AI 代理：新互联网浪潮

Bret，感谢你接受采访。

Bret, thank you for doing this.

谢谢邀请。

Thanks for having me.

很高兴我们敲定了时间，时机正好。你们最近宣布了新的一轮融资，恭喜。

I'm glad we got it on the calendar. It's a good time to do it. You guys recently announced a new round. Congratulations.

谢谢，我很感激。这是旅程中的一个里程碑，很好地证明了你们至今的成果和未来的方向。

Thank you. I appreciate it. It's a nice milestone on the journey, a good testament to what you guys have built to date and where you're headed over time.

是的，这是一个里程碑，但这么说更准确。融资只是为到达目标添加燃料。对于我和联合创始人 Clay 这样的创业者来说，我们想打造一家持久稳固的公司。所以这只是一个里程碑，但也是个反思的好时机。我们非常自豪。我认为我们在所处的领域——即面向客户体验和客户服务的 AI 智能体——是明确的领导者。但我常说这次 AI 浪潮与最初的互联网浪潮何其相似。Alta Vista 是第一个，但 Google 才是书写历史的那一个。所以我们需要从现在开始持续多年完美执行，那时我才会满意。但我很兴奋。

Yeah, it's a milestone, but that's the way to put it. Raising financing is just adding fuel to get to the place you want to go. For an entrepreneur like me and my co-founder Clay, we want to create an enduring, durable company. So it's just a milestone, but it's a good time to reflect. We're really proud. I think we're the clear leader in the space we operate, which is AI agents for customer experience and customer service. But I always talk about how similar this AI wave is to the original internet wave. Alta Vista was first, and Google is the one who writes the history book. So we need multiple years of sustained impeccable execution from here, and then I'll be happy. But I'm excited.

有个 Norm Macdonald 的笑话：历史上好人总是赢，因为你可以定义成功的标准。你怎么看互联网和 AI 的相似之处？哪些地方相似，哪些地方不同？

There's the Norm Macdonald joke that the good guys always won in history because you get to define the terms of success. What do you think about the parallels between the internet and AI? Where do the parallels exist and what breaks down?

一个非常相似的点是，互联网时代有一些非常明显的赌注，比如搜索和电商是最突出的两个，还有支付。你不需要对互联网有多深的了解，也能想到人们数字化购物可能有用。问题是谁会拥有这个市场。亚马逊和 buy.com 的策略截然不同，甚至最初销售的产品组合也不同。亚马逊显然在策略和执行上都好得多。他们有很多细节做对了。看看搜索市场。我提到了 Alta Vista 和 Google，但我们与之竞争过的一家技术公司 Inktomi 也令人印象深刻。他们有一个非常优秀的工程团队，我可以告诉你为什么 PageRank 更好，但他们团队很扎实。然而，他们的商业模式是 B2B。他们基本上把搜索引擎授权给门户网站，这就错失了创建 AdWords 的机会，而 AdWords 后来成为有史以来最伟大的商业模式。所以有太多细节决定了你是成为 Google、Alta Vista 还是 Inktomi。回到你的问题，AI 市场有趣的一点是，有几个领域显然会受到 AI 影响：软件工程、客户服务、内容营销、视觉特效行业，可能还有法律行业等。因此，这不像是我有了一个绝妙的主意。如果我告诉你“天哪，AI 用于客户服务”，这个概念是显而易见的。问题在于，你是否有正确的产品？是否有正确的市场进入模式？这就是我提到 Inktomi 和 Google 对比的原因。你是 B2B 还是 B2C？包装是什么？什么形态会成为主导？所以这是一个竞争极其激烈的时期，就像我对互联网时代的记忆一样。这非常有趣。所以你有非常清晰的市场，竞争非常激烈。这与其他市场不同。手机出现时，像网约车这样的类别并不是不言自明的市场。然后有了一些伟大的洞察，才创造了 Uber 和 Lyft 这样的公司。现在，许多最大的市场已经为人所知，因此，我们公司的价值观之一是“竞争强度”，这很不寻常。这个价值观的第一句话是“我们知道我们并非理所当然地成功”，我认为这是在这个时代成功的一个非常重要的部分。

One thing that is quite similar is there were some very obvious bets in the internet, like search and e-commerce being two of the most prominent, and payments. You don't need to be that savvy about the internet to think it might be useful for people to buy things digitally. The question was who will own that market. Amazon and buy.com had very different strategies, even different portfolios of products they initially sold. Amazon clearly had a much better strategy and execution. There are a lot of details they got right. Look at the search market. I mentioned Alta Vista and Google, but one of the more impressive technical companies we competed against was Inktomi. They had a really good engineering team, and I could tell you why PageRank was better, but they had a solid team. However, they had a B2B business model. They essentially licensed their search engine to portals, and that left out the opportunity to create AdWords, which turned out to be the greatest business model of all time. So there are so many details that dictate whether you become Google or Alta Vista or Inktomi. Going back to your question, what's interesting about the AI market is there are a few areas that are obviously going to be impacted by AI: software engineering, customer service, content marketing, visual effects industry, and probably others like the legal industry. As a consequence, it's not like I had this great idea. If I told you, 'Oh my gosh, AI for customer service,' the concept is obvious. The question is, do you have the right product? Do you have the right go-to-market model? That's why I mentioned the Inktomi vs. Google thing. Are you B2B? Are you B2C? What is the packaging? What is the form factor that will become dominant? So it's an intensely competitive time, just like my recollection of the dotcom era. That's really interesting. So you have very clear markets with very intense competition. That's different from other markets. When the mobile phone came out, some categories like ride-sharing weren't self-evident markets. Then there were a couple of great insights, and it created the Ubers and Lyfts of the world. Right now, many of the biggest markets are already known, and as a consequence, one of our company values is actually 'competitive intensity,' which is unusual. The first line in that value is 'we know we're not entitled to our success,' and I think it's a really important part of being successful in this era.

这种相似性在什么地方失效了？

Where does the parallel break down?

我认为它改变了软件公司的格局。我喜欢 Harvey 这个例子，这是一家我非常欣赏的公司。我想不出一家伟大的法律科技公司。我相信有一些，但如果你看公开市场排名前十的企业软件公司，没有一家在法律科技领域。有 ERP 系统、CRM 系统等等。它根本不是关键类别之一。部分原因是向律师销售生产力提升工具的总可寻址市场并不大。但现在有了 Harvey，如果你真的在做工作，做反垄断审查，突然总可寻址市场看起来巨大，因为法律咨询和法律劳动的可寻址市场实际上相当大。这对我来说非常有趣，因为我认为传统上对软件可寻址市场的认知已经被颠覆了。智能体不仅仅是提高人们的生产力，而是实际在做一项工作。因此，你评估软件价值的方式开始偏离传统的软件生产力指标。想想一个 Sierra 智能体，它实际上为你完成了一笔销售。你评估它的方式甚至与 AI 或软件无关。你评估它的方式基本上是那笔销售的利润率是多少？你付给一个人做那笔销售的佣金是多少？所以我认为它极大地改变了传统的软件估值观念，无论是作为风险投资人还是经济学家。它确实改变了市场。我对此感到兴奋。我认为这对我们的行业来说将是一件非常积极的事情。

I think it's shifted the landscape of what is a software company. I love the example of Harvey, a company I really admire. I can't think of a great legal tech company. I'm sure there are a couple, but it wasn't like if you went through the top 10 enterprise software companies in public markets, there's not one in legal tech. There are ERP systems and CRM systems and all these others. It's just not one of the key categories. In part because the TAM for selling productivity enhancement to lawyers is not that big. But now with Harvey, if you're actually doing the work and doing the antitrust review, all of a sudden the total addressable market looks huge because the addressable market of legal advice and legal labor is actually quite large. That's really interesting to me because I think the traditional perception of where there are addressable markets in software has been upended. Agents aren't simply productivity enhancements for people but actually doing a job. As a consequence, how you evaluate the value of a piece of software starts to move away from traditional software productivity metrics. Think about a Sierra agent that actually makes a sale for you. The way you would value that is not really even related to AI or software. The way you'd value it is basically what are the margins on that sale? What would be the commission you'd pay a person to make that sale? So I think it has dramatically shifted the traditional view of how to value software, whether as a venture capitalist or an economist. It's really changed the markets. I'm excited for that. I think it's going to be a really positive thing for our industry.

另一个好处是，我可以把你说的 Harvey、Slice 和 Lora 全部覆盖掉，这样我们就不用打广告了。但这是个很好的例子。我好奇一件事：在某种程度上，基于结果的定价具有如此可证明的 ROI，人们愿意为之支付什么非常明确。但另一方面，你在某些方面也受制于替代方案。

The other great thing is I can dub over you saying Harvey and Slice and Lora for the totality of that, so we don't have to give any plugs. But it's a great example. I'm curious about one thing: to some extent, outcome-based pricing has such demonstrable ROI, and it's very clear what people are willing to pay for it. But also you're somewhat beholden to alternatives in some ways.

所以我在想，我从 Zoom 获得的价值是什么。如果世界上只有 Zoom 这一款产品，我们也许能让 Redpoint 支付 1000 万美元，因为它对我们的日常影响太大了。但他们没法收那么高，因为有 Teams、Google Meet 和其他竞品。那么，当你考虑为你的业务或 Harvey 和 Lora 这类公司做基于结果的定价时，有没有什么框架？或者你觉得，当存在这些替代品时，价格会如何演变？你认为它最终会侵蚀早期能获得的 ROI 定价吗？

And so like I think about what value I get from Zoom and if Zoom was the only thing that existed in the world, I don't know, we could probably get Redpoint to pay $10 million for it or something, right? Just cuz it's like that impactful to our day-to-day, but they don't get to charge that because there's Teams and there's Google Meet and there's other stuff like that. And so I guess as you think about outcome based pricing for your business or for Harvey and Lora like doing work in that way, is there any framework or like how do you think price plays out in some ways when there are these alternatives that can exist? Do you think that it ends up eroding some of the, you know, the ROI pricing that you can get in the early days?

