salesforce. I could not imagine a more deeply invented software as a service category. You guys making the decision to throw out that business model and do something else for enterprise software is radical.

David：从它们的市值看，投资者目前在“网开一面”。说到战略，你们在别处谈过很多，但我想和你们再“深入聊聊”。Sierra 的商业模式和定价策略非常激进。Bret，你刚刚还是 Salesforce 的联席 CEO——我很难想象有比那更“根深蒂固”的 SaaS 业务形态。你们决定抛弃那套模式，去做一套面向企业软件的全新模式，这非常激进。

Ben: It's very telling at the very least. Lay it on us. What's the strategy and why'd you do it?

Ben：至少说明了很多问题。跟我们摊开讲吧。你们的策略是什么？为什么这么做？

Clay: We started from first principles and asked what are agents actually doing? In contrast to software as a service or software, you'd buy off a shelf at Fry's Electronics decades ago, which might help you be marginally more productive, help you get a job done. Agents in contrast are actually getting the job done for you. You're in essence hiring software to accomplish a task and get it done well.

Clay：我们从第一性原理出发，先问“agents 实际上在做什么？”这和过去的软件即服务（software as a service）或几十年前你在 Fry's Electronics 货架上买到的软件不同，那些软件顶多让你稍微更高效、帮你完成工作；而 agents 则是替你把工作真正做完。本质上，你是在“雇佣软件”来完成一项任务，而且把它做好。

As we were thinking about how do you price this, what is the business model, seat based? What is a seat? That doesn't make any sense. Consumption? Is it per message? Is it per token? Is it per conversation? None of these things actually mapped very well to getting a job done and getting a job done well. Going to principles of value-based pricing and pricing against value delivered. We arrived at what we call outcome-based pricing or resolution based pricing, where we only charge our customers when their agent successfully completes the task that it's set out to do.

在思考如何定价、商业模式是什么的时候——按席位计费（seat-based）？什么叫一个 seat？这说不通。按用量计费？按消息？按 token？按会话？这些方式都无法很好映射到“把一项工作完成、并且完成得好”。于是我们回到“基于价值定价”的原则，围绕“交付的价值”来定价。我们最终采用所谓的“基于结果定价 / 基于解决定价”（outcome-based / resolution-based），只有当客户的 agent 成功完成其设定任务时，我们才收费。

David: That's defined as human does not get involved?

David：这是否定义为：没有人工介入？

Clay: That's correct. It completely gets the job done. The reason this is important is, what we're trying to do in a way is resolve this age old tension between the cost and quality of customer experience. I think every great business wants to deliver an amazing experience to their customers. But unless you're like Hermes or the Four Seasons, it's too expensive to do. A phone call might cost \$10, \$15, or \$20. Rolling a service truck might cost hundreds of dollars.

Clay：没错。就是彻底把事情办完。之所以重要，是因为我们要解决一个长期存在的矛盾：客户体验的“成本与质量”的张力。我认为每一家优秀企业都想提供出色体验，但除非你像 Hermes 或 Four Seasons，一般来说那样做太贵了——一通电话可能要 10、15、甚至 20 美元；派一辆服务车上门可能要几百美元。

How do you bridge that gap? We think that AI changes that. AI, it turns out, can do things that people value very much. It can reason and decide. It can take action, it can speak your language, it never gets tired, it's always patient. Most importantly, it can get this job done.

如何弥合这个差距？我们认为 AI 改变了这一切。事实证明，AI 能做很多人们非常看重的事：它能推理和决策，能采取行动，能用你的语言沟通，从不疲惫，始终耐心——最重要的是，它能把事情办成。

What we love about the model is it deeply aligns with our incentives with our customers. They want to lean on their Sierra agent as much as possible because when they do, we deliver a better customer experience and save the money. We're motivated to build the most performant, capable agents that we can. It ends up a very different relationship, where, as opposed to vendor customer, we are partners trying to build this incredible agent to elevate the customer experience and also save cost.

我们喜欢这套模式，因为它深度对齐了我们与客户的激励。客户会尽可能依赖他们的 Sierra agent——这样他们既能得到更好的体验，也能省钱；而我们也被激励去打造性能更强、能力更全面的 agents。双方不再是传统的“厂商—客户”关系，而是共同打造卓越 agent、提升体验并节约成本的“合作伙伴”关系。

Ben: You guys are effectively carrying the risk. You're incurring all these costs of doing all this AI, of building all this software, of hiring all these people, but you're not getting paid unless the ultimate bottom of funnel thing happens.

Ben：等于说你们在承担风险。你们要负担做 AI、构建软件、招聘人才的所有成本，但只有当漏斗底部的最终转化发生，你们才会收到钱。

Clay: That's right. It's an expression of confidence that our platform will deliver, our agents will be the best, and that they can deliver. Again, I think it's a strong signal of the quality that we can deliver our confidence in the technology. Again, the incentives alignment is extremely powerful.

Clay：是的。这表达了我们对平台与 agents 的信心——它们能交付、而且能做到最好。我也认为这是我们对技术与可交付质量的强烈信号。同时，这种激励对齐极具力量。

Bret: If you talk to a comp expert, if you look at an enterprise software sales team, usually their compensation is 50% salary, 50% performance based, based on quota attainment. With people, we've always thought about, how do you incentivize the behaviors you want? It's a huge topic. Executive compensation, for different roles, we really just want to move software in that direction. If the AI agent's supposed to make a sale, it should be paid a commission. If the AI agent is supposed to handle customer service, it should be paid when it solves the problem. If it doesn't, it didn't do its job, if it didn't do anything valuable for you, you shouldn't pay for it.

Bret：如果你和薪酬（compensation）专家交流，会发现企业软件销售团队通常是“50% 基薪 + 50% 绩效”，且绩效与配额完成（quota attainment）挂钩。对人而言，我们总在思考如何激励想要的行为——这是个大话题，包括高管薪酬、不同岗位的激励设计。我们只是想把软件也往那个方向推进：如果 AI agent 负责销售，它就该拿“佣金”；如果它负责客服，只有在解决问题时才应该“获得报酬”。若没解决、对你没有创造价值，你就不该为它付费。

To your point, Ben, we're taking on the risk, but I also think as a consequence we're making something more valuable. It's very easy for our customers to know the value of their Sierra agent. They know how much it would've cost to have a phone call with a person. They know how much it costs to have the Sierra agent solve it. We're saving our customers hundreds of millions of dollars in operating expenses and improving their customer experience.

回应你的观点，Ben，我们确实在承担风险，但因此我们也在创造更有价值的东西。客户很容易就能评估他们的 Sierra agent 的价值：人工接电话要花多少钱、让 Sierra agent 解决要花多少钱——一目了然。我们在为客户节省上亿美元的运营开支，同时提升其客户体验。

To your point, David, when you brought up the software as a service revolution in the early 2000s, I'll pause so you can take the same direction you want, but I think this will upend the business model of enterprise software in a really positive way.

