How AI is Reinventing Software Business Models ft. Bret Taylor of Sierra

频道: Sequoia Capital
视频: https://www.youtube.com/watch?v=xlQB_0Nzoog
原文语言: en
统计: 共 61 轮

[0:02]

Next up, we have Brett Taylor. He was part of the, I think, first APM program at Google where he famously rewrote all of Google Maps in a weekend. Then he founded a company. Then he went ahead and was just CTO of Facebook for a while. And then he founded another company, which we're lucky enough to be in business with him on, which is redefining the front door for businesses. and Brett will be joined by our partner RVY who used to deliver groceries and now gives people money. About the same. Please welcome RV and [Applause] Brett. Yeah, but a lot of groceries.

下面这位是 Bret Taylor。他参加了 Google 我记得是第一届的 APM 项目，在那儿他有个著名的事迹——一个周末就把整个 Google Maps 重写了一遍。后来他创办了一家公司，再后来去 Facebook 当了一阵子 CTO。然后他又创办了另一家公司，我们也很荣幸能跟他合作，这家公司正在重新定义企业的「门面」。今天和 Bret 对谈的是我们的合伙人 RVY，他以前是送菜的，现在是给人发钱的——其实差不多啦。有请 RV 和 Bret。（掌声）是啊，不过那可是好多菜呢。

[0:41]

Okay. Um I thought Pat was saying I was the legend, so I was pretty pretty upset when apparently it was Brett. You're a legend to me. Thank you. I really appreciate you saying that. Having Jeff Dean and then Brett Taylor is like having like LeBron and Steph of engineers. It's amazing. Or Ruof, two random rugby people that only you know, you know, like that maybe an analogy you'd get. Um, all right. I want to ask you a question. There's very few people that have been described as 10,000x engineers probably in general, but also extremely extremely few people who've done that and also been good at enterprise sales.

好。嗯，我本来以为 Pat 说的传奇人物是我，所以一听说原来说的是 Bret，我还挺失落的。在我心里你就是传奇。谢谢，真的很感谢你这么说。先请来 Jeff Dean，再请来 Bret Taylor，这就好比工程师界的 LeBron 加 Steph，太顶了。或者用 Ruof 来打比方——那是两个只有你认识的橄榄球运动员，可能这个比方你能 get 到。好，我想问你个问题。能被称为「一万倍工程师」的人本来就极少，但同时又能把 enterprise 销售做得很好的人，那是少之又少。

[1:17]

Okay. Um, and I think one of the things that'll be fun today is we can like be like high flutin and talk about where the world's going, but we can also talk about like how to sell things. And so the question I have before we get to tactical advice on that is like what made you even think that you could do something other than just be an amazing engineer? Yeah, I actually think this is uh one of the greatest challenges to overcome as an entrepreneur. uh this is really reductive but I think if you talk to an entrepreneur they often have some unique insight on the business they're creating. So maybe it's a technical

好。嗯，我觉得今天好玩的地方在于，我们既可以聊得很高大上、谈谈世界往哪儿走，也可以接地气地聊聊东西到底该怎么卖。所以在进入那些实操建议之前，我想问的是：当初是什么让你觉得，除了做一个超级厉害的工程师，你还能干点别的？是这样，我其实觉得这是创业者要克服的最大挑战之一。这么说可能有点简单粗暴，但我觉得，你去跟创业者聊，他们往往对自己要做的生意有某种独到的洞察。可能是一个技术上的

[1:49]

insight. If they come from a background in say business development, maybe they see an opportunity in the market uh that they can take advantage of as a partnership or business opportunity uh in the financial services industry. Maybe you see an inefficiency in the market you can take advantage of. The issue I see with a lot of entrepreneurs is you sort of become single issue voters. It's like every time there's a problem uh in your business, if you're a product person, you go change the product. There's always the like famed redesign like the dead cat bounce before your company gets destroyed. And it

洞察。如果他出身于商务拓展，可能他在市场里看到了某个机会，能通过合作或者商业方式去抓住它，比如在金融服务行业里。也许你看到了市场里某个低效的地方，可以拿来做文章。我看到很多创业者的问题是，你会变成那种「单一议题选民」。就好比每次公司一出问题——你要是产品出身，你就去改产品。总会有那种著名的「最后一搏式重设计」，就是公司被搞垮之前那一下「死猫反弹」。而事实上

[2:21]

turns out that as you're growing your business at any given point, you may have a go to market problem, you may have a product problem, you have a competitive problem. And I think the hardest thing is when you come out of business with what you're comfortable with, you regress to sort of what makes you comfortable. For me, uh my most like vulnerable moment was when I was 29. and I became the CTO of Facebook. Uh I was one of the older people at Facebook at the time.

随着公司的成长，在任何一个时间点上，你可能面临的是 go-to-market 的问题，可能是产品的问题，也可能是竞争的问题。我觉得最难的地方在于：当你带着自己擅长、舒服的那套东西去面对时，你会下意识退回到让自己舒服的方式里。对我来说，我最脆弱的时刻是 29 岁那年，我当上了 Facebook 的 CTO。当时我算是 Facebook 里岁数偏大的人之一了。

[2:48]

Good. Well, I didn't know what the what I was about to cuss. I shouldn't cuss. I didn't know what I was doing. You told me this was recorded. I have to like I think that you should not give away state secrets about open a fine. It's good. And um I uh I wouldn't have self-described that I was sort of struggling in my job, but I think I was. I was sort of trying to do everything myself. Uh, and Cheryl Sanberg, who was the chief operating officer at the time, took me aside and kind of had a sort of come to Jesus conversation with me. Um, essentially saying, "You need to hold your team to as high of a standards you hold

挺好。嗯，我都不知道自己刚才要爆什么粗口——我不该爆粗口。你跟我说过这是要录下来的。我得……我觉得你不该把 OpenAI 的国家机密抖出来。没事的。嗯，我当时不会承认自己工作上有点吃力，但其实我是有的。我那时候什么都想自己一个人扛。后来当时的首席运营官 Sheryl Sandberg 把我拉到一边，跟我来了一场「掏心窝子」的谈话。她大意是说：你对团队的要求，得跟对自己一样高。

[3:19]

yourself. You need to stop trying to do the work yourself and you need to identify the people who aren't meeting your needs, replace them, but you need you're not scaling the way you're operating effectively." And it was kind of the conversation I needed to have at that moment. I felt really shitty for about an hour and then like woke up the next day with kind of a fire in my belly. And I realized upon reflection uh she didn't sort of diagnose it this way, but upon my reflection I was going into work every day and trying to say how do I conform this job to me and my strengths. Um and instead of thinking

你得别再什么活都自己干，你得识别出那些达不到你要求的人，把他们换掉。你现在这种干法根本撑不起规模。这正是我那个时刻需要的一番话。我难受了大概一个小时，然后第二天醒来就跟打了鸡血一样、心里燃起一把火。事后回想我才意识到——她当时倒不是这么诊断的，但我自己反思后发现——我每天去上班，脑子里想的都是「怎么把这份工作改造成适合我、迎合我长处的样子」。而不是去想