我对此有稍微不同的看法，但我也尽量直接回答你的问题。

I have a slightly different way of thinking about it, but I'll try to answer your question directly, too.

你可以拒绝我的问题。

You could reject my question.

不，这问题其实挺烦人的。换个角度回答吧。我认为 Zoom、Slack 和 Teams 这类工具出现价格压缩的部分原因是，你从横向生产力工具中获得的价值很难衡量。想象一下运营一家 12 万人的全球公司，你为 Zoom 或 Slack 这类产品按席位付费。有趣的是，你为最资深的研发工程师和新入职的毕业生支付相同的席位费，而且你选的是公司里最不具战略性的部门。因此，我认为对于横向软件——无论是生产力软件还是通信软件——定价最终会变得有些商品化，尽管也有少数例外公司能收取溢价。相比之下，如果你看面向部门的企业软件市场，比如 ServiceNow（ITSM）、Salesforce（CRM）或 SAP（ERP），这些公司每个席位产生的价值传统上要大得多，通常比 Zoom 这类软件高一个数量级以上，尽管使用人数少得多，但它更贴近业务价值。比如平衡公司账目、在财报电话会前审计财务的价值，或者一笔销售的价值。因此，你销售的业务价值更可衡量，更贴近那个业务价值。对于基于结果的定价，我给出的类比是从展示广告转向 CPC 广告。一种我不认同的观点是，基于结果的定价相当于把钱留在桌上。这就好比说现代 CPC 广告拍卖是把展示量留在桌上。但现在已经没人这么想了，因为历史已经证明，价值已向 CPC 和每次转化成本倾斜。我认为在数字经济中，你越接近可衡量、可问责的结果，你的平台就能获得越多的价值。回到你的问题：竞争会导致价格压缩吗？很可能。但总体而言，你越接近真正有价值的业务结果，你的平台就越会根据那个业务结果的价值来定价，而不是与其他技术比较。有趣的是，我不确定 Zoom 的基于结果定价会是什么样，因为你必须为每一次视频通话赋予价值。我敢肯定，有些通话在达成大交易时非常重要，有些则完全无关紧要，这并不容易。在 Sierra 的业务中，我们帮助构建用于客户体验的 AI 智能体。你知道呼叫中心的每次联系成本，以及 AI 智能体可以带来的成本节约。你知道新产品销售的价值，如果你的 AI 智能体帮助促成了那笔销售，就像你付给销售人员的佣金一样，你知道它对智能体有多大的价值。由于非常贴近这个价值，我认为这是一种很自然的收费方式。对于公司来说，这意味着他们可以按业务获得的价值来建模，而不是按技术的成本。

No, it's like really obnoxious. Answer a different question. Um, I think part of the reason that there's price compression for tools like Zoom or Slack and Teams is in part because the value you get from a sort of horizontal productivity tool is very hard to measure. If you just think about running a 120,000 person company that's a global company, you're paying per seat for something like a Zoom or a Slack or something. It's sort of funny because you're paying the same value per seat for like the most sophisticated research and development engineer and like the new grad and you're sure pick the most least strategic department of whatever that company does. And as a consequence, I think when you're thinking of horizontal software, whether it's productivity software, communication software, you end up with pricing that is somewhat commoditized and there are some rare exceptions, you know, where companies are able to charge a premium. In contrast though, if you look at the enterprise software market that are oriented towards departments, say ServiceNow for ITSM or Salesforce for CRM or SAP for ERP systems, you know, the value that those companies derive per seat, you know, for their application is traditionally much larger, usually more than an order of magnitude of software like Zoom, even though many fewer people use it, but it's closer to a business value. You know, the value of balancing your company's ledger and auditing your financials before an earnings call. You know, the value of a sale. And as a consequence, you know, the business value you're selling is more measurable. It's closer to that business value. The analogy I would give for outcomes-based is we're going from impression ads to CPC ads. And one way of looking at outcomes based pricing that I don't agree with is you're sort of leaving money on the table. That would be like making a modern cost-per-click ads auction saying you're leaving impressions on the table. And that's not the way anyone thinks of it anymore just because history's played out and the value has accrued towards CPCs and cost-per-conversion now for modern ad networks. I think in the sort of digital economy, the closest you can get to a measurable accountable outcome, the more value will accrue to your platform. And so going back to your question, will competition cause price compression? Probably. But I think in general, you know, the closer you are to a really valuable business outcome, the more your platform will be valued relative to the value of that business outcome as opposed to being compared to another piece of technology. And so, you know, it's interesting if I'm not sure what outcomes based would be for Zoom because you'd have to ascribe a value to every single video call you have. And I'm sure some are quite important when you're closing a huge deal and some are totally trivial and that's just not easy. In Sierra's business, we help build AI agents for customer experience and you know your cost per contact in your call center and the cost savings that an AI agent could drive. You know the value of a new product sale and if your AI agent helps make that sale, just like how much you pay a salesperson for doing that, you know how valuable it is to your agent. And as a consequence of being really close to that value, I think it's a really natural way to charge for it and for companies it means they can model this not proportional to the cost of a technology but proportional to the value that they're getting as a business.

我认为会发生的事情——结合竞争和技术采用——是很多 AI 智能体现在正被拿来与人类同行比较，无论是劳动力成本还是效率，比如软件工程或客户服务领域。你可能知道，传统观点认为 AI 智能体将成为这个行业的主导力量。你可以想象，10 年后你会开始将智能体与其他智能体进行比较。这将导致各种差异。成本将不再与劳动力成本比较，而是与推理成本比较，但效率可能会不同。我认为人们经常忽略的是二阶效应。回到 Sierra 的业务。很多人想到 AI 智能体用于客户体验时，就会想到呼叫中心自动化。这没错。如果一通电话现在成本 20 美元，用 AI 只需 20 美分，那是一个节省运营开支的好机会。但想象一下，你经营一家大型电信公司，整个业务基于客户生命周期价值。你有一些订阅用户使用高端套餐 10 年，另一些使用低端套餐 1 年。你的业务实际上是客户获取成本和流失率的函数。现在，你的 20 美元电话变成了 20 美分。你只是要收回这些成本吗？还是会想，我可以与订阅用户进行多少次对话，从而实际提升他们的套餐等级？降低他们看到电视广告后转投其他移动运营商的可能性？然后你突然意识到，这比你从降低运营成本中获得的节省要重要得多。我认为这实际上会改变市场。所以从某种程度上说，你可以问：一阶或二阶效应会是价格压缩吗？我实际上认为影响会比那更深远——客户参与的方式将完全改变，你不再将其视为成本中心。当一通电话的价格开始接近一次页面浏览的价格时，你会进行更多的通话。

The thing I think will happen which is a mix competition and technology adoption is a lot of AI agents now are being compared to their human counterparts whether it's labor costs or effectiveness in a market like software engineering or customer service. An area that you know perhaps like you know conventional wisdom is AI agents will come to be dominant parts of this industry. You have to imagine that in 10 years you'll start comparing agents to other agents. And that'll lead to all sorts of differences. You know, the cost won't be comparing to labor cost. You'll be comparing it to inference costs, but the effectiveness will presumably be different. And I think the thing people often miss are second order effects. So, just going back to Sierra's business. I think a lot of people think about AI agents for customer experience and they think call center automation. And that's true. And you know, if a phone call costs $20 today and it cost 20 cents with AI, wow, that's a great opportunity to recoup operating expense savings. But imagine you run a big telecommunications company and your entire business is based on lifetime value. And you have some subscribers who are on a higher tier plan for 10 years or some subscribers on a lower tier plan for one year. Your business is really a function of customer acquisition costs and attrition. And now all of a sudden, your $20 phone call went to 20 cents. Are you just going to recoup those costs? Are you going to think, how many more conversations can I have with my subscribers and actually increase the size of the plan they're on? Reduce the likelihood that they see an ad on television and switch to another mobile phone provider. And all of a sudden, you realize, wow, that's a lot more important than the operating expense savings I might have gotten from reducing my BO costs. And I think it will actually change the market. And so in some ways you could say, will the first or second order effects be price compression? I actually think it will be much more dramatic than that, which is actually what you do with customer engagement will just shift entirely and you'll stop thinking of as a cost center. When the price of a phone call starts to approach the price of a page view, you're going to do a lot more of them.

因此，我认为这将彻底颠覆市场。特别是，我认为现在很难预测估值方式，但它会更接近业务成果，而不是技术成本。我想说的是——我不是要你猜测其他业务，因为你今天在这里经营着一家很棒的公司——但我脑子里一直在翻来覆去地比较互联网和移动互联网，AI 作为机会集更像哪个？结果集是更像互联网，即大型独立公司捕获的价值大于以某种方式利用技术的现有企业，还是更像移动互联网？我们可能都同意，移动互联网最大的受益者可能是 Google（通过 Android）、Apple（通过 App Store）和 Facebook。大部分创造的价值可能都以某种方式流向了现有企业。那么，当你思考哪里会有全新的机会或价值向量时，你是否认为互联网也是如此，大部分价值流向了 Amazon 和 Google？

And as a consequence, I think it's just going to really upend the markets. And in particular, I think the way you'll value it is very hard to predict right now. But I think it will go closer towards business outcomes than the cost of the technology. I guess wrapped in that, and I don't mean to ask you to speculate about other businesses because you have a great one that you're running here today, but there is this analogy that I've sort of flipped back and forth in my mind a little bit, which is internet versus mobile of what is AI as an opportunity set. Does the outcome set look more like the internet in that the value captured by big independent standalone companies is larger than that of existing businesses that have leveraged the technology in some way, shape or form versus mobile? I think we could probably both agree that the biggest beneficiaries of mobile were probably Google with Android, Apple with the app store, Facebook. Probably most of the value that was created was probably in some type of incumbent in some way. And I guess as you think about where there might be net new opportunities or where vectors of value are going to be created, do you think the same about the internet that most of the value accrued to Amazon and Google?