回应你的观点，David——你提到 2000 年代初的 software as a service 革命——我先停一下，方便你按你的思路展开。但我认为，这（模式）将以一种非常积极的方式颠覆企业软件的商业模式。

David: I still want to ask about this.

David：我还是想继续追问这个问题。

Clay: There's the analogy with online ads where we move from CPM to CPC to pay per conversion. As you move closer and closer to the actual value, you can deliver a much more valuable service.

Clay：这可以类比在线广告，我们从 CPM 转向 CPC，再到按转化付费。你越贴近实际价值，就越能提供更有价值的服务。

Ben: You have to. It forces your company to deliver value, otherwise you go outta business.

Ben：你必须这样做。它会迫使你的公司交付价值，否则就会倒闭。

David: Have you guys started to see or discover little glimpses of it? Really to underscore radical, Bret, you said, if this model takes, it's going to change everything. This isn't just a pricing model. When Mark Benioff invented SaaS, it changed everything about Silicon Valley, not just how software is delivered. Have you guys started to see or get inklings of what the second and third order effects of the business model is going to be?

David：你们已经开始看到或发现一些苗头了吗？为了强调它的“激进性”，Bret，你说过如果这种模式站稳脚跟，它会改变一切。这不仅仅是定价模型。当 Mark Benioff 发明 SaaS 时，它改变的是整个硅谷，而不只是软件的交付方式。你们是否已经开始看到或预感到这种商业模式可能带来的二阶、三阶效应？

Bret: Clay mentioned that we're more partner than vendor. I'm going to give my historical context, which might be slightly embellished, but you can poke at it. If you look at perpetual license software or you bought a version of a piece of software, what you do, as a company, would buy it, and then typically you'd have a group of people at the company who installed it, maintained it, managed the upgrade process, ran the servers, and did a lot of stuff.

Bret：Clay 提到我们更像“合作伙伴”而不是“供应商”。我给点历史脉络，可能略带修辞，但你们可以挑刺儿。过去如果是永久授权（perpetual license）的软件，或者你买了一版软件，公司会先买许可证，然后通常会有一拨人在公司里负责安装、维护、管理升级流程、跑服务器，以及做一大堆相关的事。

When you had the equivalent of that, that was software as a service, it wasn't just that you switched from capex to opex and you had a perpetual license and you went to a subscription software, which changed accounting and all that, it also changed the people you needed. You didn't really need the site reliability engineers to keep the service up because that was actually something that you got from your software as a service vendor. You didn't actually have to worry about the servers either. You could actually probably get rid of a data center.

当等价物变成了 software as a service（SaaS），变化不仅是从 CapEx 转成 OpEx、从永久授权改为订阅（进而改变了会计处理等），还改变了你所需的人员结构。你并不再需要站点可靠性工程师（Site Reliability Engineers，SRE）来维持服务可用性，因为这事儿由 SaaS 供应商来做。你也不必再操心服务器了，甚至可能可以直接不要自己的数据中心。

Actually, if you switch entirely to software as a service, you probably don't even need a data center team at all. You end up where not only do you change the way you pay for the software, but you actually change the roles and responsibilities for it. With software as a service, you obviated the need for a lot of the lower level technical machinery of running this software, but you still had to install it, customize it, and all these other things.

事实上，如果你完全切到 SaaS，可能连数据中心团队都不再需要。最后的结果是，不仅支付方式改变了，角色与职责分工也随之改变。用上 SaaS，你免除了大量运行软件所需的底层技术性工作，但依然需要做安装、配置/定制等一系列事情。

I always use the analogy, let's just take a CRM software or something. You install it and your sales don't improve. Whose fault is it? I don't think many people would blame the CRM software. It's like, maybe you have bad salespeople. I don't know.

我常打个比方，比如拿 CRM 软件来说。你把它装上了，但销售没提升。该怪谁？大多数人不会去怪 CRM 软件。可能是你的销售队伍不给力，我也说不准。

David: Are you talking from experience?

David：你这是亲身经历吗？

Bret: No. It's a hundred percent. We use Salesforce this year. It's a great CRM, I love it. I am very loyal to that place. It's more just like an accountability thing. The software vendor provides the software. It's up to you to make it work. There's this arm's length accountability for it. As you said, Ben, we don't get paid if it doesn't work. It actually really changes the shape of what I think a software vendor would call a post-sales process. You can't just throw software off the wall and say good luck to you because it's actually in our interest to make you successful too. If you don't know how to make your agent work, we want to come help you do that.

Bret：不是。我是认真的。我们今年在用 Salesforce——很棒的 CRM，我很喜欢，我对那家公司也很有感情。这里更像是“责任边界”的问题：软件厂商提供软件，让它真正发挥作用则取决于你自己，这是一种保持距离的责任关系。正如你说的，Ben，如果它不起作用我们就拿不到钱。这实际上改变了传统意义上“售后流程”的样貌。你不能把软件一丢就说“祝你好运”，因为让你成功其实也符合我们的利益。如果你不知道怎么让你的 agent 跑起来，我们希望能进来一起搞定。

We spend as much time thinking about how to be partners to our customers after they purchase our software, as we do before. We have customers who are extremely technically sophisticated, who don't even want to talk to us, and that's great. We have some who barely have an IT team, and they help us. We want you to make it work, and we need to make sure that we match like meet those needs as they come up. I think that's exciting because I think the relationship that companies will have with Sierra will be as different as the difference in relationship that we went through with on-premises software as a service vendor.

我们在客户购买之后，花在“如何做一个好合作伙伴”上的思考和购买之前一样多。我们有些客户技术实力极强，几乎不需要和我们交流——那也很好。也有些客户几乎没有 IT 团队，他们会来找我们帮忙。我们希望你把它用起来，同时我们也需要确保自己能匹配并满足你出现的那些需求。我觉得这很令人兴奋，因为企业和 Sierra 之间的关系，将会像当年从本地部署转向 SaaS 时所经历的关系变化一样“截然不同”。

As Clay said, you're hiring an agent to do a job. That's just a very different relationship than installing a piece of software. I think it's exciting. I think the second of four order effects will be how procurement teams think of what they expect of their software vendors towards outcomes, accountability. As you said, Ben, risk is fundamentally the role here. But I also hope that if we fast forward 10 years and maybe we have the privilege of you're doing an Acquired on us and the magic of our business model is when you talk to our customers, we're the strategic advisor in AI. It's a deeper, more foundational relationship than simply a software vendor.

正如 Clay 所说，你是在“雇佣一个 agent 来完成工作”。这和“安装一套软件”是完全不同的关系。我觉得这很令人期待。我认为接下来更高阶的效应之一，会体现在：采购团队会如何重新定义他们对软件供应商的期望——更看重结果与责任。正如你所说的，Ben，风险本质上是这里的关键。但我也希望，快进 10 年，如果我们有幸被你们做一期 Acquired，届时这套商业模式的魔力会体现在：当你去采访我们的客户时，Sierra 已经是他们在 AI 领域的“战略顾问”，这是一种比“软件供应商”更深、更基础的关系。

David: Which is funny. I feel like the best enterprise software organizations, sales folks, and leaders have always been that to their customers, but the business model wasn't aligned.