[3:55]

like what's the most important thing to do in this job today even if it's not something I find delightful or interesting. So, I sort of came at the job uh and I I always modeled it as giving myself advice like I was sort of my own board of directors and sometimes it would be people, sometimes it would be business, sometimes it would be technology. I was the CTO so hopefully it wasn't that some of the time. Um and interestingly enough, um I got more joy from it. Um I realized at that point I actually just appreciated having an impact more than I appreciated any act of doing my job. And so I ended up with

「今天这份工作里最重要的事是什么」，哪怕那件事我并不觉得有趣或者享受。所以后来我换了个角度去面对工作。我一直把它想象成给自己提建议，好像我是自己的董事会——有时候建议是关于人的，有时候是关于业务的，有时候是关于技术的。我是 CTO 嘛，所以但愿不会大部分时间都在纠结技术那块。有意思的是，我反而从中获得了更多快乐。我那时候意识到，比起做这份工作里任何一件具体的事，我其实更看重自己能带来影响。结果我就形成了

[4:28]

this really nice reinforcement of, wow, I'm better at my job than I was last month because I'm not just doing what I like or what I feel comfortable doing. Um, and that ended up like a really virtuous cycle. So I've had a I've been a little bit of a chameleon. I've, you know, it's interesting. If I meet people from Facebook, they think of me as an engineer. If I meet people from Salesforce, they think of me like a suit, you know, like or I don't know, generously. I something the boss. And I'm not any one of those things. I just sort of showed up every day to, you know, be uh to have an impact. And as I

一个特别好的正反馈：哇，我这个月比上个月更胜任这份工作了，因为我不再只挑自己喜欢的、做着舒服的事去干。这后来就变成了一个非常良性的循环。所以我有点像个变色龙。挺有意思的，要是我遇到 Facebook 出来的人，他们眼里我是个工程师；要是遇到 Salesforce 出来的人，他们眼里我是个穿西装的、或者说得好听点——是个领导。可这些身份我哪个都不是。我只是每天来上班，想着怎么去带来影响。就像我

[5:00]

said, I just I gain a ton of joy from it. And the reason I bring it up for the entrepreneurs in the room is I think at every stage of your company at the beginning, product is all that matters. You know, finding your first and happy customers. As you scale, what you need to be great at changes. And I think having uh the self-awareness and self-reflection to actually change what you spend your time on is actually I think one of the greatest challenges of not having your company sort of get away from you and continuing to be effective leader at that company. I I would tell you I think it's something just so you

刚说的，我从中收获了巨大的快乐。我之所以跟在座的创业者们提这个，是因为我觉得，公司在每个阶段——一开始，产品就是一切，关键是找到你最早的、满意的那批客户。等你做大了，你需要擅长的东西就变了。我觉得，能有足够的自我觉察和反思，真正去改变自己把时间花在哪儿——这恰恰是最大的挑战之一，决定了你的公司会不会失控，以及你能不能继续做那家公司里一个有效的领导者。我跟你说，这事儿——你记着就行——

[5:33]

have it. It's something I have taken and sharing with my kids which is like don't let something you're good at become who you are. Right. And I think that that is something I've learned from you that I really appreciate and I think you're doing in the day-to-day. So tell people about Sierra. Why did you and Clay start it? What's the vision? What are you guys up to? Uh at Sierra, we help companies build customerf facing AI agents. Um we started on the hypothesis that uh every company's main digital interface will be an AI agent, you know, and maybe in in uh 20 years ago, your main digital presence was your website, your dot. Uh

这是我也拿去跟我孩子分享的一句话：别让你擅长的某件事，变成了你这个人本身。对。这是我从你身上学到的、特别受用的一点，我觉得你在日常里也确实是这么做的。来跟大家讲讲 Sierra 吧。你和 Clay 当初为什么要做这家公司？愿景是什么？你们在忙些什么？在 Sierra，我们帮企业打造面向客户的 AI agent。我们一开始的假设是：每家公司主要的数字界面，都会变成一个 AI agent。你想，20 年前你主要的数字门面是你的网站、你的「dot com」。

[6:10]

remember listing on Yahoo directory was the bees knees, you know, back in the day. Um I do I do remember that. Yeah. Thank you for Yeah. Thank you for noticing. Yeah. You know, and then for a lot of brands, whether it was like your social profile or your mobile app, you know, companies have a lot of different digital touch points today. Our hypothesis, I don't know if you measured it like what percentage of digital interactions happen on a touchcreen now versus a keyboard. It never replaces, but it certainly sort of displaces. Our hypothesis is you fast forward 5 years, the vast majority of digital

想当年能上 Yahoo 目录那可是顶顶厉害的事。我记得，我是真记得。是啊，谢谢你……谢谢你还记得。是啊。后来对很多品牌来说，不管是社交主页还是手机 App，企业如今有了一大堆不同的数字触点。我们的假设是——不知道你有没有量过这个数据——现在有多大比例的数字互动是发生在触摸屏上、而不是键盘上的。它从来不是完全取代，但确实会挤占掉一部分。我们的假设是：往后快进 5 年，绝大多数的数字

[6:42]

interactions will happen via an agent. And as a consequence, every company will need their singular agent, not plural, the one that has their brand at the top that represents the whole of their customer experience. And um so we've uh worked with a lot of different companies from ADT home security to Sirius XM to help them build their primary customer-f facing AI agent. Um a lot of it is AI and it's an agent, but a lot of it is also business. You know, like what is the customer experience you're providing for ADT home security? your alarm might be going off, your alarm might be broken for SiriusXM, someone might want a

互动都会通过一个 agent 发生。其结果就是，每家公司都会需要它那个「唯一」的 agent——是单数，不是复数——就是顶上挂着它品牌、代表整个客户体验的那一个。我们和很多不同的公司合作过，从 ADT 家庭安防到 SiriusXM，帮他们打造面向客户的主力 AI agent。这里面很大一部分是 AI、是 agent，但也有很大一部分其实是业务。比如，你给 ADT 家庭安防提供的客户体验到底是什么？可能是你家警报响了，可能是警报坏了；对 SiriusXM 来说，可能有人想要

[7:16]

better price on their plan uh because their promotional period expired. So, it's actually really interesting because just like your website, I don't think most people think of websites as a piece of technology nowadays, right? It's almost just an expression of your brand and your business, that's actually kind of the agents that that companies are building on our platform. Uh but I hope, you know, if we're successful, if you run across an branded AI agent in the wild, I hope it's powered by our platform. That's what we're trying to do. And one of the things that comes up a lot is like what are the foundation

一个更划算的套餐价格，因为他的促销期到期了。所以这其实特别有意思，因为就跟你的网站一样——现在大多数人已经不把网站当成一项技术了，对吧？它几乎就是你品牌和业务的一种表达。这其实也正是那些公司在我们平台上打造的 agent 的样子。但我希望，如果我们做成了，将来你在世界上随便撞见某个挂着品牌的 AI agent，我希望它背后跑的就是我们的平台。这就是我们想做的事。还有一个经常被提起的问题是：foundation model 会去做哪些事，而应用层的公司又该去做哪些事，对吧？

[7:44]

models going to do versus what are the application layer companies going to go do, right? And I think you have probably the most unique vantage point on this being the chair of OpenAI as well as building an application layer company. So can you help our founders here think about what is going to belong to them versus the foundation models? Yeah, I've um I'll say strong opinion loosely held uh on this. So I think there's really three big markets that I see and I'll end with the one I think is the most exciting. Uh first you have the foundation models and Jeff said something similar to this but I think