其实我还没算过这笔账。我不是在反驳你。我是说，它们是有史以来最有价值的公司之一，在标普 500 里排前五，但这很有意思。我记得《长尾理论》那本书，但它创造了一个不可思议的经济体。如果你看股市里前 100 家软件公司，前五名是 Meta、Google、Amazon、Apple、Microsoft，但接下来的 40 家是 SaaS 公司，其中很多除非你在我们这个行业，否则大多数人从未听说过。长尾的总和是多少？我看看像 Shopify 这样了不起的公司。Shopify 是受益者，但它们在赋能创业者。我想起 Stripe 的 Patrick 谈到提高互联网的 GDP。所以我认为互联网尤其是一个极长的长尾，它确实以戏剧性的方式改变了经济，改变了分销机制。你看，Meta、TikTok 没有互联网就不会存在。你是归功于互联网，还是把一切都归因于它？所以我很乐意回答你的问题，但我想说，我实际上不知道该怎么想，因为现在每家公司都是数字公司。

As opposed to... it's actually I just haven't done the math on it. I'm not arguing with you actually. I mean, they're two of the most valuable companies of all time, and you know, in the top five of the S&P 500 and all that, but it's so interesting. And I remember the book The Long Tail and all that, but it's created an incredible economy. So if you took the top 100 software companies in the stock market, the top five are Meta, Google, Amazon, Apple, Microsoft, but the next 40 are SaaS companies, many of which most people have never heard of unless you're in our business. What is the sum of the long tail? And I look at incredible businesses like Shopify as an example. Shopify is a beneficiary of that, but they're empowering entrepreneurs on top of it. I think of Patrick at Stripe talking about increasing the GDP of the internet. So I think the internet in particular is an extremely long tail, and I think it's really changed the economy in dramatic ways and changed distribution mechanics. You look at certainly Meta, TikTok wouldn't exist without the internet. Do you give credit to the internet or do you ascribe it all there? So I'm happy to answer your question, but just to say I don't know actually how to think about that because every company is a digital company at this point.

这很有意思。这是一个有趣的观点，也许我需要重新组织我的问题，因为我认为你可能是对的，平台本身之上创造的总价值使那个相形见绌。

It's interesting. That's an interesting point, and maybe I need to reframe how I asked the question because I think you might be right that the totality of the value created on top of the platform itself dwarfs that.

可能确实如此。我很希望有比我更聪明的人来做这个分析。我认为是的，但我不知道。这也很难定义：一家拥有 6 万名软件工程师的银行是软件公司还是银行？还有那家 HVAC 公司，现在能以更有意义的方式接触客户，收入从 100 万美元增长到 2000 万美元。你把功劳归给谁？这很有趣。

It may actually. I'd be interested for someone smarter than me to do that analysis. I think it does, but I don't know that. It's also hard to define: is a bank with 60,000 software engineers a software company or a bank? And the HVAC company that can now reach customers in a more meaningful way, that grew from $1 million in revenue to $20 million in revenue. Who do you give credit to for that stuff? That's interesting.

但我想问的是，有一种……移动互联网的 Salesforce 在很多方面被证明就是 Salesforce。它某种程度上是延续的，至少在那个领域内，显然也发生了颠覆，比如 Uber 和出租车、Airbnb 和酒店等等。我想，当你考虑那些大型软件公司时，你提到了 ServiceNow 和 SAP，当然还有生态系统中的 Salesforce。你认为其中大部分会持续存在吗？如果我们挑出，比如说，10 个行业，以及那些横向或非横向的应用玩家？你认为大部分会存活下来，成为类似于移动互联网的下一代，还是我们会看到大量颠覆，全新的公司会捕获 ITSM、CRM、ERP 或任何三个字母的缩写？

But I guess the question I had is there's sort of this... I guess the Salesforce of mobile proved to be Salesforce in a lot of ways. It was kind of a continuing, at least within that, and obviously there's disruption that ended up happening with Uber and taxis or Airbnb and hotels or whatever it is. I guess as you think about those big software players, and you mentioned ServiceNow and SAP, obviously Salesforce in the ecosystem as well. Do you think that the lion's share of those will be the ones that persist if we picked, I don't know, the 10 industries and who the horizontal or not horizontal, but like who the application players on there? Do you think the lion's share of those will survive and be the next generation of it akin to mobile, or do you think we'll look at a lot of disruption and net new companies are going to be who captures ITSM or CRM or ERP or whatever three-letter acronym?

我认为会有颠覆。但我不认为其中任何一家一定会被颠覆。实际上，我认为这需要大量的执行。有一次我和 Shopify 的 Toby 开玩笑说，你找一个愤世嫉俗的工程师，他说：“你不就是云里的一个数据库吗？”你回答：“差不多吧。”是啊，这有点低估了我为这个云数据库投入的十多年工作，但确实差不多。如果你简化地看软件即服务，它就是云里的一个数据库，上面有很多工作流，而智能体最终会做这些工作流。所以，我没有在 ERP 系统上做很多工作，但如果你想象一下采购流程、合同以及构成它的所有事情，然后安永的审计师用它来审计你的季度财报。其中有多少会由 AI 智能体完成？当网页浏览器中的表单和字段不再被大量使用时，那个平台的价值是什么？不是零。我实际上认为那个分类账，你拥有的平衡账本，其实非常有用。所以它不是零。同样地，制造网页浏览器中表单和字段的公司能否制造智能体？是的，我绝对可以看到这一点。然而，我们在技术中一次又一次看到的是，有各种花哨的名字，比如创新者困境、跨越鸿沟。我不再记得它们都意味着什么了，但有效的是，当你是一个现有企业时，你会在某种程度上沉迷于你的产品和商业模式。结果，当你下面发生大的平台转变时，你最终通常会留下一堆公司的坟墓，这些公司通常由于商业模式原因无法完成转型。比如 Siebel Systems 的本地部署软件相对于 Salesforce 这样的公司。

I think there will be disruption. But I don't think any one of those will definitely be disrupted. I actually think it's just going to require a lot of execution. I was joking with Toby at Shopify one time that you find a cynical engineer and like, 'Aren't you just a database in the cloud?' and you're like, 'Kind of.' Yeah, I mean, it kind of undersells the decade plus of work I put into this database in the cloud, but yeah, it kind of is. If you look at software as a service reductively, it is a database in the cloud with a lot of workflows on top, and what agents will end up doing is those workflows. So, I haven't done a lot of work on ERP systems, but if you just imagine the procurement processes and contracting and all the things to make up it, and then the Ernst & Young auditor using it to audit your financials for a quarterly earnings report. How much of that will be AI agents? And then what is the value of that platform when the forms and fields in the web browser aren't used very much anymore? Not nothing. I actually think that ledger, the balanced books that you have, is actually quite useful. So it's not zero. And then similarly, could the company who made the forms and fields in the web browser make the agents? Yeah, I could definitely see that. The thing that we've seen time and time again in technology though is, there's all the fancy names for innovator's dilemma, crossing the chasm. I can't remember what they all mean anymore, but effectively, when you're an incumbent, you sort of become addicted to your product and business model. And as a consequence, when there's a big platform shift underneath you, you end up usually with a graveyard of companies who, usually for business model reasons, couldn't make that transition. Siebel Systems with on-premises software relative to a company like Salesforce.

你可以看到像微软这样的公司，在互联网和云领域经历了起起伏伏，最终变得非常强大。所以如果你足够大，你就有多次机会应对这些转型。我的猜测是，我们会看到几个故事——也许有一个像萨提亚那样的故事，一家公司通过转型变得比以前更强大；也会有一些像 Siebel Systems 那样的故事，公司眼睁睁看着慢动作的车祸发生在眼前，却因为阻碍大公司创新的自然引力而无法及时转向。我非常感激自己曾短暂担任过上市公司的 CEO，因为你能真切感受到那些引力。投资者说他们着眼长期，但每个季度你都要出去接受一次业务成绩单。我非常钦佩像 Adobe 的 Shantanu 这样的人，他经历了从永久许可软件到可确认收入的转型。这真是一门手艺——真正的商业模式转型。它会改变你的资产负债表，改变你的会计方式。所以突然间，你的盈利看起来或高或低，尽管我总是对会计感到好笑。现金流是一样的，但所有这些数字都在变动。除此之外，你还必须成为一个出色的讲述者，带着你的员工、投资者和客户一起前进。这非常艰巨。所以我认为很多创业者会说，‘是的，我们要颠覆现有企业。’实际上，在那个位置上坐过一段时间后，我充满了同理心。我认为一些伟大的领导者会从这场变革中脱颖而出，就像萨提亚因将微软转型到 Azure 云时代而获得的声誉一样，但这绝不是必然的结果。事实上，对于大多数现有企业来说，如果他们不拥抱软件的发展方向，默认结局就是被颠覆。

You could see companies like Microsoft had fits and starts with the internet and cloud and came out quite strong. So if you're big enough, you get multiple at bats with these transitions. My guess is we will have a few stories—maybe a Satya-level story of a company that came out even stronger on the other side—and we'll have some stories like Siebel Systems, of companies that saw the slow-motion car wreck in front of them and couldn't turn the wheel fast enough, just because of the natural gravitational forces that keep large companies from innovating. The thing I feel really grateful to have briefly been a public company CEO is that you really feel those gravitational forces. Investors say they're long-term, and every single quarter you go out and get a report card on your business. I have so much admiration for people like Shantanu at Adobe, who went through that transition from perpetual licensed software to ratable revenue. It's so much a craft—true business model transitions. It changes your balance sheet, changes your accounting. So all of a sudden you look more or less profitable, even though I always laugh about accounting. It's like the cash flow is the same, but all these numbers shift around. And on top of that, you have to be a great storyteller and bring your employees along with you, your investors along with you, your customers along with you. It's formidable. So I think a lot of entrepreneurs are like, 'Yeah, we're going to disrupt the incumbents.' And actually, having sat in that seat for a while, I have a ton of empathy. I think some great leaders will come out of this, and just like the reputation Satya has for transitioning Microsoft into the Azure cloud era, it is absolutely not a foregone conclusion. In fact, for most of these incumbents, the default will be that they will be disrupted if they don't embrace where software is going.