David：这很有意思。我觉得最优秀的企业软件组织、销售人员和领导者一直都在对客户扮演那样的角色，只是商业模式并没有与之对齐。

Ben: They weren't really directly incentivized too.

Ben：而且他们并没有得到真正“直接”的激励。

Bret: Yeah, ask ahead of procurement when you get the bill of materials. They're like, what is this stuff?

Bret：是啊，你去问采购（负责人），当你拿到物料清单（BOM）时，他们会说：这些都是什么玩意儿？

Clay: Bret mentioned commission. We work with one very large furniture retailer. We're paid on commission if we attach a premium delivery service to furniture delivery. It was like, we're literally paid on sales commission today. It just broadens the aperture of what's possible with the software. What are all the jobs that a customer facing agent could do? That's where we get really excited about the possibilities here. It's like, what will that look like when a company can show up and at its best in every moment with its customers, with an agent that is fluent and helpful, and can actually get stuff done for you across all parts of the customer lifecycle?

Clay：Bret 提到了“佣金”。我们与一家非常大的家具零售商合作；当我们把“高端配送服务”绑定到家具配送上时，我们就按提成拿钱。换句话说，我们如今真的是拿“销售佣金”。这让“软件能做到什么”的边界被大大拓宽——一个面向客户的 agent 能做哪些工作？这正是让我们对其潜力感到兴奋的地方。想象一下：当一家公司在客户生命周期的每个环节，都能通过一个流畅、能帮上忙、而且真能把事情办成的 agent，在每一刻都以最佳状态出现，那会是什么样子？

Ben: You know what's funny is we just used an old word that is already obsolete in your model, which is post-sale. You were talking about how, oh, post-sale, we're incentivized to go and work with our customers. In fact we demand it. We have to because it's our revenue too. It's actually not post-sale. You've signed a deal, but you haven't made the money.

Ben：有意思的是，我们刚刚用了一个在你们模式里已经过时的旧词——“post-sale（售后）”。你们刚才说，哦，在“售后”阶段，我们被激励去与客户协作；事实上我们还“必须”这么做，因为这也是我们的收入来源。但严格说这并不是“售后”：合同签了不代表你赚到钱了。

David: Right. Sales don't happen until their sales happen.

David：没错。只有客户真的“成交”了，你的“销售”才算发生。

Ben: Right. The whole notion of there's this firm dividing line between pre-sale, then the sale, and then post-sale, being about renewal. Ultimately, what post-sale is about is are we going to get the renewal next year, in three years, or whatever. This breaks that.

Ben：对。传统观念里，“售前—成交—售后（续费）”之间有清晰分界；“售后”的本质在于：我们能否在下一年、三年后等拿到续费。而你们这套模式打破了这种分法。

Bret: It does and actually leads to a couple of other things, which is two things, which is speed to delivery matters a ton and making it super, super easy to set up matters a ton too. To your point, Ben, if it's complicated or slow, you're not earning money until it's live and successful, so we focused on a couple of things. We're typically going live in a small handful of weeks. It can be as low as two or three weeks for an agile firm. It can go as high as a couple months, maybe more traditional company that has a lot of internal gates to go live.

Bret：确实如此，而且由此还带来两点非常重要的事情：其一是“交付速度”极其关键，其二是“极致易用的搭建”同样极其关键。呼应你的观点，Ben，一旦复杂或缓慢，在上线并成功之前我们都赚不到钱——所以我们把重点放在这两点上。通常我们会在几周内上线；对敏捷的公司，2–3 周就能搞定；对更传统、内部上线关卡较多的公司，可能需要几个月。

The other thing though is we spend a lot of time to enable not just technology teams to make these agents, but also their customer experience teams. One of the reasons why IT projects go slowly is you end up with this slow loop of figure out what the requirements are, throw them over the wall, have someone implement it. That wasn't right. Go back and forth.

另外，我们投入了大量精力，不仅让技术团队能做出这些 agents，也让“客户体验团队”能直接上手。IT 项目之所以常常推进缓慢，一个原因是形成了低效的“慢循环”：先梳理需求、再丢给对面实现、发现不对又来回返工。

Clay: Talk separately to the marketing team.

Clay：还得再单独去和营销团队沟通一轮。

Bret: If you think about the furniture retailer, Clay mentioned and like, who's the expert in these premium delivery packages, what's been effective in the past, that's someone on the business team, not someone on the tech team. We have all these no-code tools, so these teams can go in and actually build their agents themselves. No AI expertise, no tech expertise, but it all goes back to the thing you said, Ben, which is we need to empower our customers to make these successful interactions live because it is a gate to our revenue model, but it aligns all these incentives. The reason why we're so focused on going live in days or weeks is because we're as interested in you in that being the case. I really love the incentive alignment that it drives in our product.

Bret：再拿刚才提到的那家家具零售商做例子：谁才是“高端配送方案”的专家、历史上哪些做法有效？答案通常在“业务团队”，而不是“技术团队”。因此我们提供了整套无代码工具，让这些团队可以自己构建他们的 agents——无需 AI 专业背景、也无需技术背景。归根结底，正如你说的，Ben：我们必须赋能客户，让这些成功的交互尽快“上线实战”，因为这既是我们收入模型的“闸门”，也能对齐各方激励。我们之所以强调“以天或周为单位上线”，是因为我们和你们一样在乎尽快落地。我非常喜欢这套模式在我们的产品里带来的“激励对齐”。

Ben: In theory, it should open up way more experimentation for customers. If you don't have to sign a big contract, then be locked into that vendor, have a big implementation time, and owe them a certain amount of money no matter what, it's like, okay, I'll dot five vendors and we'll see which one actually moves the needle for our business. Now that ignores the complexity of there is real setup time and there's human focus as your bottleneck. If we can solve some of these problems, then in theory, companies should just adopt way more partners and see what works in the way that cloud allowed you to just allow your engineers to quickly spin something up versus provisioning a server.

Ben：理论上，这应当让客户能进行更多实验。如果你不必签一纸“大合同”、不被某家厂商锁定、无需长周期实施、也不必“无论如何都欠对方一笔钱”，那就可以同时试用五家供应商，看看谁真的能“动针”（显著改善业务指标）。当然，这里忽略了现实中的复杂度：确实存在搭建时间，而且“人的注意力”往往是瓶颈。如果我们能解决这些问题，那么理论上，企业就会像云计算让工程师“快速拉起一个服务、而不是先去配服务器”那样，更大胆地引入更多合作伙伴，去验证什么真正有效。

Bret: It cuts both ways too. I think one thing, if you talk to a head of technology at a big firm right now, they've probably done too many proofs of concept and don't have enough live successful. It cuts both ways. It's fine to experiment if you have the wherewithal to make decisions and move quickly. When we advise our clients, we often say, have the business metric you're going to go achieve and just go achieve it. Running a lot of experiments can be useful, but often just having the like top down initiative to do it is just as important.