我觉得你在这件事上的视角大概是最独特的——你既是 OpenAI 的董事长，又在做一家应用层的公司。所以你能不能帮在座的创始人们理一理：哪些东西会归他们，哪些会归 foundation model？可以。这事儿我是「立场坚定，但随时准备松手」。我看到的大致有三个大市场，我会把我觉得最让人兴奋的那个留到最后说。首先是 foundation model。Jeff 也说过类似的话，但我觉得

[8:16]

there's going to be a ton of consolidation in the foundation model market. It's fundamentally just a capital intensive business just like building data centers. Effectively it is building data centers you know at the core of the expense. And as a consequence, it just, you know, if you have large-scale capex, you can generate more capex and it sort of benefits scale. And as a consequence, I think it'll end up a little bit like the cloud infrastructure market, a handful of players, relatively low margin, but very very high scale and sort of collecting taxes from everyone in the AI ecosystem.

foundation model 这个市场会出现大量的整合。它本质上就是个资本密集型生意，跟建数据中心一个道理。从成本核心来看，它实际上就是在建数据中心。结果就是，你要是有大规模的 capex，你就能产出更多 capex，它天然偏向规模效应。所以我觉得它最后会有点像云基础设施市场：玩家就那么几个，利润率相对较低，但规模极其庞大，相当于在向整个 AI 生态里的每个人「收税」。

[8:47]

Uh I think the the next market is is really making tools, the proverbial pickaxes and the gold rush. Um and uh I think you know a lot of uh companies created prior to large language models like you know data bricks and snowflake will sort of be in that category but a lot of new ones as well. Um there's a lot of nuance there. I think that will be a market that can really uh be threatened by the foundation model companies. I imagine every cloud infrastructure company has some there's probably an AWS or Azure equivalent to what they do. And so you end up with a naturally sort of more challenging, but

嗯，下一个市场是造工具的，就是淘金热里那些卖铲子和镐头的。我觉得很多在大语言模型出现之前就成立的公司，比如 Databricks、Snowflake，会归到这一类，但也会冒出一大批新公司。这里面门道很多。我觉得这会是个很容易受到 foundation model 公司威胁的市场。我猜每家云基础设施公司做的事，大概都能找到一个 AWS 或 Azure 的对应版本。所以你天然就处在一个更艰难的位置，不过

[9:24]

it can be very high scale. Um, but it's just a you definitely are closer to the sun, you know, the closer you are to the models themselves. And then there's the AI applied AI market. And I actually think this will manifest its form as as agents. Um, so uh, you know, there's companies like Harvey that make an agent for the legal profession. There's companies like Sierra that make it for customer experience. Uh, there's companies that make it for marketing.

它的规模可以非常大。只是说，你离模型本身越近，就越是「离太阳越近」。然后就是应用 AI 这个市场了。我其实觉得它最终会以 agent 的形态呈现出来。比如有 Harvey 这样的公司，给法律行业做 agent；有 Sierra 这样的公司，给客户体验做 agent；还有给营销做的公司。

[9:47]

There's companies that do for visual effects. Uh my view is this is sort of the new software as a service. Um companies will the form factor of purchasing AI will be purchasing an agent that does a job. Um I I think there's a it's a really exciting time. I actually I think I have assume a lot of folks in this room are working on sort of applied applications in AI. Um the reason I think it's exciting are sort of twofold. One is I think it's the way software should be consumed. Uh I think right now a lot of particularly large enterprises uh in the excitement after chat GBT licensed a lot of models and

还有做视觉特效的公司。我的看法是，这基本就是新一代的 SaaS。以后采购 AI 的形式，就是采购一个能干某项活儿的 agent。我觉得这是个特别让人兴奋的时代。我猜在座很多人做的就是 AI 的应用层。我之所以觉得它令人兴奋，大概有两个原因。第一，我觉得这才是软件本该被消费的方式。我觉得现在很多公司、尤其是大型 enterprise，在 ChatGPT 出来之后的那股兴奋劲儿里 license 了一大堆模型，

[10:23]

and as we all know software is like a lawn. You have to tend to it. So I think there's a lot of buyer remorse right now from a lot of spend and a lot of proofs of concept but not a lot of value. Um the second thing is I think it changes the markets. you know, there's these markets where like selling enhanced productivity to attorneys is not like a huge tam and so you sort of run up against just a natural thing like even if you get you know 80% market share how big could this company be you know but if you're actually selling antirust review or or something I don't know much about the law so I just made that sound

而我们都知道，软件就像一片草坪，你得不停地打理它。所以我觉得现在很多人有「买家后悔」的情绪——花了一大笔钱、做了一大堆概念验证，但没产生多少价值。第二点是，我觉得它改变了市场格局。你看，有些市场——比如把「提升效率」卖给律师——它的 TAM 本来就不大，所以你很快就会撞上一个天花板：哪怕你拿下 80% 的市场份额，这家公司又能做到多大呢？但如果你卖的其实是反垄断审查之类的东西——我对法律不太懂，所以刚才那个例子是我瞎编的——

[10:56]

credible really you you are the you know crisis to the stars and you don't know anything about the law that seems you know it turns out the crisis to the stars um you know uh that has a ton of value um because you're essentially you know providing something that very high cost labor produced before and as a consequence I think if you look at the total addressable market of you know all these vertical specific domain specific agents the market is gigantic um and actually it's interesting if you look at the public markets if you look at the top uh you know five software companies that dominate um the S&P 500 that

这是很有说服力的——你是那种行业里数一数二的律师，但你对法律一窍不通，这听起来很矛盾，但事实证明，那种顶尖律师其实有巨大的价值，因为你本质上是在提供过去靠高成本人力才能产出的东西。正因如此，我觉得如果你去看所有这些垂直领域、特定行业的 AI agent，它们加起来的整个潜在市场是极其庞大的。而且很有意思，如果你看公开市场，看主导标普 500 指数的前五大软件公司，

[11:34]

dominate the stock market. None of them are sort of pure play software as a service companies. Obviously, Microsoft is and it has some enterprise software. Amazon has AWS, but they're infrastructure companies. They're consumer. Um, and if you look at the software as a service companies, Service Now, Salesforce, SAP, Oracle, they're all sort of in that sort of 200 to30 billion market cap range, you wonder in this new world of agents, will we see our first trillion dollar sort of applied enterprise software company? And I I think the answer is yes. Uh because you're going from selling productivity

也就是主导整个股市的那几家，它们没有一家是纯粹的 SaaS 公司。Microsoft 当然算，它也有一些 enterprise 软件；Amazon 有 AWS，但他们更像基础设施公司，也做消费业务。而真正的 SaaS 公司——ServiceNow、Salesforce、SAP、Oracle——市值大概都在两千亿到三千亿美元这个区间。你会想，在 agent 这个新世界里，我们会不会看到第一家市值上万亿美元的应用型 enterprise 软件公司？我觉得答案是肯定的，因为你正在从卖生产力提升，