你过去谈到过这一点，刚才也暗示了，但商业模式方面的因素似乎要难得多，因为涉及那么多利益相关方。你谈到了股东、员工和客户，以及管理预期、带领大家应对所有会计上的特殊性。在你看来，这似乎比技术层面的挑战更大。

You've talked about this in the past and you kind of alluded to it there, but the business model elements of it seem far harder with all those constituencies. You talked about shareholders and employees and customers and just managing expectations and taking people through all the accounting idiosyncrasies. It seems like that in your mind is even more challenging than the technological elements.

毫无疑问，在我心里是这样。实际上，我并不否认技术有时确实很难。1995 年，构建一个可扩展的网站非常困难。现代实践还不存在；像 memcache 这样的东西还不存在，仅仅让数据库扩展就是一件大事。我曾与一位在 Salesforce 创立时在 Oracle 工作的人交谈过，他们当时认为数据库中的多租户是一个新颖的概念。所以，是的，我不想低估技术。但你不需要成为 AI 研究员就能看到，让智能体在一定程度上有效工作很快就会变得非常容易。现在确实很难，但技术会进步。所以，如何构建一个对利益相关者有用的产品是一回事，但商业模式是什么？商业模式转型之所以困难，就在于其中的变革管理。如果你投资组合中的任何一家 SaaS 公司，可能都有相同的激励结构：你有年度经常性收入，销售人员被激励去增加它，你可能还有一个团队被激励去减少流失。所以你是在增加你的年金，并努力降低流失率，以免失去年金的价值。所有软件即服务公司都是这样运作的。现在，从单个客户的角度来考虑。他们付给你——我取个整数——他们每年付给你 100 万美元，而你有一项颠覆性的新技术，10 年后意味着他们会付给你 1000 万美元。但如果你在那一年推出它，收入会降到 20 万美元。你实际上要怎么做？如果你是一家上市公司，这非常困难，因为突然间你的业务看起来在放缓，即使你只是在为了更光明的未来而“吃蔬菜”。如果你是一家尚未盈利或现金流为正的初创公司，你的烧钱率会不会突然飙升？如果这些都不适用——你是一家私营公司，却同时具备所有这些特征——突然间你的员工可能会中途跳槽，因为他们是在赌未来。看看从 Windows 收入到 Azure Active Directory 的转型。每一次都是一场地面战，在逐个客户的基础上将人们过渡到那个新世界。对于纸上谈兵的战略家来说，很容易说，‘当然，这是正确的做法。’然后问题是，如果你不在乎混乱的中间过程，是的，这完全显而易见，但世界上每个人都在乎混乱的中间过程。因此，这需要极其卓越的领导力才能完成这些转型。我认为人们对此考虑得不够。部分原因是因为很多人以前没有真正大规模运营过企业。我以前讲过这个故事，在谷歌，我们的第一个园区是 SGI（硅谷图形公司），在他们实际上要倒闭的时候我们搬了进去。在 Facebook，我们搬进了 Sun Microsystems 的园区，当时他们被甲骨文收购并迅速关闭。这两家公司在我相对短暂的职业生涯中都曾是成功的公司，建造了园区，而在我当时相对短暂的职业生涯中，它们已经倒闭，正在出售园区。这只是因为他们的技术和商业模式曾经很出色，但没有过渡到下一个。而这正是我们行业公司的默认结局。只有极少数公司能够转向多个产品、改变商业模式，你可以用两只手数过来。我认为作为创业者，牢记这一点非常重要，因为如果你的目标是创建一家比你活得更久的公司——这当然是我的目标——你必须创造一种文化，而不是只会一招鲜。

Without question in my mind. I actually don't—there are points where tech is really hard. In 1995, it was really hard to make a scalable website. Modern practices didn't exist; things like memcache didn't exist, and just getting databases to scale was a big deal. I was talking to someone who worked at Oracle when Salesforce was starting, and they were talking about multi-tenancy in a database as a novel concept. So yeah, I don't want to minimize the technology. But you don't need to be an AI researcher to see that making agents that work somewhat effectively will get really easy relatively soon. It's really hard right now, and technology will improve. So how you actually build a product that's useful for a stakeholder is one thing, but then what is the business model? The reason why it's hard to make a business model shift is just the change management of that. So if you're any SaaS company in your portfolio, you probably have the same incentive structure: you have your annual recurring revenue, and sales people get incentivized to increase it, and you might have a team that is incentivized to reduce attrition. So you're adding to your annuity and trying to keep attrition down so you don't lose money value in that annuity. That's how all software as a service companies work. Now think about it from the perspective of an individual account. They're paying you—I'll just make it even numbers—they're paying you a million dollars a year, and you have a disruptive new technology that in 10 years will mean they're paying you $10 million. But if you rolled it out that year, it would go down to $200,000. How do you actually do that? If you're a public company, that's really hard because all of a sudden you're going to look like your business is slowing down, even though you're just eating your vegetables for a greater, brighter future. If you're a not-unprofitable or not-cash-flow-positive startup, all of a sudden does your burn rate go way up? If none of those apply—you're a private company that somehow is all those things—all of a sudden your employees might jump ship in the middle because they're betting on the come. And you just look at the transition from Windows revenue to Azure Active Directory. Each one of those is a ground game at the customer-by-customer level to transition people into that new world. And it's so easy for armchair strategists to be like, 'Of course, that's the correct thing to do.' And then the question is, well, if you don't care about the messy middle, yeah, it's totally obvious, but everyone in the world cares about the messy middle. So it requires just incredibly exceptional leadership to go through those transitions. And I think people don't think about it enough. I think in part just because a lot of folks haven't really operated businesses at scale before. I've told this story before, but at Google, our first campus was SGI, Silicon Graphics, and we moved into their campus after they effectively were going out of business. And at Facebook, we moved into Sun Microsystems' campus after they were acquired and summarily shut down by Oracle. Both of those companies in my relatively short career were successful companies and built campuses, and in my relatively short career at the time had gone out of business and were selling their campus for parts. It was just because their technology and business model was once great and they didn't transition to the next one. And that is the default for companies in our industry. There's just a very small handful of companies that have been able to move to multiple products, change business models, but you can list them on maybe two hands. And I think it's very important as entrepreneurs we keep that in mind, because if your ambition is to create a company that outlives you—which is certainly my ambition—you have to create a culture that's not like a one-trick pony.

我听过您谈论当今 AI 生态的三个不同板块：基础模型、工具和应用 AI。有一个问题是，基础模型会在哪些领域渗透到应用中，哪些领域又更难渗透。当您看到像 Claude Code 这样介于基础模型和实际应用之间的东西时，您同时拥有 OpenAI 的董事会席位和自家公司，视角独特。您如何看待公司在基础模型之上作为应用 AI 系统存在的空间，以及基础模型会在哪些地方捕获价值？

I've heard you talk about the three different segments of the AI ecosystem today: foundation models, tools, and applied AI. One question that comes up is where foundation models will push into applications and where it might be harder. When you look at something like Claude Code, which sits between the foundation model and the application, you have a unique purview with a board seat at OpenAI and your own company. How do you think about where companies can exist as applied AI systems on top of foundation models versus where foundation models will capture value?

我提供一个思考技术演进的框架。在新技术早期——无论是智能手机还是网页浏览器——很多价值在于让它“能用”。1995 年或 1998 年，做一个网站或可扩展的数据库很难。到 2025 年，做网站或可扩展数据库轻而易举。所以早期，初创公司和现有企业卖的是新平台能做什么以及它能用。如果从 1998 年到 2025 年你的价值主张还是这个，你作为公司就不存在了，因为随着平台和生态成熟，它变成了商品。我假设这也会发生在推理智能体、行动智能体、语音智能体等今天很难的事情上。所以构建应用 AI 智能体的公司今天需要专注于让它能用，但也需要思考在此基础上自己的价值。每个 SaaS 公司本质上就是云上的一个数据库，但他们在数据库之上创造了巨大价值。没人再关心数据库是否可扩展。我做尽职调查时会问：一旦技术能用、不再是差异化因素，产品变成什么？有些领域会有显而易见的答案，这些答案其实不关乎 AI，而是技术与业务应用的结合。比如 Ramp——我们是 Sierra 的客户，也是它的超级粉丝——我不在乎他们用什么数据库。他们提供的价值由技术促成。我在乎的是我能为 Sierra 的每个员工配一张公司信用卡，这很棒。这就是将要发生的转变。

I'll give a framework I think about for technology evolution. In the early days of a new technology—whether the smartphone or the web browser—a lot of the value is in making it just work. It was hard to make a website or a scalable database in 1995 or 1998. In 2025, it's trivial to make a website or a scalable database. So in the early days, startups and incumbents sell the vision of what that new platform is and that it works. If that remains your value proposition from 1998 through 2025, you won't exist as a company because it becomes a commodity as the platform and ecosystem mature. I assume that will happen with reasoning agents, action-taking agents, voice agents, and all the things that are really hard today. So companies building applied AI agents need to focus on making it work today, but they also need to think about their value on top of that. Every SaaS company is just a database in the cloud, but they've created a ton of value on top of that database. No one cares that the database scales anymore. The question I would have for due diligence is: once the technology works and is no longer the differentiator, what does the product become? Some domains will have obvious answers that aren't really about AI, but about the intersection of technology with the business application. For example, Ramp—we're a customer at Sierra and a huge fan—I don't care what database they use. The value they provide is facilitated by technology. What I care about is that I can bring a corporate credit card for every Sierra employee, and it's great. That's the transition that will happen.

思考基础模型公司可能进入哪些应用领域——如果历史不会重演但会押韵，基础设施即服务公司在哪些地方消除了上层 SaaS 应用的需求？

Thinking about where foundation model companies might move into applications—if history doesn't repeat but rhymes, where have infrastructure-as-a-service companies obviated the need for a SaaS application on top?