Bret：这也是把双刃剑。我觉得如果你现在去问一家大公司的技术负责人，他们很可能做了太多 PoC（概念验证），真正上线成功的却不够多。这是双向的。如果你具备快速决策与推进的能力，做实验没问题。我们给客户的建议常是：先明确你要达成的业务指标，然后就去实现它。做很多实验当然有用，但很多时候，自上而下地推动落实同样重要。

Ben, there's this other thing I think related to what you said. There's this term in enterprise software, best of platform or best of breed. Best of platform is like that proverb, No one gets fired for buying IBM. It's basically saying, look, if you have a huge enterprise license agreement with one of the big incumbent vendors, Microsoft or whatever. They have a new offering for an ERP system. Buying that, your CEO's not going to be like, you bought an ERP system for Microsoft, what are you, crazy? No one will ever say that. That's where procurement processes and first process are tend towards platforms.

Ben，还有一件与您刚才所说相关的事。在企业软件里有个词对：best of platform（平台优先）和 best of breed（最佳单品）。best of platform 就像那句俗语：No one gets fired for buying IBM（买 IBM 不会被炒）。意思是说，如果你和 Microsoft 等巨头签了很大的企业许可协议，他们推出了一个新的 ERP 系统，你买了它，你的 CEO 不会说“你居然买了 Microsoft 的 ERP，你疯了吗？”没人会这么说。所以采购流程以及前期流程往往倾向于选择平台。

The more a technology is considered a commodity, I think the more it tends towards best of platform because you get essentially commodities of scale. In your big enterprise license agreement, you can get better discounts. You can do all these things. You don't need to onboard a new vendor security, blah, blah, blah.

一项技术越被视为“商品化”，就越倾向于选择 best of platform，因为你基本能享受规模经济。在大型企业许可协议下，你能拿到更好的折扣，还能做一堆配套操作；你也不需要为引入新供应商再走安全审查等繁琐流程，等等。

When new technologies come out, this pendulum swings from best of platform towards best of breed. The reason for that is incumbents typically aren't that great at these new technologies. We talked about business model changes. If you're a software as a service vendor, you have a strategic impediment to embracing new business models. Similarly, just because you're good at making a database in the cloud for ITSM system doesn't mean you're necessarily good at making AI agents. There are technology barriers, there are business model barriers.

当新技术出现时，这个钟摆会从 best of platform 摆向 best of breed。原因在于 incumbents（既有巨头）通常并不擅长这些新技术。我们谈过商业模式变革：如果你是 SaaS 厂商，你在拥抱新商业模式上会有战略性掣肘。同样，你擅长为 ITSM 系统在云上做数据库，并不意味着你擅长做 AI agents。这里既有技术门槛，也有商业模式门槛。

Right now, I think to your point, Ben, people are experimenting a lot more, but also bluntly putting the value of the AI agent in displacing labor costs is so much greater than the software costs. People will go towards the highest quality software right now, which is in companies like Sierra. I don't think that will happen forever. At some point it will be talking on this podcast and it's like, oh, yeah, AI agents, I made one, I made 12 this weekend. It will no longer be technically hard, and then you start to swing back towards platforms. This is the race.

就现在而言，呼应你的观点，Ben，大家在做更多实验；但直说一点，AI agent 在替代人工成本上的价值远大于软件本身的成本，因此大家会优先选择最高质量的软件，也就是像 Sierra 这样的公司。我不认为这种情况会永远持续。总有一天大家在播客里说，“哦，对，AI agents——我做了一个，上周末我还做了 12 个。”届时它将不再是技术难题，钟摆又会摆回平台阵营。这就是竞赛。

Right now, there are best of breed companies like Sierra. Can we gain enough of a clientele and customer base and customer success that in 10 years we are the incumbent, or will we not prove our value to enough people such that when the best practices, these technologies become more commonplace, that incumbents can adopt it? You see this time and time again. I think it's why almost every great technology company was born in a period of technology disruption. The internet gave birth to everything from Salesforce to Google, to Amazon. The mobile phone gave WhatsApp, Uber, DoorDash, Instacart.

当下有像 Sierra 这样的 best of breed 公司。问题是：我们能否积累足够多的客户群与成功案例，让 10 年后我们成为新的 incumbent，还是说我们未能向足够多人证明价值，以至于当最佳实践与技术普及开来，被 incumbents 轻松采用？这种故事反复上演。我认为，这也是为什么几乎每一家伟大的科技公司都诞生于技术颠覆期：互联网孕育了 Salesforce、Google、Amazon；移动时代给了 WhatsApp、Uber、DoorDash、Instacart。

Right now, I just look at all these saplings that are growing right now and which of them will grow into the next generation. Anyway, that's how I think about it. For what it's worth, it's like a race. Right now, quality is all that matters, and it's why our company's growing so well, but it is not something we're entitled to for a decade. We need to essentially create the scale that is necessary as this technology becomes commonplace.

现在我看到许多正在成长的小树苗，未来哪些会长成下一代参天大树？总之，这是我的思考。说到底，这像一场赛跑。眼下唯有质量最重要——这也是我们公司增长这么好的原因——但这不意味着我们能“稳吃十年”。随着技术走向普及，我们必须打造所需的规模。

Ben: You've both led really big teams and you both have been the crack incubation project in the past. Bret, I'm thinking Google Maps or Clay, recently with Project Starline, which is now Google Beam. Is that right?

Ben：你们俩都带领过非常庞大的团队，也都做过“尖刀型”的孵化项目。Bret，我想到的是 Google Maps；Clay，最近是 Project Starline——现在叫 Google Beam，对吗？

Clay: Google Beam. Yeah. Beam me up, Scotty.

Clay：Google Beam。对，“Beam me up, Scotty.”（传送我上去吧，Scotty。）

Ben: Sweet. In building Sierra in this AI era, is there anything different about being leaders of people and leaders of teams versus these almost famous teams that you've led in the past?

Ben：太棒了。在这个 AI 时代打造 Sierra，相比你们过去带领那些几乎“家喻户晓”的团队，在“带人”和“带队”上有什么不同吗？

Clay: I think first and foremost in building a startup, you operate just at an entirely different scale, orders of magnitude smaller scale than Brett and I were operating at in certainly our most recent jobs. There's a proximity to all of the details, all of the work that is important in building and running the business and that I think is part of leadership and demonstrating. I was like, look, we are in a very small boat together and out to build something great. Bret will, on a Sunday night, check in a thousand lines of pristine code, and I will be in the weeds of pricing proposals, our contracts, the exact copy, our marketing language, and so on.