[12:07]

enhancement to selling outcomes and outcomes are valuable. Uh some outcomes are extremely valuable. Um and so I think there's been a little understandably a little too much excitement around the models themselves. Not that I I mean obviously at Open Eye were very excited about it, but as an entrepreneur I think that it's sort of the the pickaxes are obvious, but actually the value is elsewhere in my mind. I think the value is taking these this amazing next generation technology solving a very valuable business problem that was high cost before and selling something valuable for an order of magnitude less money than the current

转向卖结果，而结果是有价值的，有些结果价值极高。所以我觉得，大家对模型本身的兴奋有点过头了，这也可以理解——当然在 OpenAI 我们对模型非常兴奋，但作为创业者，我觉得卖铲子这件事是显而易见的，可在我看来真正的价值在别处。我觉得价值在于：拿这项了不起的新一代技术去解决一个非常有价值、过去成本很高的商业问题，然后把这个有价值的东西用比现有成本低一个数量级的价格卖出去，

[12:45]

cost is really easy. Um so I just think it's a great business and I'm I'm super excited about the future of the software industry as a consequence. What do you think is the right way to price those outcomes when you think about them? Right. I think that there's all the former seatbased models, right? And now Sierra obviously does it differently. How applicing model and then also talk about how applicable you think that is or isn't to other agent companies. Yeah, at Sierra we do we call it outcomesbased pricing. So uh for our median customer that typically means when the AI agent resolves the issue for

这其实很容易。所以我觉得这是门很棒的生意，也因此我对软件行业的未来超级兴奋。那你觉得给这些结果定价的正确方式是什么？过去都是按席位收费的模式，对吧？现在 Sierra 显然做得不一样。能聊聊你们的定价模型吗？再说说你觉得它对其他 agent 公司适用还是不适用。好。在 Sierra，我们把它叫作 outcomes-based pricing（按结果付费）。对我们的中位数客户来说，这通常意味着：当 AI agent 自主帮客户解决了问题，

[13:19]

the customer autonomously there's a pre-negotiated rate for that and if we do have to escalate to a person it's free. uh we do that just to align with the the business model of our customers and uh I actually view it as sort of a natural evolution of software. We went from uh box software with a perpetual license to you know once you delivered software in a browser most everyone's on the same version. you needed to sort of invent a new business model to sustain R&D which led to software as a service subscription based business and we just thought from first principles and said hey if you're selling software that

就按事先谈好的费率收费；如果我们不得不转交给人工处理，那就免费。我们这么做是为了跟客户的商业模式对齐，而且我其实把它看作软件的一种自然演进。我们从盒装软件加永久授权，走到了把软件交付到浏览器里——一旦这样，几乎所有人都用同一个版本，你就得发明一种新的商业模式来支撑研发投入，于是有了基于订阅的 SaaS 商业模式。我们就从第一性原理出发去想：如果你卖的软件是

[13:53]

completes a job what is sort of the secular business model for that and it felt like let's pay for a job well done you know sales people get paid a sales commission you know why not the AI as well um it it does really disrupt the way you build a software company in my mind though um if you look at particularly the enterprise software ecosystem there tends to be sort of an armslength relationship between software vendors and customers. Uh typically you'll sell a piece of software and be done. Uh then you'll have systems integrators come in and spend you know for large scale enterprise software deployments

完成一项工作的，那它最自然的商业模式是什么？感觉答案就是——为干得好的活儿付费嘛。销售人员能拿销售提成，那为什么 AI 不行呢？不过在我看来，这确实彻底颠覆了你做一家软件公司的方式。尤其看 enterprise 软件这个生态，软件厂商和客户之间往往是一种保持距离的关系：你通常把一套软件卖出去就完事了，然后系统集成商进场——对于大型 enterprise 软件部署——花上

[14:27]

sometimes 6 months or a year to implement that software. Um and that's well I mean by the time the software is being used you know the vendor has been probably not gone they're probably going to sell them more things but effectively gone for uh 6 to 12 months that so as a consequence there's almost like an active disassociation with the act with the outcomes um and so you end up with a confluence of I think a lot of different factors that I think are interesting uh there was a question just before I walked up here about you know uh agents that can code what will that do to the professional services industry broadly

半年甚至一年去实施那套软件。等到软件真正开始被用起来，厂商虽然不能说彻底走人了——他们大概率还会再卖给客户更多东西——但实际上已经有六到十二个月不在场了。结果就是，厂商几乎是在主动跟最终结果脱钩。所以你会遇到一堆我觉得很有意思的因素交织在一起。我上台之前刚好有人问到，会写代码的 agent 会对整个专业服务行业产生什么影响。

[15:01]

um there's a question of if the right thing for AI software is outcomes, how do you actually align your company with that kind of accountability? I think it sort of blows up a lot of the model. Um but that's why, you know, as Steve Jobs said, it's more fun to be a pirate than it is to be in the Navy. And I think it's a really good time to be a startup because, uh most people in this room are not encumbered by a business model. Um which is a funny way of thinking about it. But if you look at the history of software, Salesforce beating seable systems uh you know service now beating the plethora of ITSM companies that came

还有一个问题是：如果 AI 软件该卖的是结果，那你到底要怎么让公司去承担那种责任、跟结果对齐？我觉得这会把很多旧模式都炸开。但这也正是为什么——就像 Steve Jobs 说的——当海盗比当海军有意思。我觉得现在是做创业公司特别好的时候，因为在座大多数人都没被某种商业模式束缚住。这么说挺有意思的。但你看软件的历史，Salesforce 打败 Siebel Systems，ServiceNow 打败它之前那一堆 ITSM 公司，

[15:34]

before it's actually much closing a technology gap in your product is is hard but not impossible. Changing your business model is really hard. Uh, and you look at Microsoft transitioning from Windows to Azure and how incredibly awkward that transition was and Satcha deserves a bunch of credit. Shantenu did a great job at Adobe going to ratible revenue, but there's like a graveyard of CEOs who've been fired because they couldn't make that transition in part because public company investors are horribly impatient in these things. And so I just think that what's exciting about AI is I think it's going to give rise to new delivery

其实，在产品上追平技术差距是难，但并非不可能；真正难的是改变你的商业模式。你看 Microsoft 从 Windows 转向 Azure，那个转型有多么别扭，Satya 确实功不可没。Shantanu 在 Adobe 转向订阅式经常性收入也做得很出色，但有一大批 CEO 就是因为没能完成这种转型而被炒掉，墓地里堆满了他们——部分原因是上市公司的投资者在这些事情上极其没耐心。所以我觉得 AI 让人兴奋的地方在于，它会催生新的交付

[16:14]

models, new technology models, but as importantly, new business models. And I think that if you think about what are the opportunities for startups relative to incumbents, don't lose sight of the business models. I actually think it's usually um one of the greatest advantages that startups have relative to incumbents. Yeah, I I think historically speed has been something, but I think on this point it's like you also have this enormous benefit of, you know, you're not encumbered by the old business model. Well, you referenced the uh previous companies that you've been in and you started. One of the things

模式、新的技术模式，但同样重要的是，还有新的商业模式。我觉得，如果你去想创业公司相对 incumbent（在位巨头）的机会在哪，千万别忽视商业模式。我其实觉得，这通常是创业公司相对 incumbent 最大的优势之一。对，我觉得历史上速度一直是个优势，但在这一点上，你还有一个巨大的好处，就是你不被旧的商业模式束缚。对了，你提到了你之前待过和创办过的公司。我们之前聊过一件事，

[16:41]

we've talked about is like there are things that are similar about building a company now and there are things that are different. What are some of the things that are the same as the previous companies that you've built that you're doing now? And what are some of the things that are different? Uh well, at Friend Feed, my not so successful social network, we built our own servers. Can you tell the story? Can you tell the story about the Facebook IPO real quick?