有几个经验法则。第一，它是开发者产品吗？开发者工作在基础设施上，所以周围工具会有引力。你会看到在开发者平台上，AWS 或 Azure 经常有直接竞品，类似开源竞品。第二，工具领域——比如数据标注——可能面临基础模型公司进入的风险，因为如果你想成为基础模型提供商的大客户，这是一个相邻领域，自然扩张。但越接近真正的垂直业务应用，这种可能性就越大。迄今为止，还没有一家公司成功同时销售横向基础设施和垂直应用。我不认为这只是技术问题——无意冒犯我提到的 SaaS 公司——但并不是亚马逊或 Azure 做不了那些产品。他们当然能，但产品管理文化完全不同，面向不同的买家，商业模式也不同。销售解决方案和销售基础设施之间有一千个不同理由，这实际上造就了两种不同的公司。我不认为这在 AI 时代会改变。

There are a few rules of thumb. First, is it a developer product? Developers work on the infrastructure, so there's a gravitational pull from the tools around it. You see that with developer platforms where AWS or Azure often have direct competitors, similar to open source competitors. Second, the tools space—like data labeling—might be at risk of foundation model companies moving into it because if you're trying to be a great customer of a foundation model provider, it's an adjacency and a natural expansion. But the closer you get to a vertical business that's really an application, it could happen. To date, no company has successfully sold both horizontal infrastructure and vertical applications. I don't think it's just technology—no offense to the SaaS companies I mentioned, but it's not that Amazon or Azure couldn't build those. They could, but it's a very different product management culture, selling to different buyers, with a different business model. The sum of a thousand reasons why selling a solution is different from selling infrastructure creates two different companies. I don't see that changing in the age of AI.

您刚才间接提到了我们经常看到的一个现象——我称之为“部署工程”文化，也就是“palunteerism”，这个词现在用法和原来在 Palantir 的意思不太一样了。我想听听您的看法，因为这可能和您说的“初期需要缝合这些组件”有关，今天可能还带有人工服务的成分，但也许不会永远如此。

是的，我先说说为什么我觉得它现在很流行，然后再谈 Sierra 的具体情况。我认为采用 AI 的变革管理相当艰巨。

如果你有一个智能体要对接 20 或 30 个系统，既有系统集成的复杂性，也有变更管理的问题——以前是谁在做这件事，如何设置护栏等等。随着时间的推移，我认为所有这些都会产品化，但与此同时，AI 项目失败的主要原因之一是采用或部署。所以我认为很多公司，尤其是年轻的 AI 公司，发现更亲力亲为或高接触度与为客户带来成果之间存在强相关性，这解决了部分采用差距。

If you have an agent that takes action against 20 or 30 systems, there's both the systems integration complexity and the change management of who was doing that before, how do you put guardrails around it, all these other things. Over time, I think all this will be productized, but in the meantime, one of the main ways an AI project fails is adoption or deployment. So I think a lot of companies, younger AI companies in particular, have found a strong correlation between being more hands-on or high-touch and driving outcomes for customers, which solves some of that adoption gap.

在 Sierra，我们的模式非常灵活。我们为客服团队（比如运营团队）提供无代码产品，这样你无需任何技术知识就能构建智能体。我们有一个用于构建智能体的平台即服务，叫做 Agent SDK，像 Ramp 这样的工程团队用它来构建 AI 功能。此外，我们还有一个可选的智能体开发团队，可以手把手帮你完成整个过程。这样我们不仅提供产品，还能真正让你成功使用它。

At Sierra, we have a model that's quite flexible. We have a no-code product for customer experience teams, like operations teams, so you can build an agent without knowing any technology. We have a platform as a service for building agents called our Agent SDK that engineering teams like Ramp use to build AI functionality, and then we have an optional agent development team that can help handhold you through that process. It's really there so that we can show up not just with a product but actually make you successful with it.

这当然不是 Palantir 模式本身。这只是我们表达的方式：我们会适应你使用这项技术的方式。可能是你的运营团队，也可能是你的工程团队。也许你不知道如何采用，我们会让你无论如何都成功。今天宣布融资时，我特别自豪的一点是，超过一半的客户营收超过 10 亿美元，超过 20% 的客户营收超过 100 亿美元，这对我们这个年龄的公司来说很不寻常。我认为这是因为我们接触的是许多受监管最严格的行业中的公司，它们有大量约束，我们会讨论基于 AI 的护栏是否足够，是否需要确定性护栏，如何处理合规和审计。这里面有很多产品，但很大一部分是集成到这些控制机制当前的工作方式中，以及在你因为采用这项技术而改变 BO 策略时，我们如何帮助你进行变更管理。对我们来说，这真的是努力成为合作伙伴而不仅仅是供应商，目的是推动成果，确保这不是 AI 观光，而是这些智能体真正上线并快速带来商业价值。

It's definitely not the Palantir model per se. It's just our way of saying we're going to accommodate the way you want to use this technology. Maybe it's your operations team, maybe it's your engineering team. Maybe you don't know how to adopt it, and we're going to make you successful no matter what. One of the things I was really proud of when we announced our financing today is that over half of our customers have over a billion in revenue, and over 20% have over 10 billion in revenue, which is unusual for a company of our age. I think it's because we're approaching companies in many of the most regulated industries that have a ton of constraints, and we'll have conversations about whether AI-based guardrails are enough, whether you need deterministic guardrails, how to deal with compliance and auditing. There's a lot of product in there, but a lot of it is integrating into how those controls work today, how we can help you with that change management as you're changing your BO strategy because you're adopting this technology. For us, it's really trying to be not just a vendor but a partner, and it's really there to drive outcomes and ensure that this isn't AI tourism but these agents are actually going live and driving business value quickly.

我不知道这个行业会走向何方。我不太了解 Palantir 的人，不过我们团队里有他们的一些前员工。但我认为他们引领潮流是件好事。有趣的是：我们曾在 Facebook 搬进旧的 HP 大楼，他们以创造第一个开放式办公室布局而闻名。我想他们用的是隔间，所以不完全是开放式办公室。而 Marissa 在 Google 做的助理产品经理项目也被复制了。我认为无论角色是否完全相同，都要归功于 Alex 和 Palantir 团队。我喜欢硅谷对这些不同交付和组织模式的探索，我认为我们会从中受益。我们会学到很多。没有公司想成为专业服务公司，所以对那种模式的 caricature 并不好，但我喜欢它隐含的责任感。

I don't know where it will go as an industry. I don't know the Palantir folks really well, though we have a few of their alumni on the team. But I think it's great for them for starting a trend. It was interesting: we moved into the old HP building once at Facebook, and they were famous for creating the first open office floor plan. I think they had cubes, so it's not quite as open office. And what Marissa did at Google with the associate product manager program has been replicated. I think credit to Alex and the Palantir team for whether or not it's exactly the same role. I love these explorations of different delivery and organizational models in Silicon Valley, and I think we'll benefit from it. We'll learn a lot from it. No company wants to be a professional services firm, so there's a caricature of what that becomes that is not great, but I like the accountability that it implies.

是的。而且它允许在客户所处的阶段与他们相遇，这在某种程度上向大型组织销售时尤其重要。我认为他们做得最好的一件事就是教条地选择未来方向，并纯粹地坚持‘不，我们是云，我们在云上工作，这就是我们要做的，我们不会为你做本地部署’，并将这一点投射到那些不愿意在他们看到的必然路径上相遇的客户身上。有趣的是，你想在旅程中带领这些客户并与他们相遇，我认为这在当下非常重要。我们在不同行业也看到了这一点，但我想每个市场都会有点不同，每家公司都会以不同的方式执行。

Yeah. And it allows meeting customers where they are in their journey, which I'm sure is particularly important as you sell to larger organizations in some ways. There's going to be... one of the things I think they did best was dogmatically picking where the future was headed and just being a purist about 'no, we're the cloud, we're working in the cloud, this is what we're going to do, no we're not going to do on-prem for you' and projecting that onto customers who didn't want to meet them on the path of inevitability they saw coming. It's interesting where you want to bring these customers along and meet them in the journey, which I think is super important at this moment in time. We've seen it across different industries as well, but I guess every market is going to be a little bit different and every company's going to execute in a different way.

嗯，有几个元观点我觉得非常值得思考。另一个我希望知道的数据是：在企业软件市场，每年花在软件许可上的百分比与实施上的百分比是多少？

Well, there are a few meta points that I think are really interesting to think about. Another number I wish I had: in the enterprise software market, what percentage annually is spent on software licensing versus implementation?

我认为大概是实施费用的 3 倍左右。

I think it's like 3x implementation or something.

我认为这很可能正确。其中一部分只是糟糕的决策，过度定制。我和很多非常资深的 CIO 谈过，他们意识到了这一点，你更希望开箱即用，这样总拥有成本更低。但有趣的是，现在有了软件工程智能体，它们可能会降低实施的边际成本。有一个完整的生态系统，公司围绕一些非常高端的咨询、更常规的实施以及一系列系统集成商而建立。然后还有软件公司采用这种前向部署的模式。但你是否也将这一切与 AI 智能体结合起来？实施成本会下降吗？我认为可能会，然后做这件事的人可能会有所变化。这将如何发展并不完全清楚。但在此基础上，你再加上基于结果的定价。房间里的大象是，如果实施得不好，你就拿不到报酬。所以你必须有一个更负责任的模式来促进这种商业模式。我认为所有这些……我们在系统集成商社区有一批非常棒的合作伙伴。但这是与他们和我们之间关于走向更负责任的对话。我对世界各地的公司抱有希望：实际部署这些技术的成本应该会大幅下降，我们努力促成这一点，因为我认为公司不需要巨大的前期成本就能在 AI 中找到价值非常重要。但同时有很多变量在变化，我只能想象，如果 10 年后你我还坐在这里，我问同样的问题，答案会不同。我不知道，但我想答案会不同。

I think that's probably right. And some of it is just bad decision-making, too much customization. I've talked to a lot of very sophisticated CIOs who are aware of that, and you kind of want more out of the box so your total cost of ownership is lower. But it's interesting too because you have software engineering agents now which will presumably reduce the marginal cost of implementation. You have an entire ecosystem of companies built around some very high-end consulting, some more routine implementation, and there's a spectrum of systems integrators there. And then you have software companies with this sort of forward-deployed motion. But do you combine that all with AI agents as well? Does the cost of implementation go down? I think probably, and then who the people who do it might shift a bit. It's not totally obvious how that will play out. But then on top of that, you combine that with outcomes-based pricing. The elephant in the room with outcomes-based pricing is if it doesn't get implemented well, you don't get paid. So you have to have a more accountable model to facilitate that business model. I think all of that... we have a really great set of partners in the systems integrator community. But it's a conversation with them and us of moving towards more accountability. I'm hopeful for companies around the world: the cost of actually deploying these technologies should go down a lot, and it's something we try to manufacture just because I think it's really important that companies don't have huge upfront costs to find value in AI. But there are a lot of variables shifting at the same time, and I just have to imagine if you and I are sitting here 10 years from now and I ask that same question, I imagine the answer will be different. I don't know, but I imagine the answer will be different.