Clay：首先，做一家初创公司，你所处的规模完全不同——比我和 Brett 最近几份工作所处的规模要小好几个数量级。你会更贴近所有细节、所有对建设和运营公司重要的工作，我认为这本身就是领导力与示范：就像在说，“看，我们同乘一条小船，但要造出伟大的东西。”Bret 会在周日晚上提交上千行干净的代码；而我会深陷于定价方案、合同、具体文案、营销语言等细枝末节。

I think first and foremost, it's just a level of being in the details. One thing we've tried to bring to Sierra that I think echoes running these larger teams with broader sets of functions is when you're operating at the intersection between what is possible and what is not yet possible, this zone of the barely doable, it's super important that to the largest extent possible, you'd be able to control your own technology destiny. We talked earlier, we don't do our own pre-training, we don't build our own foundation models.

我认为最重要的就是“深入细节”的程度。我们努力在 Sierra 落地的一点，也呼应了过去管理大型跨职能团队的经验：当你工作在“可行”与“尚不可行”的交界处、也就是“勉力可为”的地带，尽可能掌控自己的技术命运就变得极其重要。正如我们先前提到的：我们不做自有 pre-training，也不自建 foundation models。

We do have a small research team, and I think that's somewhat uncommon for an applied application layer company, but many of the breakthroughs that we've had that have enabled us to deliver such quality and cost savings and more have come through novel agent architectures and really going down a click or two in, in the stack to innovate it at lower levels of the technology stack.

我们确实有一支小型研究团队——对一家应用层公司来说这不太常见。但我们之所以能在质量、成本节约等方面取得突破，很大程度上来自新型的 agent 架构，以及沿技术栈下探一两层、在更底层进行创新。

Ben: Has that felt familiar to you? It seems like your whole career has been in frontier technologies if you look at all the AR and VR.

Ben：这对你来说是不是很熟悉？回看你做过的那些 AR、VR，感觉你的职业生涯一直在前沿技术上。

Clay: Very much so. That's actually one of the parts I love most about building Sierra. You are at this frontier where it's like, can we even make this work? Can we get this thing to do this thing reliably and well? It's not a simple matter of programming. It's not typing into a keyboard or I guess asking cursor to do something until a piece of software emerges. It's exploration, it's discovery, it's posing hypotheses and validating or invalidating those again and again and again.

Clay：非常熟悉。这其实是我最热爱打造 Sierra 的部分之一。你所处的是前沿：我们真能把它做成吗？能否让它稳定、出色地完成这件事？这不只是简单的编程——不是在键盘上敲一敲，或者让 cursor 做点什么，软件就会“冒出来”。而是探索与发现，不断提出假设并反复验证或证伪。

There's a real element of science, exploration, and figuring out how to make this thing work. In addition to then translating those inventions into things that are directly useful for our customers, contrasting what we do today with augmented reality glasses, the development cycle for a wave guide or a display was years or maybe just under a decade. What I love about this is the immediacy. We can have a breakthrough in our agent architecture on Monday, implemented on Tuesday, have it deployed with hundreds of our customers on Wednesday, and directly see the impact of that work. I love the immediacy of that. To have both this invention, discovery, and the unknown, it's very exciting to be in. The direct, practical application of it is super fun and one of the best parts of building the company,

这里面有很强的科学与探索成分——要把“它如何运作”真正摸清楚。与我们当年做增强现实眼镜相比（比如一个波导或显示器件的研发周期动辄数年、甚至接近十年），如今最大的魅力在于“即时性”：周一 agent 架构有了突破，周二就能实现，周三便能在数百家客户处部署上线，直接看到成果的影响。我喜欢这种即时性；既有发明与发现、又充满未知，这种体验令人兴奋。而它直接、务实的落地过程也非常有趣，是创业中最棒的部分之一。

Ben: Bret, how's leadership felt different for you this time around versus Salesforce or Facebook?

Ben：Bret，相比在 Salesforce 或 Facebook，这次你的领导体验有什么不同？

Bret: Plus one to everything Clay said. I think creating a company in the age of AI is interesting because we talked about how software engineering is impacted by AI, but everyone's job is as well. I think one thing culturally that feels meaningful is having active conversations about how to use AI to do our jobs differently. It's an awkward conversation, but if you're a software engineer and you're not using something like Cursor to do your job, you're probably being half as productive or even worse than you could be.

Bret：给 Clay 刚才说的全部 +1。在 AI 时代办公司很有意思——我们谈到 AI 如何影响软件工程，但实际上每个岗位都受到影响。从文化层面看，一个很重要的改变是：积极讨论“如何用 AI 以不同方式完成工作”。这类对话可能有点尴尬，但如果你是软件工程师、却不使用像 Cursor 这样的工具，你的生产率大概只有本可达到的一半，甚至更低。

There's almost like, you want people to adopt these tools because they want to, and you need to volunteer them to do it too. I don't think we can succeed as a company if we're not the poster child for automation and everything that we do, and that feels really different. I have a lot of empathy. I'll just take a real simple example. Salesforce had 80,000 employees. After the pandemic, getting people back to the office was a total pain in the ass. People had moved, people had this, lifestyle changes, and all these things. Every big company goes through it. People who say it's easy haven't run an 80,000-person company. Different people have different approaches. It's just hard.

差不多是这样：你既希望大家主动愿意采用这些工具，也需要他们“自觉”去做。如果我们自己都不是“自动化的样板”，我不认为公司会成功——这种要求确实很不一样。我很能感同身受。举个简单例子：Salesforce 有 8 万名员工，疫情后推动复办公作简直“麻烦透顶”。有人搬了家，有人改变了生活方式，诸如此类。每家大公司都会经历这些；觉得这事儿容易的人，大概没带过 8 万人的公司。每个人、每家公司做法不同，但就是难。

At Sierra, we're in the office company. We just said, if you don't want to be in the office, don't work here. It's super easy. We're a new company. It's just so easy to do these things at a small scale. I observed just having everyone in our company, you didn't use ChatGPT deep research before your sales meeting? Are you kidding me? That's a best practice that everyone should do. Imagine doing that with 10,000 salespeople to roll that out.

在 Sierra，我们是到办公室办公的公司。我们直接说：不想来办公室，就别来这儿工作。对于一家新公司、小规模实施这些规则非常容易。我也观察到，我们要求每个人在销售会议前都用 ChatGPT 做深入研究——没做？你在开玩笑吗？这应该是“人人遵循的最佳实践”。想象把这事儿在 1 万人的销售团队里推广，难度可想而知。

I think about it a lot. Just having the vantage point of having come from larger, I just have a ton of empathy for lack of a better word, the cultural change management of absorbing these technologies into larger organizations. We're trying to be the poster child of it. Because we are a partner to so many larger firms, I have a lot of empathy for the challenges of adopting technology into cultures. I think it's really, really hard, and I have a ton of respect for leaders who are able to do it at a larger scale.

我经常思考这些。也因为我来自更大的组织，我对“大型组织将这些技术融入文化所需的变革管理”——姑且这么说——有很深的共情。我们在努力成为这方面的“样板”。作为众多大型企业的合作伙伴，我很能理解把技术融入企业文化的挑战。这真的非常难，我也非常敬佩那些能在更大规模上做到这一点的领导者。

David: I'm curious. Maybe as a good final question for you guys on this front. About you guys as co-founders, I imagine that must have been extremely intentional because it's not like either of you, given your careers, couldn't have just gone and built a company yourself, probably funded it yourself. You didn't need the team slide to raise money, so to speak, having you both on there.