就是现在创业有些地方跟以前很像，有些地方又很不一样。在你现在做的事情里，哪些跟你以前创办的公司是一样的？又有哪些是不一样的？嗯，在 FriendFeed，我那家不太成功的社交网络，我们当时是自己搭服务器的。能讲讲那个故事吗？能先快速讲一下 Facebook IPO 那个故事吗？

[17:03]

just about friend feed for everyone where uh you got recognized. Um can you I'll I'll do that then. I'll do it then. I'll do that. Okay. It's a good story. It's a good story. We stand between these people and drinks, man. You got to make them laugh like let's go. You know, hit them with something. You're pulling out the deep cuts. Um you know, I think a lot has changed just in the infrastructure world. You know, I think literally we built our own servers uh at friend feed and you know like brought them into the colo. We brought our site down once by tripping over a power wire in the colo. Uh that

就先给大家讲讲 FriendFeed，讲讲你是怎么出名的。你能不能——我等下讲，等下讲，我来讲。好。这是个好故事，好故事。哥们儿，我们现在挡在这些人和酒水之间呢，你得逗他们笑啊，来点儿料，整点儿活。你这是在翻压箱底的料啊。我觉得光是基础设施这块就变了很多。我们在 FriendFeed 真的是自己搭服务器，搬进机房（colo）里。有一次我们在机房绊到一根电源线，把整个网站搞挂了。那次

[17:34]

was Sajjie Sajie's fault. We love him. No, I'm just kidding. It was like so it's just a different era. Um you know I do think it is uh as we've uh virtualized infrastructure uh and built you know just the software development life cycle we have today from uh cloud infrastructure like Azure and Amazon web services to GitHub and GitHub actions and you know all the incredible open source libraries um and then you have sort of the virtual infrastructure to help support your companies like you know ramp you know and rippling you end up where uh there's a certain like there's a very different level of focus now in starting a

是 Sanjeev 的错——Sanjeev 的错。我们爱他。不，开玩笑的。那真的是完全不同的年代了。我确实觉得，随着我们把基础设施虚拟化，把今天这套软件开发生命周期搭起来——从 Azure、AWS 这样的云基础设施，到 GitHub、GitHub Actions，再到那些了不起的开源库——然后你还有像 Ramp、Rippling 这样的虚拟基础设施来支撑公司运转，结果就是，现在开公司有一种非常不一样的专注度。

[18:16]

company. Uh it used to be you just had to do all these things to like build the scaffolding of your company and it was actually like a non-trivial amount of work. And the thing I find just so exciting about like the future of entrepreneurship is you there's just so little overhead now in terms of just like intellectual overhead. Uh, and I remember actually just how hard it was to get like credit cards for our employees. Like they had to all get a credit check. I'm like, are you kidding me? Like it's it's our bank account.

以前你得把所有这些事都做一遍，去搭建公司的脚手架，那其实是相当不小的一摊活儿。而我觉得创业这件事的未来里最让我兴奋的，就是现在的负担少了太多——尤其是脑力上的负担。我还记得当年光是给员工办信用卡有多难，他们每个人都得做信用审查。我当时想，开什么玩笑？这用的是我们公司的银行账户啊，

[18:43]

Like what are they? And you know, now you can, you know, print a virtual, you know, uh, credit card with, you know, things like ramp. It's so easy. Um, so I think we've just seen every aspect of starting a company that was about sort of like the operational sort of scaffolding infrastructure to do so. Uh, many startups that you all have funded are now doing that. And so as a consequence, you can focus on just purely what defines your company and what makes it different. My color commentary on that is I actually think one of the mistakes a lot of smaller companies make is spending their resources and time on things that aren't

他们到底在查什么？而现在，你用 Ramp 这种工具就能直接生成一张虚拟信用卡，太简单了。所以我觉得，过去开公司里那些属于运营层面、搭脚手架、铺基础设施的方方面面，如今很多你们投的创业公司都在帮你把它们搞定了。结果就是，你可以纯粹去专注于真正定义你公司、真正让你与众不同的东西。我对此的一点补充评论是：我其实觉得很多小公司犯的一个错误，就是把资源和时间花在那些

[19:20]

core to their business. Uh and it's a it's more than a hidden cost. I think like every ounce of time you're spending on something that fundamentally should be a commodity to your business is time you're not spending on like selling your product to more customers or like working on like true differentiation and it takes a lot of discipline because a lot of those things are like catnip you know it's like fun to work on that part of your business and I always like to say like you know what Clay who's somewhere around here we always talk about like what will make us happy a year from now and that's always our

对自己业务并不核心的事情上。这不只是一笔隐性成本——我觉得，你花在那些本质上应该是商品化（commodity）的东西上的每一分钟，都是你没有花在把产品卖给更多客户、或者打磨真正差异化上的时间。这需要很强的自律，因为那些事很多都像猫薄荷一样诱人——做那部分业务其实挺好玩的。我老爱说，Clay（就在这附近某处）和我总会聊：什么能让我们一年后感到开心？这一直是我们的

[19:51]

metric and then some you could take it to 10 years if you want to be like no one knows what the hell is going to happen in 10 years. By the way, if you don't know what's going to happen in 10 years, nobody knows what's going to happen in 10 years. And I do think it's very clarifying and I just think it's I it always surprises me how much I tend towards some of those things that are, you know, wholly undifferiated at this point. Okay. Well, you mentioned some of the companies that Sierra works with, right? And some of those are companies that have been around for a long time, right? They are established companies.

衡量标准。然后你也可以把它拉到十年——如果你想抬杠的话：没人知道十年后会发生什么鬼事。顺便说一句，如果你不知道十年后会怎样，那别人也都不知道十年后会怎样。我确实觉得这个问题能让人想得很清楚，而且总让我意外的是，我自己也常常会被那些到了今天已经完全没有差异化的事情吸引过去。好。那你刚提到了几家跟 Sierra 合作的公司，对吧？其中有些是已经存在很久的公司，是那种根基很深的老牌公司。

[20:18]

What does the T and ADT stand for? Telegraph. Telegraph. Okay. So, how do establish companies? It's cool, by the way. Home security over telegraph. I actually imagine someone coming to like the Sequoia of that day. It's like home security over telegraph. It's like YouTube for cats, you know, and like and uh and it and they built that business. It really was. And now they're actually doing home security with AI agents. What a cool privilege. I think it's awesome.