嗯，同样的事情也发生了。

Well, and the same thing happened.

我想这并非全新现象，就像我们之前从 TCO 预付费模式转型时那样。我记得职业生涯中有过云系统集成商，比如 Cloud Sherpas 和 Appirio。每个人都必须以某种方式成为云系统集成商。他们必须想办法让账算得过来，无论是分期支付 1000 万美元还是一次性预付 1000 万美元。这让我想到一件事：软件开发领域正在发生一场民主化运动，使得组织能够以比以往任何时候都更轻量的方式构建定制化解决方案。我肯定你有过客户尝试自己动手，或者他们现在还在这么做，而你的解决方案显然比构建一个调用薪资数据的 HR 机器人要复杂一些。另一方面，软件开发的民主化将让人们能够更有意义地将应用商业化。最终，我们是否会看到更多软件被销售到组织中，因为民主化使得专业化成为可能，比如面向 HVAC 行业的 SMB ITSM？还是说存在某种限制，人们不想在自己的组织里拥有 5000 个不同的商业供应商？

I guess it's not entirely novel when we were shifting from TCO upfront licenses. I remember in my career there were cloud systems integrators, like Cloud Sherpas and Appirio. Everyone had to become a cloud system integrator in some way. They had to figure out how to make the math work if it was $10 million over time versus $10 million upfront. This brings up something I've thought about: we have this democratizing force in software development that enables organizations to build bespoke solutions much more easily than ever before. I'm sure you've had customers who tried to DIY, or maybe they still are, and your solution is a bit harder than building an HR bot that calls payroll data. On the other side, democratizing software development will allow people to commercialize applications more meaningfully. At the end of the day, will we see more net software sold into organizations because it's democratized, allowing specialization like an SMB ITSM for the HVAC industry? Or is there a limiting force on how many solutions people can buy, and you just don't want 5,000 different commercial vendors in your organization?

嗯，这是个很好的问题。我认为有两股趋势，或者说不止两股，正在同时发生。首先，整个经济在过去 25 年里一直在数字化，而我认为 AI 的出现只会加速这一进程，因为它正在数字化以前并非数字化的职业，比如律师助理行业。这将增加对软件的需求。同样，从我出生以来，软件开发人员就一直短缺，而现在我们有了软件工程智能体。在某个时候，随着这些智能体变得更高效，我们大概会弄清楚到底需要多少软件工程师。我们从未经历过这种情况。我不知道答案，因为我们不知道软件的实际需求。然后，正如你所说，软件行业正在被颠覆：什么是软件公司？我的直觉是，基础设施提供商在销售智能。而我的直觉是，以前做应用的 SaaS 公司，明天会成为做智能体的公司，它们会构建专用智能体。公司可能会围绕某个业务线或行业拥有五个智能体。这是我的假设。我认为大多数公司会愿意从软件公司购买智能体，而不是自己动手。原因是，即使软件工程取得了进步，如果你构建了软件，你就拥有它，并且必须维护它。软件就像草坪：它不能静止不动还能继续发挥作用。以 ERP 为例。假设有一个新的会计准则，比如六、七年前出台的 606。软件行业的整个经济逻辑是，每个人都需要那个新会计准则，而不仅仅是你。一家软件公司可以摊销推出该准则的复杂性。除此之外，如果你要报告收益，你怎么知道它是正确的？你实际上是把正确性和责任外包给了一家公司。这在我看来非常合理。这个想法从来都不复杂，但摊销创新成本的激励仍然有意义，因为即使你可以让 AI 来编写，你怎么知道它是正确的？顺着这个思路想下去，公司就应该这么做。

Yeah, it's a wonderful question. I think two trends are happening, or maybe more than two, at the same time. First, the whole economy has been digitizing for the past 25 years, and I think the emergence of AI will just increase the pace of that evolution because it's digitizing professions that weren't really digital before, like the paralegal profession. That will increase demand for software. Similarly, we've had a shortage of software developers since I've been alive, and now you have software engineering agents. At some point, presumably as these agents become more productive, we'll figure out what the limit of how many software engineers we need is. We've never experienced that. I don't know the answer because we don't know the actual demand for software. Then, as you said, you have the disruption of the software industry: what is a software company? My intuition is that you have infrastructure providers selling the intelligence. And then my intuition is that the software-as-a-service companies who made apps before will be companies that make agents tomorrow, with purpose-built agents. Companies might have five agents around a particular line of business or industry. That's my hypothesis. I think most companies will want to purchase agents from software companies rather than DIY. The reason is that even with advancements in software engineering, if you build software, you own it and you have to maintain it. Software is like a lawn: it can't sit statically and continue to do its job. Take the ERP example. Let's say there's a new accounting standard, like 606 came out six or seven years ago. The whole economics of the software industry is that everyone needs that new accounting standard, not just you. A software company can amortize the complexity of rolling that out. On top of that, if you're reporting earnings, how do you know it's correct? You're essentially outsourcing the correctness and accountability to a firm. That seems quite rational to me. The idea was never that complicated, but the incentives of having amortized costs of innovation continue to make sense because even if you could ask an AI to write it, how do you know it's right? Pull that thread, and it just seems like a company should do that.

把边缘情况分散到不同的供应商和内置的专业知识中，会让人感到安心。

There's some comfort in having the edge cases spread out across a bunch of different vendors and expertise built in.

当你购买一个新产品时，知道还有谁在使用它，并且你也信任他们的判断，难道不会感到安心吗？这种动态是真实的。我认为如此，这就是为什么我强烈直觉它会继续下去。不过，了解我们行业的形态确实很有趣。我对此思考了很多，也谈论了很多，因为我自认是一名计算机程序员。看到自己的职业被一项技术如此颠覆，真是非常有趣——我们从软件的创作者转变为代码生成机器的操作者。你需要或想要多少机器操作员？他们需要什么工具？那份工作有多令人满意？我的直觉是，它实际上会更有趣、更令人满意。我只是喜欢为自己的生活增加杠杆，能够像指挥交响乐一样指挥一群软件工程智能体。我和办公室里的很多人聊过，他们在上班前启动一个智能体，然后进来查看拉取请求。这真的很酷。我不知道这在行业中会如何发展，但我相信会有智能体公司出现。我相信它们会带来巨大的价值。

And don't you feel secure when you buy a new product that you know who else uses it, and you trust their judgment too? That dynamic is real. I think so, that's why my strong intuition is it will continue. It is really interesting though to know what is the shape of our industry. I've thought a lot about this, and I've talked about it a lot because I identify as a computer programmer. It's really interesting to see your own profession so upended by a piece of technology as we migrate from being authors of software to operators of code-generating machines. How many machine operators do you need or want? What are the tools that they need? How satisfying is that job? My intuition is it will be actually more fun and more satisfying. I just love adding leverage to my own life, and to be able to symphony conduct a bunch of software engineering agents. I talked to so many people at the office who, before they head into work, kick off an agent and then come in to look at the pull request. It's pretty cool. I don't know how it'll play out in the industry though, but I believe that there will be agent companies. I believe they'll be delivering a ton of value.

事实上，我认为价值可能大得多，因为它们实际上是在完成任务，而不是模糊地提升生产力。说到企业软件，任何在这个行业待得够久的人都做过关于 ROI 的演示。你知道，几乎都一样——我不是说那是胡说八道——但你有 X 个销售人员，他们都提高了 Y 的生产力，你应该付钱给我们。哦，我们只要求一半的价值。房间里每个人都持怀疑态度。而现在，这些智能体实际上在自主执行任务。我认为这既极其有价值，作为一家公司也极其有价值。

And in fact, I think perhaps a lot more value because they're actually accomplishing tasks as opposed to amorphous productivity enhancements. And you talk about enterprise software. Anyone who's been in this industry long enough has had to do a presentation about ROI. You know, it's almost the same. I don't say nonsense, but you have X salespeople and they all get Y more productive, and you should pay us. Oh, we're only asking for half of that value. And everyone in the room is skeptical. And now you have these agents actually autonomously performing tasks. I think it's both extremely valuable, and I think it's extremely valuable as a company.

是的，这很有趣。那种编排的事情……我可能会得出的答案是——当人们大声说出来时，这让他们惹上麻烦——那就是：人更少可能会更有趣。本质上，可能只需要更少的人以某种方式编排多智能体。但我们不知道。

Yeah, it's interesting. That orchestrating thing... It may be the answer that I think I would probably land on, which has gotten people in trouble when they've sort of said out loud, is like: it might be far more fun with fewer people. There just might be less people inherently orchestrating the multi-agents in some way. But we don't know.