David：我很好奇。或许这可以作为这部分的收官问题。关于你们作为联合创始人这件事，我想这一定是非常有意的选择。以你们各自的履历，完全可以各自去创办一家公司，甚至自己出资都行。严格说，你们根本不需要靠融资演示里的 team slide（团队页）来募资——把你们俩放上去就够了。

Ben: Or better put, you could have only had the team slide.

Ben：更准确地说，只用一张 team slide（团队页）都够了。

David: For either of you, it would've sufficed. It must have been very intentional. How did you guys think about it?

David：你们中的任何一位单独出马都足够了。所以这肯定是非常有意的决定。你们当时是怎么考虑的？

Bret: I've been trying to work with Clay unsuccessfully every single day since I left Google in 2007.

Bret：自从 2007 年我离开 Google 起，我几乎每天都在尝试拉 Clay 一起共事，但一直没成功。

Clay: Yeah, this was 20 years in the making.

Clay：是啊，这算是“酝酿了 20 年”。

Bret: The short version of this is the only way I could convince Clay to actually work with me to start a company with him, I was like, fine, I'll do it. I'm just kidding. It was like that though. We started in the same program. Marissa Mayer hired us both at Google as associate product managers. We were more or less friends ever since. It was a relatively small group of people.

Bret：简短版是：为了说服 Clay 真正跟我一起创业，我就说——好吧，我来干。开个玩笑，但差不多就是那样。我们当年同在一个项目里，Marissa Mayer 把我们俩都招进了 Google，当时的职务是 associate product managers（助理产品经理）。从那时起我们基本就一直是朋友。那个圈子规模也不大。

Ben: Legendary program.

Ben：传奇项目。

Bret: Totally. We had this monthly poker game that happened roughly twice a year, just because people were busy. We've been friends for a while. Every single place I went, I would call Clay. I'd be like, you got to come here, it's great. Sundar has as high of opinion as Clay as I do, and it was just hard to make everything work.

Bret：确实。我们还有个“每月一次”的扑克局，但因为大家都忙，基本一年就打两次。我们做朋友很久了。我每到一个新地方都会给 Clay 打电话，说：你得来这儿，太棒了。Sundar 对 Clay 的评价和我一样高，但要把一切条件都凑齐真的很难。

We had lots of dinners. Clay may have a different version of this, but I'm just like, I just kept on getting rejected. When I said I was leaving Salesforce, we ended up having this long lunch. We both found out, we shared a passion for large language models.

我们吃了很多次饭。Clay 可能有他的版本，但在我看来就是：我不停地被他婉拒。直到我说要离开 Salesforce，我们吃了顿很长的午餐，发现彼此都对 large language models（大型语言模型）充满热情。

Ben: What year was this?

Ben：那是哪一年？

Clay: This was December, 2022.

Clay：2022 年 12 月。

Ben: Okay, so ChatGPT had just come out.

Ben：好，所以那时 ChatGPT 刚刚发布。

Bret: Just come out. I had announced I was leaving Salesforce, ChatGPT comes out a week later, and we're all just talking about it. I was like, I didn't know what I was going to do, but now I know I'm going to work on this. I don't know what yet.

Bret：刚出来没多久。我宣布要离开 Salesforce，大概一周后 ChatGPT 发布，大家都在聊它。我当时想：我本来还不知道要做什么，但现在我知道我要做这个——虽然具体做什么还没想清楚。

Clay: You thought I was the AR/VR guy, which I was, but also in Labs I had been obsessed with language models and things like Notebook LM, which came out of it. We were both like, okay, we are both obsessed with what is unfolding right now in technology and where this goes. Over that lunch, we hatched plans to start the company together.

Clay：你以为我是那个 AR/VR 的人——确实是——但我在 Labs 时也一直迷恋语言模型，以及由此催生的 Notebook LM 等项目。我们俩当时的感觉是：好吧，我们都痴迷于眼前这场技术变革及其走向。那顿午餐上，我们就商量好一起创业了。

Bret: We had no idea what to do. We figured it out much later in March because you got to get outta your job and do all these things. We just knew we just had the premise, which is this technology's going to change everything. It's going to create a bunch of business opportunities. Let's go. Ride into the darkness and figure it out later.

Bret：一开始我们并不知道具体要做什么。真正定下来得是到了 3 月份——毕竟还要从原岗位卸任，还有一堆事情要处理。我们当时唯一确定的前提是：这项技术会改变一切，会带来一大堆商业机会。那就上路吧，先骑马闯进黑夜，边走边想。

Just to metapoint, I'm just a huge believer in the power of partnership. You've interviewed a lot of entrepreneurs. It's hard. It's stressful. You take everything personally. It is so nice to have a partner to do it with because when you're having the moment and you need to rant at the sky, we can call each other up. I just couldn't imagine doing it solo. I just don't.

再说一个元层面的观点：我非常相信“合伙人关系”的力量。你们采访过很多创业者——这件事很难、压力很大，而且你会把一切都往自己身上揽。能有个伙伴一起做，真的太好了；当你情绪上来、想对着天吼两嗓子时，我们可以互相打电话。我完全无法想象一个人单干——真的不行。

David: It's funny, part of the reason I asked the question, I didn't want to lead the witness too much, but we talked about it at the beginning of our Google episode, the vast majority of companies we cover is the singular founder, the Mark Zuckerberg, even Microsoft, like Bill had Paul Allen.

David：很好玩，我之所以问这个问题，部分原因是不想过多“引导”你们的回答；我们在 Google 专题的开头聊过，我们所讲到的公司里绝大多数都是“单一创始人”主导的，比如 Mark Zuckerberg；就连 Microsoft，像 Bill 也有 Paul Allen。

Ben: They have co-founders, but they're not the main.

Ben：他们有联合创始人，但并不是那个“主角”。

David: The main guy. You guys grew up at Google, which was like a true partnership. Yeah. I would just wonder if that formative experience in seeing Larry and Sergey together rubbed off on you a little bit.

David：就是那个“主心骨”。你们是在 Google 成长起来的，那更像是一种真正的合伙关系。我在想，在你们形成期看到 Larry 和 Sergey 搭档的经历，是否在某种程度上影响了你们。

Bret: It is actually funny you say that too because Clay and Bret, Bret and Clay, has the Larry and Sergey. People talk about us as a unit. They joke around that we spend way too much time together.

Bret：你这么说还真挺有意思，因为 Clay 和 Bret、Bret 和 Clay，有点像 Larry 和 Sergey。大家把我们当一个整体来看，还打趣说我们总是腻在一起，花了太多时间在一起。

Clay: So much time together. Instead of having a holiday party, we have a Sierra birthday party every year in March. Bret and I said a few remarks and someone said, you guys seem to have a really nice dynamic. This is one of the spouses there. I said, yeah, it helps that we actually like each other and like spending time together.