ADT 里那个 T 代表什么？Telegraph（电报）。电报。好。那这些老牌公司——顺便说，这其实挺酷的，用电报做家庭安防。我都能想象当年有人跑去找那个时代的 Sequoia，说我要做「电报版家庭安防」，就跟说「给猫看的 YouTube」一样，然后他们真把这门生意做起来了。真的就是这样。而现在他们居然在用 AI agent 做家庭安防。这是多酷的一种荣幸啊。我觉得太棒了。

[20:43]

And you know, it's a better acronym than AD AI, you know. Um Okay. So, but what do they have to do? What are the ones that are going to succeed? what are they going to do to in a world of changing technology and what are the ones that are not going to succeed? What is holding them back? So, and how can this group help all of those established companies succeed? Yeah. So, I think it's a really interesting opportunity for uh companies in competitive markets like home security um or a company like SiriusXM that's got incredible unique original content, but they have a lot of cord cutting and a lot of competition around them. You have

而且你看，这个缩写也比「AD-AI」顺口多了。好。那么，他们得做些什么？哪些公司会成功？为了在这个技术不断变化的世界里成功，他们要做什么？哪些公司不会成功？是什么在拖它们后腿？在座这群人又能怎么帮所有这些老牌公司成功？好。我觉得，对于身处家庭安防这种竞争激烈市场的公司，或者像 SiriusXM 这样——它有非常独特的原创内容，但面临大量用户「剪线」（弃用有线、转向流媒体）和激烈竞争——这其实是个很有意思的机会。你

[21:21]

this technology and AI that drives incredible amounts of productivity in departments that had previously been incredibly underserved uh by software. uh everything from you know largecale contact centers to we talked about things like legal departments large operations departments um and as a consequence companies that lean into this can really change the cost structure of their business um uh if you think about I mean there's companies that have like 20,000 person contact centers and you know this costs hundreds of millions of dollars every single year and if you can take all of that opex and

现在有了 AI 这项技术，它能在那些过去被软件严重忽视的部门里带来惊人的生产力提升——从大规模的客服中心，到我们提到过的法务部门、大型运营部门等等。结果就是，那些愿意全力拥抱这件事的公司，真的能改变自己业务的成本结构。你想想看，有些公司光客服中心就有两万人，每年要花掉好几亿美元，如果你能把这一整块运营开支拿过来，

[21:57]

put it back into your business you can lower prices you can invest in growth and so it's really interesting I think it is one of those opportunities where these shovelready applications of AI even if it's indirect because they can drive such structural changes to the uh essentially essentially the unit economics of so many companies businesses I think it will change market share and so I think that it's a little bit like the birth of the internet you know and you know Walmart did very well uh since the birth of the browser even in the face of Amazon, there were a lot of retailers like Blockbuster who did

把省下来的钱投回业务里，你可以降价，也可以投入增长。所以我觉得这真的特别有意思。这是那种现成可用的 AI 应用带来的机会——哪怕影响是间接的，因为它们能给那么多公司的 unit economics 带来结构性的改变，我觉得它会重新洗牌市场份额。所以在我看来，这有点像互联网刚诞生那会儿。你想想，自从浏览器问世以来，Walmart 即便面对 Amazon 也活得很好，但也有很多像 Blockbuster 那样的零售商

[22:34]

not. And so I think uh because most executives running these companies now lived through the web browser era, saw the birth of the smartphone uh and saw the growth of things like Instacart and Door Dash and others. They see this. They're not and they're they're modeling what is the future of our industry going to look like. And you know, you mentioned enterprise sales. I think the key thing about selling anything is speaking the language of your customer and having empathy. Um, I think one of the things I see a lot with entrepreneurs is they'll go into a meeting and sell their product. Um, if

就没撑下来。所以我觉得，因为现在掌舵这些公司的大多数高管都经历过浏览器时代，见证过智能手机的诞生，也看着 Instacart、DoorDash 这些公司一路成长起来，他们看得懂这一切。他们不是傻子，他们正在推演自己这个行业的未来会长成什么样。然后你刚提到 enterprise 销售，我觉得卖任何东西最关键的一点，就是说客户听得懂的话、对客户有同理心。我经常看到一些创业者，走进会议室就开始推销自己的产品。但是

[23:12]

you ever see a really good salesperson, the first thing you'll see them do is ask a lot of questions. But then the next thing you need to do is actually listen and understand what they're saying and then reflect the value that your technology provides through the problems that they just described. And I actually think that actually a lot of the technologies in AI will immediately benefit a lot of these businesses, but they're being pitched by 20 AI vendors every single day. And they all sound the same. They literally all sound the same.

如果你见过真正厉害的销售，你会发现他们做的第一件事是问一大堆问题。但接下来你真正要做的，是认真听、是搞懂对方在说什么，然后把你的技术能提供的价值，映射到他们刚刚描述的那些问题上去。我其实觉得，AI 里有很多技术能马上给这些企业带来好处，但问题是他们每天都被二十家 AI 厂商轮番推销，而且这些厂商听起来全都一个样，真的是一模一样。

[23:41]

you could see like the you know first call deck of startups that do completely different things and there's like the same words on the slides and so this is the I think the really opportunity right now is how do you actually stand out in such a saturated market I actually Clay and I talk a lot about this because we both started web design companies in high school you know like during the dot bubble I went door todoor flower shops he did more like graphical things for I think he made a little more money than you did in high school yeah I it's he did his successful. He also got sued by the California Raisins, but that's a whole

你会看到那些做着完全不同业务的创业公司，他们的第一次路演 PPT 上写的词几乎都是一样的。所以我觉得现在真正的机会在于：在这么饱和的市场里，你到底怎么才能脱颖而出。其实我和 Clay 经常聊这个，因为我俩高中时都开过网页设计公司，就在那个互联网泡沫时期。我当时挨家挨户去给花店跑业务，他做的更偏图形设计那一块，我觉得他高中赚的钱比你多一点。是的，他确实做得很成功。他还被 California Raisins 给告了，不过那又是

[24:14]

another story. Um, so true story, his name's Clay. He named his company Claymation. Bad look. Um, sorry, Clay. I didn't know this a public. It's public now. Um, love you, man. Uh, and this guy doesn't know anything about the law. Yeah. uh we we talk a lot about the experience that you we were young and I not in the center of it during the dotcom bubble where you didn't just have Google you had alt vista all the web uh you had intome which was a B2B search player that was uh quite we respected quite a bit uh you had amazon.com and you had buy.com and half.com you had eBay and you know you had uh PayPal and all the other folks

另一个故事了。这是真事，他叫 Clay，结果他把公司命名为 Claymation（黏土动画）。这名字起得不太妙。抱歉啊 Clay，我之前不知道这事能公开说，现在反正都公开了。爱你，老兄。而且这哥们儿压根不懂法律。是的。我们经常聊起那段经历——我们当时还年轻，没在 dotcom 泡沫的最中心，那时候不光有 Google，还有 AltaVista、AllTheWeb；还有 InfoSpace，一个挺受认可的 B2B 搜索玩家；还有 amazon.com、buy.com、half.com，还有 eBay，还有 PayPal 以及其他一堆

[24:57]

pursuing payments I think now like victors write the history book. So you you look at all these magnificent coming out of that era and you forget that it was like a fight to the death. Um the idea that hey maybe we should search the internet was not exactly novel. Um the but the actual like detailed execution of those ideas was paramount. And I think you know going back to like how do you help companies like ADT succeed? the first thing you need to do is like show up as a partner to them and help them solve their acute business problems. And that's a uncomfortable sort of conversation because I think a lot of