也许吧。但我们不知道对软件的需求有多大。没错。我们从未体验过，也从未满足过。所以我不知道。回到我之前的类比：将客户电话的成本从 20 美元降到 20 美分，再降到 2 美分。这改变了动态。想象一下，如果网站的一次页面浏览成本是 20 美元，网络会完全不同。你根本不会有博客。那将是一个非常不同的世界。现在，没有人会考虑一次页面浏览的边际成本。我确定有成本，对吧？你加起来，但边际成本低到你不去想它。它就会改变。二阶和三阶效应是巨大的。所以当你考虑扩大一家软件公司时，你想增长，对吧？所以你拥有更少人的唯一原因是它不会促进你的增长。而对我来说，更少总是更好并不明显。问题是：你用更少的人就能饱和你的市场吗？我只是认为，这个市场能吸收多少创新——我们从未体验过。现在我们可能有机会了，但认为人会更少——我持保留态度。但下一个谷歌会不会说：‘我想要和以前程序员一样多的软件机器操作员，然后我就能生产出价值数万亿美元的东西’？我不知道。我不知道答案是什么。

You know, maybe. But we don't know how much demand there is for software. That's right. We've never experienced it; we've never satisfied it. So I don't know. And going back to my earlier analogy about reducing the cost of a customer phone call from $20 to 20 cents to 2 cents. It just changes the dynamic. Imagine if a page view of a website cost $20. The web would be really different. You just wouldn't have blogs. It would be a very different world. Now that you know, no one thinks about the marginal cost of a page view. I'm sure there is a cost, right? You add it up, but so marginal you don't think about it. It's just going to change. The second and third order effects are huge. So as you think about scaling up a software company, you want to grow, right? And so the only reason you'd have fewer people is if it didn't contribute to your growth. And it's not obvious to me that less is always better. The question is: do you saturate your market with fewer people? And I just think that the saturation of how much innovation this market can absorb—we've never experienced that. And now we might have the opportunity to, but the idea that it's strictly fewer people—I'm like maybe. But is the next Google the one that says, 'I want just as many software machine operators as I had programmers before, and I'm just going to produce something that's worth trillions in value as a consequence'? I don't know. I don't know what the answer is.

我觉得我们有点过于深入企业软件了，但你是曾经最大的——可能是最大的独立应用软件纯玩公司的联席 CEO。从软件消费的角度来看，想想我自己的行为：我已经明显减少了访问网站的次数，因为我直接问 Claude 或 ChatGPT。是的，你可以配音。我会用 Harvey。你可以用——我可以问 ChatGPT 我的问题。我不再去网站了。我不再像过去那样去维基百科查，比如 Bret Taylor 的工作经历之类的。你认为企业软件消费是否必然朝着那个方向发展——我们有一个中央门户（ChatGPT），我们向它提问关于 Salesforce、Snowflake、Workday 或 ServiceNow 的问题，而不再像过去那样访问各个域名？

I feel like we're nerding out about enterprise software, but one of you were the co-CEO of maybe the biggest—probably the biggest standalone application software pure play company out there. From a consumption of software standpoint, if I think about my own behavior, I've shifted to definitely visiting far fewer websites because I just ask Claude or ChatGPT. Yeah, you can dub. I'll do Harvey. You can do—I can ask ChatGPT what question I have. And I don't go to the websites. I don't go to Wikipedia to look up, you know, Bret Taylor's history of work or whatever it is in the same way that I did in the past. Do you think that it's an inevitability that enterprise software consumption moves in that direction where we have some central portal—ChatGPT—that we're asking the questions of Salesforce or Snowflake or Workday or ServiceNow, and that we're no longer going to the domains in the same way that we have in the past?

很可能。我的意思是，我认为最终你会拥有这个——实际上，我先从消费者的角度来说，我认为消费者在很多方面引领着企业——你最终会有一个智能体来帮助你消化和综合信息，帮助你采取行动。当我计划今年夏天的假期时，我完全用了 ChatGPT，效果很棒。我确信有 20 个网站我本可以去但没去，但我也去了——我最终通过这个服务找到了我们住的 Airbnb。所以它并没有减少 Airbnb 网站的流量，但很多旅游推荐网站和其他网站我确实没去。所以我认为这会有赢家和输家。但当你看看企业软件，我总喜欢把它看作要完成的任务：你雇佣这个软件来做什么工作？如果你看看我的母校，CRM 软件，它管理线索、机会和你的销售管道。仍然会有销售人员，他们仍然需要销售，他们需要负责，需要获得报酬。那会有巨大的价值。而那些要完成的任务——你如何完成它们——可能会发生巨大变化。尽管我很喜欢 Salesforce，但一个好的销售人员并不想整天盯着那个屏幕。他们想去销售，对吧？所以我认为这个行业的未来对使用者来说会更愉快。而且我认为它会真正改变它的形态。关于你和我刚才说的 ChatGPT，有一个有趣的点：聚合器——Ben Thompson 关于消费者互联网的理论在这里非常有趣。因为你看互联网，有需求生成，比如社交媒体和相关的广告平台，还有需求满足，比如网络搜索和按点击付费广告以及 AdWords 等等。而 ChatGPT 特别地正在改变很多这些东西。所以如果你看看直接面向消费者的零售市场，Shopify、Instagram 和 TikTok 之间的关系非常微妙，我认为你会看到消费者行为的巨大转变，这会产生大量的下游效应。就像人们花很多时间思考搜索引擎优化，依赖应用商店分发的人花了很多时间思考这个——有些人对此非常满意，有些人非常不满——我们正在进入一个新世界，这些智能体将为许多不同品牌调解消费者体验，我认为人们仍在思考其影响。我和一个在电信行业工作的人聊过，他们为优惠做了很多工作，比如激励你更换有线电视或手机套餐。

Probably. I mean, I think you end up with this—I actually think just speaking from a consumer perspective first, and I think consumer leads enterprise in a lot of ways—you'll end up having an agent who helps you digest and synthesize information, help you take action. When I planned our vacation this summer, I used ChatGPT exclusively, and it was amazing. And I'm sure there were 20 websites I didn't visit that I could have, but I also visited—I ended up finding the actual Airbnbs we stayed in through the service. So it didn't take away traffic from the Airbnb sites, but a lot of the travel recommendation sites and others that I didn't. So I think there'll be winners and losers in that. But then as you look at enterprise software, I always like to think of it as jobs to be done: what is the job you hire this piece of software to do? And if you look at my alma mater, CRM software, it manages leads, opportunities, and your sales pipeline. There's still going to be salespeople, and they still need to sell, and they need to be accountable, they need to get paid. That's going to be a ton of value. And those jobs to be done—how you do them—will probably change dramatically. And as much as I love Salesforce, a good salesperson doesn't want to stare at that screen all day. They want to go sell, right? So I think the future of that industry will hopefully be more delightful for the people using it. And I think it will really change the shape of it. One interesting point about what you and I just said about ChatGPT that I think is interesting though is kind of the aggregators—the Ben Thompson sort of theory of the consumer internet is really interesting here. Because you sort of look at the internet and you have demand generation, which is like social media and the ad platforms associated with them, and you have demand fulfillment, which is web search and pay-per-click ads and AdWords and things like that. So much of that with ChatGPT specifically is really changing. And so if you look at the direct consumer retail market, the relationship between Shopify and Instagram and TikTok and all that is quite nuanced, and I think you're just going to see a real big shift in consumer behavior which will have a ton of downstream effects. And just like people spend a lot of time thinking about search engine optimization, and people who are dependent on the app store for distribution have spent a lot of time thinking about that—some are really happy with it, some are very upset about it—we are entering a new world where these agents will mediate the consumer experience for a lot of different brands, and I think people are still working through the implications of that. I was talking to someone who works in the telecommunications industry about all the work they do for offers, like to incentivize you to switch cable or mobile phone plans.

如果是消费级智能体，优惠会消失吗？一切都会回归到最低价方案吗？在旅游行业，就像我描述自己度假时那样，哪些企业会从中受益，哪些会受损？如果你是一家聚合商，一直专注于 Google，你要如何将业务迁移到 ChatGPT？我的猜测是，就像围绕帮助人们有效购买广告、搜索引擎优化而建立的企业生态一样，现在还有一个全新的世界尚未被创造出来。但我也认为这确实会改变经济，因为很多变化都始于消费者行为，然后层层传导。我认为我们正处于这种行为变化的非常早期阶段。

And if it's consumer agents, do the offers go away? Does everything revert to the lowest price plan? In the travel industry, as I was describing my vacation, what businesses win from that, what lose? If you are an aggregator and you've been focused on Google, how do you translate that business to ChatGPT? My guess is just like there's an ecosystem of companies built around helping people purchase ads effectively, search engine optimization, there's this whole world that hasn't yet been created around that. But I also think it'll really change the economy because so much starts with consumer behavior and trickles down from there. And I think we're in the very early innings of that behavior change.

是的，从个人层面来看，我收到的内容个性化程度已经变得非常好，这很有趣。比如外发邮件，感觉很简单，但到某个点就会饱和。我收到的超个性化外发邮件数量，最终会达到一个饱和点……

Yeah, it's interesting to see at a personal level the personalization that exists in terms of the content I receive has just gotten so good. It could be outbound emails feels like a very simple thing, but at some point there's a saturation. The number of hyperpersonalized outbound emails I get, there's just a saturation point that I ultimately...

那你的个人智能体的饱和点是什么？

And what's the saturation point for your personal agent?

没错，是无限的，对吧？我想这是个有趣的问题。你怎么看？比如你和 Clay 打了个赌，关于智能体之间的交互何时会超过人机交互的 50%。你能给大家讲讲吗？然后你认为这有什么影响？

Yeah, exactly. It's infinite, right? I guess it's an interesting question. What do you think? Like you and Clay have a bet about at what point agent-to-agent will out be more than 50% of human-to-agent, I guess. Can you maybe outline that for people? And then what do you think the implications of that are?

是的。Clay 和我打了很多赌。到目前为止，Clay 赢了每一个。

Yeah. So Clay and I have a lot of bets. Clay has won every single one of our bets so far.