Clay：我们确实花了很多时间在一起。我们不是办节日派对，而是每年 3 月办一场 Sierra 的生日派对。那次我和 Bret 说了几句，有位家属说：你们俩的互动氛围真不错。我就说：是啊，关键在于我们确实彼此欣赏，也喜欢待在一起。

David: The other funny thing is I'm not sure which of you made the better decision after your APM stint of Bret, you obviously created a lot of market cap where you went, Clay, you also, by not going anywhere, created a lot of market cap.

David：另一件有趣的事是，我也不确定你们在 APM 经历之后谁做了“更好”的选择：Bret，你在所到之处显然创造了大量市值；Clay，你则通过“没有离开”同样创造了大量市值。

Bret: Yeah. It turns out both Facebook and Google are pretty good company.

Bret：是啊。事实证明，Facebook 和 Google 都是相当不错的公司。

Ben: Bret, in many ways I feel like you have the single best career of anyone in Silicon Valley in the last 50 years. Do you ever reflect on that, pinch yourself, and go, how the hell did this happen?

Ben：Bret，从很多方面看，我觉得你可能拥有过去 50 年硅谷里“最棒的职业生涯”。你会不会回想时掐掐自己，感叹“这到底是怎么发生的”？

Bret: That's very kind of you. The thing actually I feel most grateful for is to have been inside of some of these remarkable companies. There is a parallel to actually the Acquired podcast. What I've always loved about listening to your overviews of companies is the genuine affection for the companies, business models, what makes them great, and what makes them tick. It exudes from the way you talk about these companies. I feel that way about Google, Salesforce, Facebook, and my own companies that I've started because they're all so different, yet they're all successful.

Bret：你太客气了。我最感恩的是曾在这些了不起的公司内部工作过。说来和 Acquired 这个播客也有相通之处——我一直喜欢听你们讲公司时流露出的那种真切热爱：对公司、对商业模式、对“为什么它们伟大、是什么驱动它们运转”的热爱，你们的表达中处处可见。我对 Google、Salesforce、Facebook，以及我创办过的公司都有同样的情感——它们彼此差异巨大，却都取得成功。

I remember first going into a Salesforce management team meeting and being like, I don't understand anything going on here. It was just so different. Certainly, Quip, the company I'd started but like Facebook and Google, yet it was this remarkably successful company. I was like an anthropologist. I was like Jane Goodall observing the gorillas or something like, what is going on here? I'm taking notes. I'm like, so when he says this, this person does that, and why is that good? I need to figure this out.

我记得第一次走进 Salesforce 的管理层会议，心里想的是：我完全听不懂这里发生了什么——一切都太不一样了。它当然不同于我创办的 Quip，也不同于 Facebook、Google，但却极其成功。那时我像个人类学家，仿佛 Jane Goodall 观察大猩猩一样：这里到底在发生什么？我不停做笔记：当他这么说时，那个人就会那样做——为什么这很好？我得把这逻辑摸清楚。

You end up realizing just the shape of consumer companies and enterprise companies, and I thought I knew what great go-to-market looked like until I went to Salesforce and realized that I had just simply never seen greatness before. I just feel like it's been such a privilege to learn from people like Marissa, Larry, Sergey, and spend a lot of time with Mark Zuckerberg. Mark Benioff is one of the closest mentors I've had in business.

你最终会逐渐看清“消费类公司”和“企业类公司”的“形状”。我原以为自己懂得什么叫出色的 go-to-market，直到去了 Salesforce 才意识到：此前我根本没真正见识过“伟大”。能向 Marissa、Larry、Sergey 这些人学习，并与 Mark Zuckerberg 长时间共事，我觉得非常幸运。Mark Benioff 是我在商业上最亲近的导师之一。

The resume, whatever, and the very kind words you said, but actually for me, just having been there and actually gotten to see what you all cover every day, but first person and actually contribute to it, what a privilege. It's been a fun just to observe some of the great companies of Silicon Valley.

履历也好、你们的溢美之词也罢，于我而言，真正的幸运是亲历其境、以第一人称视角看见你们每天所讲述的那些故事，并实际为之做出贡献——这是巨大的荣幸。能近距离观察硅谷一些伟大的公司，这一路也格外有趣。

Ben: Meanwhile, Clay, you, you got to know the absolute crap out of Google while Bret is doing all that. Eighteen years, is that right?

Ben：与此同时，Clay，当 Bret 在做那些事时，你把 Google 研究得透透的。18 年，对吧？

Clay: Over 18 years, I worked on basically every part of the company, search, ads, then I ran product and design for workspace, and played an enterprise software person on TV for a couple of years because it was both the consumer applications and then Google Apps for work. There was that awkward period of G-Suite before it was workspace, a name I much prefer. I spent most of the last 10 years working for Sundar, building forward-looking things for the company, AR and VR, Google Lens, one of the earlier applications of applied AI, and then most recently rehydrating Google Labs, at least the name as an incubator of forward-looking bets for the company. That's where things like AI Studio, Notebook LM, and some of the more recent AI applications came out of touching on a similar thread as Bre.

Clay：在这 18 年里，我基本上参与过公司的每个部分：搜索、广告，后来负责 workspace 的产品与设计；也有几年“上电视当企业软件人”，因为既管消费级应用，也管 Google Apps for work。还有从 G-Suite 到 workspace 的那个尴尬过渡期——我更喜欢后者这个名字。过去 10 年大部分时间我都在为 Sundar 工作，给公司打造前瞻项目：AR 和 VR、Google Lens（较早的 applied AI 应用之一）；最近则“复活”了 Google Labs——至少在名字上——把它作为公司前瞻性押注的孵化器。AI Studio、Notebook LM 以及一些更近期的 AI 应用都从那里出来，与 Bre 的那条脉络有相似之处。

I just feel such gratitude to have seen greatness up close to have been in some small part of building the company. To have had within 18 years of Google in a way, two, three, four different careers or jobs where I built hardware from scratch and visited assembly lines in China to see headsets and wearables being assembled. That was something that when I joined in 2005 as an APM working on some part of the ad system, it never would've occurred to me. The flexibility, the opportunity, and the privilege of operating with such scale, building something and having it in the hands of hundreds of millions, if not billions of people is truly something. I love my time there. I'm immensely grateful for everything I learned and most of all friends and just amazing colleagues I made along the way.

我非常感激能近距离见证伟大，并在建设公司中尽上一点绵薄之力。在 Google 的 18 年里，某种意义上我经历了两三四段“不同的职业”：从零做硬件，去中国的装配线看头显和可穿戴是如何组装的——这些在 2005 年我以 APM 的身份加入、做广告系统的一小部分时，是完全想不到的。能以那样的规模运作、把东西做出来并交到成百上千万乃至数十亿用户手中，这样的灵活性、机会与特权，确实非同寻常。我热爱在那里度过的时光；我对学到的一切、以及一路结识的朋友与出色同事，心怀万分感激。

David: Amazing. All right, wait, I got one more question before we wrap up. I can't let this be friend lunches, poker games during the Google Plus era. What was your conversation like then? Bret, did you know you were going to win?