做支付的公司。我觉得现在是胜利者在写历史书。所以你回头看那个时代跑出来的这些了不起的公司，会忘了当年那其实是一场你死我活的厮杀。「嘿，也许我们应该去搜索整个互联网」这个想法本身并不算多新颖，真正关键的是这些想法在执行细节上做到了极致。回到「你要怎么帮 ADT 这样的公司成功」这个问题，你要做的第一件事，就是以合作伙伴的姿态出现，帮他们解决最棘手的业务难题。而这种对话其实挺让人不舒服的，因为我觉得很多人

[25:35]

people don't feel equipped to do it. Uh you know, one trick that I would just recommend is before you meet a customer, do deep research on them. And by that I mean like the actual feature deep research and chat dbt. Um you know, show up educated in these conversations. And I think you find the more you spend researching your customer and understanding their needs and less time you thinking about like the next feature going to launch, the better that conversation will go. And I think you'll end up finding that, you know, sweet spot between like where is the intersection of what you do differently

并不觉得自己有能力去做这件事。我想推荐一个小窍门：在见客户之前，先对他们做一次 deep research。我说的就是字面意义上的 deep research 功能，就在 ChatGPT 里。带着功课去谈，做个有备而来的人。我发现，你花在研究客户、理解他们需求上的时间越多，花在琢磨自己下一个要发的功能上的时间越少，那场对话的效果就越好。最后你会找到那个甜蜜点——也就是「你与众不同的地方」

[26:05]

and actual business value. That's awesome. Let's go to audience questions. I just realized that I've taken up a lot of Brett's time. So, uh, please go ahead. Thank you. Um, hi, I'm Robin. What are your learnings in building branded AI agents for specific verticals and how is it different to building horizontally? Uh, I really believe in verticals in AI. Uh I think that if you think about what the job proverbally of an AI agent is uh for a telecommunications company and you think about what's important to them uh versus uh say a commercial bank uh versus say a residential uh bank uh versus an insurance company who might be

和「真实的商业价值」这两者的交集。太棒了。我们来听听观众提问吧。我刚意识到我占用了 Bret 太多时间了，所以请提问吧。谢谢。嗨，我叫 Robin。你们在为特定垂直行业打造品牌化 AI agent 方面有哪些心得？这和做横向（horizontal）的产品有什么不同？我真的非常看好 AI 里的垂直方向。你想想看，一个电信公司的 AI agent 它的职责本质上是什么、他们最在意的是什么，再对比一下，比如说一家商业银行、一家零售银行，或者一家保险公司——保险公司可能

[26:59]

concerned about claims processing versus a health insurance company that's uh you know answering things like explanation of benefits. every single one of those applications is actually quite specialized and you know right now uh I think that the companies that can actually provide value quickly around these core workflows of these businesses that are actually valuable to them uh actually sort of when I talk about sort of like coming into those conversations and actually providing value really quickly just have a leg up. Um I am fairly skeptical of horizontal AI platforms.

关心的是理赔处理，而健康保险公司则要处理像「福利说明」这类事情。这每一个应用其实都相当专业化。我觉得现在那些能围绕这些企业真正有价值的核心工作流、快速交付价值的公司——也就是我刚说的，能走进那种对话、迅速拿出实打实价值的公司——它们就是占了上风。我对横向的 AI 平台是相当怀疑的。

[27:34]

Um, it doesn't mean they won't succeed by the way. There's always an exception to every rule. But, uh, you know, things like lang chain I think probably will succeed. They're open source and you know, you end up sort of you need to sort of develop a horizontal business in a different way. But it just, you know, pure play enterprise business. I think it's very hard to go in with a horizontal platform because there's just lots of good enough. Um, there's lots of homegrown solutions. There's lots of not invented here. Um, and you know, having a slightly better mouse trap is rarely uh, you know, doesn't pass the like

不过这并不代表它们就一定不会成功，凡事总有例外。比如像 LangChain，我觉得它大概率会成，它是开源的，你最终得用一种不同的方式去把横向业务做起来。但如果是纯粹的 enterprise 业务，我觉得拿一个横向平台进去会非常难，因为市面上已经有太多「够用就行」的东西了，有太多自研方案，还有太多「非我所创」的抵触心理。手里只是握着一个稍微好一点的捕鼠器，这种东西很少能过得了「维生素还是止痛药」

[28:07]

vitamin painkiller uh, threshold generally speaking. So, my rule of thumb is like the closer you get to uh, solving a problem for a company, the more successful your your business will be, assuming you're talking about an enterprise business. Um, and you know, the other thing that I I really learned uh from Mark Ben off at Salesforce is like if you have one successful customer, make it two, and if you have two, make it 10. And if you have 10, make it 100. Um, I do think there's a certain amount of tactical excellence that comes in when you're building a B2B software business. And ivory tower cleverness rarely wins. It doesn't mean

的那道门槛——一般来说都过不了。所以我的经验法则是：你越接近于替一家公司解决一个具体问题，你的业务就会越成功，前提是我们说的是 enterprise 业务。还有一件事，是我从 Salesforce 的 Marc Benioff 那儿学到的：如果你有一个成功的客户，就想办法变成两个；有两个，就变成十个；有十个，就变成一百个。我确实觉得，做 B2B 软件生意是需要一定的战术执行力的，那种空中楼阁式的小聪明很少能赢。这并不是说

[28:44]

you shouldn't have a long-term strategy by the way, but uh there's a certain if you focus on making your customers successful, that's where the truth is and where your like product should go. Um sometimes you can be led astray by that mechanism. You like the the first one was not representative. That's rarely the concern though. Usually it's uh being very very clever and sort of ivory tower with your strategy. So that's why I love verticals.

你就不该有长期战略，当然要有，但有一点是确定的：如果你专注于让你的客户成功，真相就在那里，你的产品也应该往那个方向走。当然，这套机制有时也可能把你带偏，比如你的第一个客户其实并不具代表性，但这种情况其实很少见。真正常见的问题，反而是你的战略太「聪明」、太空中楼阁了。所以这就是我为什么这么爱垂直方向的原因。

[29:10]

Uh maybe one more. Hey Brad. Um I was really interested in the question around how you price uh AI agents. Specifically, have you seen customers being more open to you taking a cut of cost savings or of incremental revenue that you've created? And what are the main kind of push backs that you that you're hearing from customers? Because impact pricing has been historically quite hard. Um so wondering what you're seeing there. Yeah. Um we see interest in both. Um, you know, if you're owned by a private equity firm and have a big bunch of debt, you probably say the word Ebatel like 20 times a day. It's like

再来一个问题吧。嘿，Bret。我特别想问关于你们怎么给 AI agent 定价的问题。具体来说，你有没有见到客户更愿意让你从他们省下的成本、或者你帮他们创造的增量收入里分一杯羹？你听到客户最主要的抗拒点又是什么？因为按成效（impact）定价历来都挺难做的，所以我想知道你们那边看到的情况。是的，这两种我们都看到有人感兴趣。你想，如果你被一家私募股权公司控股、背着一大笔债，那你大概一天得把「EBITDA」这个词念上二十遍，那就是