David：太棒了。等等，结束前我还有一个问题。我不能让这段经历只停留在“朋友饭局、Google Plus 时代的扑克牌局”。当时你们都聊些什么？Bret，你那时就知道自己会赢吗？

Bret: We've woven in and out. When I started Quip, Clay was working on Google Apps, so we were still cordial. We did talk shop very much during it.

Bret：我们时而交错、时而分开。我创办 Quip 时，Clay 在做 Google Apps，所以彼此关系一直很好。期间我们确实常常谈工作上的事情。

Ben: He was a giant, and Quip was so small. At that time it was like... Sorry, sorry. It was the notion of its time.

Ben：他是巨人，而 Quip 当时很小。那会儿它就像……抱歉抱歉。它是那个时代的 notion。

David: This is Ben's strategy. He builds you up and then he cuts you down.

David：这就是 Ben 的套路——先把你捧上去，然后再给你来一刀。

Bret: It was interesting. Lars Rasmussen, who was one of the guys who created Google Maps with me, also went to Facebook and was part of our poker circle too. We mixed a lot and actually a testament to relationships being deeper than rivalry in some of these places. It was still very fun. We gave each other a little shit, so it was fun.

Bret：很有意思。Lars Rasmussen（和我一起创建 Google Maps 的人之一）后来也去了 Facebook，也在我们的扑克圈子里。我们常常在一起，这其实也证明在这些地方，人际关系往往比竞争更深。那段时间依然很好玩，彼此也会互相吐槽几句，所以很有趣。

Clay: My favorite years at Google were definitely not the Google Plus years. I'll just say that.

Clay：我在 Google 最喜欢的那些年，绝对不是 Google Plus 的那些年——就说到这儿。

David: Bret, you probably weren't even allowed out of the building to go play poker because of lockdown, right?

David：Bret，封锁期间你大概连出楼去打扑克都不被允许，对吧？

Bret: The code rack.

Bret：代码架。

David: Yeah, totally.

David：是啊，完全没错。

Ben: This is something that I think is totally lost to history unless you guys lived it like you. At Facebook, it was an existential threat. We are so scared that Google's actually going to get this right. At Google, the earthquake memo, i mean Google for three years completely reoriented priorities as a company saying, we have to nail social. It wasn't just like a side thing for either company. This was the battlefield, and it ended up actually being a nothing burger. But at the time, it really mattered to both sides.

Ben：这件事我觉得若不是你们亲历者来讲，历史大概会完全忽略。当时在 Facebook，这是生存威胁——我们非常害怕 Google 真的把它做对；而在 Google，则有那份 earthquake memo（“地震”备忘录），也就是说有三年时间里 Google 以公司层面完全重排了优先级，强调“我们必须把社交做好”。对两家公司而言，这都不是什么边缘项目；那就是主战场。虽然最后的结果其实是个 nothing burger（雷声大雨点小），但在当时这对双方都极其重要。

Bret: It did. This is the thing. This is what's going on with AI right now too. When smart people had all these companies realize the size of these markets and at the time, how will sharing within these social graphs and private networks impact the net, search, and all these other things, it feels existential on all sides. I think it's easier to trivialize, it's very easy to make Google Plus jokes. It was a genuine effort and...

Bret：确实如此。关键就在这儿——现在 AI 身上也在发生同样的事。当这些公司的聪明人意识到这些市场的规模，以及在当时社交图谱与私密网络中的分享会如何影响整个互联网、搜索和其他一切时，双方都觉得这是关乎存亡的问题。事后要轻描淡写很容易，拿 Google Plus 开玩笑也很容易，但那确实是一场认真投入，而且……

都缺少真正认真的思考。

David: Easy for you to make.

David：对你来说当然容易。

Bret: Yeah, certainly. I've certainly had my mistakes in the past. We wanted to go through this all today. I think there are a lot of parallels though because when you have a technology incumbent faced with a big new wave of technology, Microsoft famously fumbled on mobile despite Windows phone and Windows Mobile being ahead of many of the other operating systems at one point, did very well in cloud, but both were treated with a lot of gravity at that company. Right now, just that analogy, part of it was born of the personal rivalries and the staff weaving between Facebook and Google, which was somewhat unique to that time.

Bret：对，没错。我过去当然也有失误。我们今天只是想把这些都过一遍。我觉得这里有很多相似之处：当一家技术上的既有强者面对一波全新的技术浪潮时——Microsoft 在移动端的失手众所周知，尽管 Windows phone 和 Windows Mobile 曾在某个阶段领先过许多其他操作系统；它在云上又做得很好——而且这两件事在公司内部都被极其严肃地对待。就拿这个类比来说，当年还有一部分是由个人层面的竞争，以及 Facebook 与 Google 之间人员的交错流动所推动的，这在当时多少有其独特性。

Put another way, I think I joke, there's a corporate strategy and then there's pure ego. I think it was a mix of a lot of the two. I think you can see the same thing in AI right now. Everyone's trying to recognize this wave of technology is going to dramatically change markets and what do we want to be when we grow up. You're going to see the equivalent of earthquake at a lot of different companies right now, given the wave of AI.

换句话说，我常打趣：一方面是公司战略，另一方面是纯粹的自我（ego）；当时两者的成分都很多。现在的 AI 身上你也能看到同样的情况：每家公司都在试图认清这波技术将如何剧烈改变市场，并思考“我们长成之后要成为什么”。鉴于这波 AI 浪潮，你现在会在许多公司看到相当于“earthquake”的动作。

Ben: Bret, Clay, thank you so much for coming on with us.

Ben：Bret、Clay，非常感谢来到我们的节目。

Bret: Thanks for having us.

Bret：谢谢邀请。

Clay: Thank you so much for having us.

Clay：非常感谢你们的邀请。

Ben: Listeners, we'll see you next time.

Ben：各位听众，我们下次见。

David: We'll see you next time.

David：下次见。

Note: Acquired hosts and guests may hold assets discussed in this episode. This podcast is not investment advice, and is intended for informational and entertainment purposes only. You should do your own research and make your own independent decisions when considering any financial transactions.

Note：Acquired 的主持人和嘉宾可能持有本期节目中讨论的相关资产。本播客不构成投资建议，仅供信息参考与娱乐目的。在进行任何金融交易前，请自行研究并独立作出决策。

2025-08-18 Acquired.How is AI Different Than Other Technology Waves?

2025-08-18 Acquired.How is AI Different Than Other Technology Waves?

热门主题

Recent Articles

2003-02-21 Warren Buffett's Letters to Berkshire Shareholders

2004-02-27 Warren Buffett's Letters to Berkshire Shareholders

2005-02-28 Warren Buffett's Letters to Berkshire Shareholders

2006-02-28 Warren Buffett's Letters to Berkshire Shareholders

2007-02-28 Warren Buffett's Letters to Berkshire Shareholders