[29:44]

what you do at private equity owned companies. And you probably care a lot about cost savings. Um, though, and if you're That's basically all you do. You just walk around. Yeah. Yeah. He actually really simplified it. He got it. Yeah. It's like, see, I I know finance too. Yeah. Yeah. Yeah. you have not ever let an oified version of yourself slow you down you know please and uh in contrast you know if you're in a very competitive market like if you talk to any CEO they generally care more about growth than they do cost savings and in fact I think one of the counterintuitive things there's lots of

私募控股公司里干的活儿。你大概会非常在意成本节省。不过——如果你是……那基本上就是你每天干的全部事，你就到处走来走去念这个。是的，是的，他其实把这事概括得很到位，他抓住了精髓。是的，看吧，我也懂金融的。是的是的。你从来没让一个「官方版的自己」拖慢过你的节奏，对吧，拜托。与之相对，如果你处在一个竞争非常激烈的市场，你去问任何一个 CEO，他们通常更在意增长，而不是成本节省。事实上，我觉得有一件挺反直觉的事，外面有很多

[30:19]

talk about with models becoming cheaper will demand for infrastructure go down no in fact demand will go up just because new use cases will emerge I think the same will be true of most cost savings in AI I think most sophisticated companies will recoup those soft cost savings and reinvest in growth. And so it won't actually be cost savings necessarily. It will actually uh flow to other parts of the business just because that's what capitalism does, right? It's like you know what saving a lot of money for what reason? Like you want to grow, you want to beat your competitors. Um so sort of first principles and and our

在讨论：随着模型越来越便宜，对基础设施的需求会不会下降？不会，事实上需求反而会上升，因为会有新的用例冒出来。我觉得 AI 里大多数成本节省也是同样的道理。我认为大多数成熟的公司会把这些省下来的成本重新收回去，再投到增长上。所以它最终其实并不算是「成本节省」，这笔钱反而会流向业务的其他部分，因为资本主义本来就是这么运作的，对吧。就像——你省下一大笔钱是为了什么？你是想增长，你是想打败你的竞争对手。所以从第一性原理出发，再加上我们的

[30:51]

lived experience is that you know topline matters more. There's just no doubt about it. And if you can drive a measurable topline outcome that's always more valuable. But cost savings do matter because fundamentally if you you know look at sort of productivity growth in the world over the past 30 years it hasn't been as uh dramatic as I think many people thought it would be after the '9s where we saw a huge uh spike in productivity growth. So fundamentally just like in the economy we're driving efficiency productivity in the economy and so the answer is definitely both.

亲身经验来看，营收（topline）这件事更重要，这一点毫无疑问。如果你能驱动一个可衡量的营收成效，那永远是更有价值的。但成本节省确实也重要，因为说到底，你回头看过去三十年全球的生产率增长，其实并没有像很多人在九十年代之后预期的那么夸张——那会儿我们见过一次生产率增长的巨大飙升。所以从根本上说，就跟整个经济一样，我们是在为经济驱动效率和生产率。所以答案绝对是「两者都要」。

[31:21]

The interesting thing about cost savings is I assume those will be compressed over time. You know, right now most of your AI startups are being compared to the labor costs. 10 years from now when it's all AI agents, you're going to be compared to other AI agent. So the other thing about cost savings is it's a temporary drug, you know, because the costs of actually all these things are going to change because of the presence of of technology. Um, so it's a little bit like, you know, your mobile phone plans or actually always think about as like the computers we used to buy growing up in the the 80s and 90s. they

关于成本节省，有个有意思的点：我估计这部分的价值会随着时间被不断压缩。你看，现在大多数 AI 创业公司是拿来跟人力成本作比较的。但十年后，当一切都是 AI agent 的时候，你就要去跟其他 AI agent 比了。所以成本节省的另一面是，它是一种短效药，因为这些东西的成本本身会因为技术的存在而不断变化。这有点像你的手机套餐，或者我其实更愿意拿小时候、就是八九十年代我们买的那些电脑来打比方——它们

[31:55]

would get cheaper and better every single time. And that's sort of where we are in this AI market. So our view is that like figure out what the outcomes are that your customers want and it really depends on the application honestly. Um even how you price it is interesting. One of my favorite stories, I don't know if this is actually true, but I uh I heard it and even if it's not true, it changed I think which was uh when LinkedIn was at one point um trying to grow the recruiting business, they started to do they started out with some usagebased model, but it turns out HR departments don't have a lot of uh

每隔一段时间就变得更便宜、更好。我们现在在 AI 市场里大概就处在这么个位置。所以我们的看法是：搞清楚你的客户到底想要什么样的成效，而且老实说这真的很取决于具体应用。甚至连你怎么定价都很有意思。我最喜欢的故事之一——我不确定这事是不是真的，但我听来的，而且就算它不是真的，我觉得它也改变了——就是当年 LinkedIn 有一阵子想把招聘业务做大，他们一开始用的是某种按用量计费的模式，但结果发现 HR 部门并没有太多

[32:27]

license to spend like you know in a variable amount. So making it a subscription model just help with the procurement process. I love that example because it's a really nuanced and insightful go to market point which is that department HR is generally a cost center in most companies unlike marketing departments which have like you know money flowing from the the ceiling you know this is a department that has to go through a pretty rigorous procurement department so actually going to a subscription model was necessary to actually sell a product into that department it's that type of thing I think is where it's like is it cost

那种可以随用量浮动去花钱的额度。所以把它改成 subscription 模式，反而帮他们理顺了采购流程。我特别喜欢这个例子，因为它是一个非常微妙、又很有洞察力的市场打法要点——也就是说，HR 在大多数公司里通常是一个成本中心，不像市场部那样，钱就跟从天花板上往下掉似的。HR 是那种必须走一套相当严格采购流程的部门，所以改成 subscription 模式，反而成了把产品卖进这个部门的必要条件。我觉得正是这类事情才是关键所在，就像——到底是成本

[33:01]

savings or topline yes both also like who are you selling to? How do they buy? Uh who are the gate uh gatekeepers to approving how they spend? That's how you grow like an enterprise software business. You really need to understand who your decision makers are, who's approving it, and actually like how do they budget? Um and I have a lot of counterintuitive things, but it's like sometimes easier for companies to prepay for things than to pay on demand.

节省，还是营收？答案是两者都是。同时也要看：你卖的对象是谁？他们是怎么采购的？谁是审批他们怎么花钱的那个把关人？这才是你把一门 enterprise 软件生意做大的方式。你真的得搞清楚你的决策者是谁、谁来拍板、以及他们到底是怎么做预算的。我这里有一堆反直觉的心得，但其中一条是：有时候对公司来说，预付费反而比按需付费更容易些。

[33:26]

Counterintuitive, but just I think that's like you really need to get into the details of your buying cycle to answer that in a in a accurate way. I'm going to wrap us up. Brett, thank you. Uh they say you should do business with people you like, trust, and admire, and you are that for me. So, I really appreciate you doing this, and all of us do as well. Thank you. Thank you.

这听起来有点反直觉，但我觉得你真的得深入去琢磨自己的采购周期，才能准确地回答这个问题。差不多该收尾了。Bret，谢谢你。人们常说，做生意要找你喜欢、信任、又敬佩的人合作，而你对我来说就是这样的人。所以真的很感谢你来参加，我们大家也都很感激。谢谢你。谢谢。