他是不是更像一个最大化主义者？他一直是更极端的乐观主义者吗？

Is he like a maximalist? Has he been more of the maximalist optimist?

他一直是更极端的乐观主义者。我不知道我是不是……我喜欢把自己看作乐观主义者，但在这种情况下，他更……

He has been the more maximalist optimist. And I don't know if I'm just... I like to think of myself as an optimist, but in this case, he's taken the more...

总得有人更极端。而且他每次都对了。所以我觉得这次他也会赢。

Someone has to be more. And he's been right every time. So I think he's going to win this one, too.

你知道，我们所有的智能体都有一个品牌在顶端，它们驱动着公司的客户体验。所以 ADT 有一个你可以通过电话或聊天交流的智能体，DirecTV 和 SiriusXM 也是如此。如今这些对话大多数是人与智能体之间的。我们曾想象一个世界，OpenAI、Google 和 Apple 的某种组合会有一个个人智能体，你可以派它去执行任务，比如去更改你的 SiriusXM 套餐，因为你想要不同类型的内容。我们打赌，当大多数对话从人与人转向其他 AI 智能体时，那个临界点何时到来。我认为这很可能会发生。我觉得你提到的 Claude 和 ChatGPT 的体验是真实的。这感觉像是这些体验的马斯洛需求层次：先思考，然后上网替我研究，之后再替我采取行动。我还没到把信用卡给它让它去订旅行的程度，但感觉那不会是在十年后。我觉得会快得多。而且我不认为我在那方面过于乐观。当你做这些服务时，它们在某种程度上变得无头化。如果你与客户的大部分互动都是通过他们的智能体进行的，那就会颠覆你对从用户界面设计到商业模式的一切思考方式。我不够聪明，无法预测二阶效应。我认为如果只是把当前世界投射到那个未来，很容易想象那个世界。但那不会发生。它会逐渐发生，并且会真正改变许多消费公司的商业模式。我们在 Sierra 希望提供的主要价值之一是帮助我们的客户为那个未来做好准备。他们的智能体与人类兼容，与其他智能体兼容。它们可以嵌入到他们的移动应用中，可以通过 WhatsApp 使用。如果智能音箱卷土重来，你可以把你的智能体发布到那些智能音箱上。我认为现在这非常重要。我们真的想成为每个客户值得信赖的 AI 顾问。我认为预测未来走向非常复杂。但第一步是让你的客户体验在那个新世界中可用。然后你可以决定那个世界的搜索引擎优化是什么。我不知道。但你最不希望的就是在那个新世界中无法进行商务活动、无法获得新客户。这是我们提供的长期价值主张之一。

You know, all of our agents have a brand at the top and they power company's customer experiences. So ADT has an agent you can talk to over the phone or over chat, and DirecTV and SiriusXM. And most of those conversations today are people talking to the agent. We had imagined a world where some combination of OpenAI, Google, and Apple will have a personal agent that you'll send on a task, who might for example go change your SiriusXM plan because you want a different type of content. We have a bet of when the line will cross where the majority of the conversations will switch over from people to other AI agents. I think this probably will happen. I think that the experience you were talking about with Claude and ChatGPT is real. I think it feels like Maslow's hierarchy of needs of those experiences: first think, second go out on the internet and do research on my behalf, and then after that take action on my behalf. I'm not quite to the level where I would give it my credit card and ask it to go book a trip, but it doesn't feel like that's a decade from now. That feels way sooner, I think. And I don't think I'm that overly optimistic on that front. And when you do these services, they sort of become headless in some ways. If the majority of your interactions with your customers are via their agents, it just upends how you think about everything from user interface design to what your business model is. And I'm not smart enough to predict the second order effects. I think it's easy to imagine that world if you just took the current world and projected it into that future. But that's not what's going to happen. It's going to be gradual and it will really change the business model for a lot of consumer companies. One of the main values we hope to provide at Sierra is helping future-proof our customers for that. Their agents are compatible with people. They're compatible with other agents. They can be embedded in their mobile app. They can be used via WhatsApp. If smart speakers make a comeback, you can publish your agent to those smart speakers. I think that's really important right now. I think we really want to be a trusted AI advisor to each of our customers. And I think it's so complex to predict where the future is going. But step one is make your customer experience available in that new world. Then you can decide what is the search engine optimization of that world. I don't know. But the last thing you want is to not be able to conduct commerce and gain new customers in that new world. And that's one of the longer-term value propositions that we provide.

你在描述 Sierra 时已经触及了这些方面。但你更多地使用了客户体验，而不是支持或销售。我想我们在这条路上走到了哪里？我意识到每个客户都会略有不同，支持与销售之间的界限可能有些模糊。但你看到越来越多的人让它从支持解决型转向开始进行一些更偏销售的对话了吗？我们在这条路上走到了哪里？

You've touched on elements of this in how you've described Sierra. But you've used customer experience more than support, more than sales. I guess where are we on that journey? And I realize every customer is going to be slightly different, and the lines probably blend a little bit between what's doing support versus what's doing sales. But are you seeing more and more people letting it shift from being the support resolution thing to starting to have some more sales-oriented conversations? Where are we on that journey?

当然。几乎无一例外，我们的客户希望处理入站客户支持，但希望他们的智能体成为客户体验的前门。我们最早的部署之一是一家鞋业公司。第一次会话是：“我要去夏威夷参加婚礼。什么凉鞋能配我的伴娘裙？”这不是一个客户支持查询。这完全是产品发现、考虑购买。事实证明，当你有一个带有品牌顶端的自由形式 AI 智能体时，你会和它交谈。有时你可能有一个传统的客服问题，但很多时候，你只是像和公司员工交谈那样和它聊天。而且很难对你的 AI 智能体进行细分。客户的意图并不总是显而易见的。所以我们看到的很多客户的模式是，他们可能因为一个问题或想要一个解决方案来找我们，然后很快线索就被拉向了更广泛的客户体验。我认为这非常令人兴奋。我确实相信每家公司的 AI 智能体将和他们的网站或移动应用一样重要。它将涵盖所有这些功能。

Absolutely. Mostly without exception, our customers want to handle inbound customer support but want their agent to be the front door for their customer experience. One of our earliest deployments was for a shoe company. The very first session was: 'I'm going to a wedding in Hawaii. What sandals will go with my bridesmaid dress?' Which is not a customer support query. It's very much a product discovery, a considered purchase. It turns out that when you have a free-form AI agent with a brand at the top, you talk to it. And sometimes you might have a traditional customer service question, but a lot of the time, you're just talking to it the way you talk to an associate who worked for that company. And it's really hard to segment your AI agent. It's not always obvious what the customer's intent is. So the pattern that we've seen for a lot of our customers is they may come to us for one problem or where they want a solution, and then quickly the thread gets pulled to be a broader sense of the customer experience. And I think that's really exciting. I actually really believe that every company's AI agent will be as important as their website or their mobile app. It will come to encompass all of that functionality.

如果你想想一家公司的网站，你可以买东西，了解他们的历史，了解高管团队，如果是上市公司，可能还有投资者关系板块。我认为你的 AI 智能体将能做所有这些事。因为到某个时候你会问，为什么不能呢？为什么我不能问 ADT 智能体任何关于他们 150 年悠久历史的问题？我应该能问。这在我看来相当令人兴奋。

And if you think about a company's website, you can buy things, you can learn about their history, you can learn about the executive team, it's a public company, there's probably an investor relations section. I think your AI agent will do all of that. Because at some point you ask like why not, you know, why can't I ask the ADT agent anything I want to about their storied 150-year-old history? I should be able to. And that's quite exciting in my mind.

所以，如果我们成功实现了使命，当你在现实中遇到一家公司的品牌 AI 智能体时，我们希望它由我们的平台驱动。我们希望它为你个性化，了解你与该品牌的历史互动，真正像礼宾服务一样，能够对你想要采取行动的任何事情采取行动。当你拥有这种体验时，你会觉得“哇，这家公司懂我”。我认为这是现在可以用 AI 大规模做到的事情，而以前根本无法做到。

And so, if we're successful in achieving our mission, when you encounter a company's branded AI agent in the wild, we'd like it to be powered by our platform. And we'd like it to be personalized to you, to know your historical interactions with that brand, to really be like a concierge type experience, to be able to take action on anything you want to take action on. And when you have that experience, for you to feel like, "Wow, this company gets me." And that's something that I think you can do at scale with AI now that you just could not do before.

在我们结束之前，你之前提到的一个观点我很想听听你的看法，那就是你本质上认为自己是一名软件开发者。我想如果五年前，也许三年前，有人问我该进入哪个领域，或者大学该学什么？我会说，“嗯，我不知道。这似乎是当今社会最大的稀缺资源，而且它的价值只会上升。”现在如果那个人回来找我，这是个虚构的人，我不认为我真的有过这样的对话。我会说，“啊，我的错。”现在，我有点偏差了，对吧？但我仍然认为，就像你说的，管理多个智能体的抽象层之类的东西很可能存在。我想也许你给年轻人的建议，我有很多年轻听众，让他们思考自己的职业生涯，以及社会和专业领域的哪些部分会在下一次工业革命中幸存下来——这次革命将以超快速度进行，不是 100 年，而是 10 年左右。我不知道正确的时间跨度是什么，但你对那些试图在不确定的未来中导航的人有什么建议或想法吗？

As we wrap, one of the things you said earlier that I would love your perspective on is identifying as a software developer at your core. And I think if someone had asked me five years ago, maybe three years ago, like what field they should go into or what should they study in college? I would be like, "Well, I mean, I don't know. This seems like the biggest scarcity we have in society today and the value of it only goes up." And now if that person came back to me, this is a fictional person. I don't think I actually had this conversation. I'd be like, "Ah, my bad." Now, like I was a little off there, right? But I still think, I mean to your point, like the abstraction of managing multiple agents and all that stuff very well could exist. I guess as maybe you counsel young people, and I have a lot of younger listeners, to think about their career in some ways and like what pockets of society and professionalism are going to persist this next industrial revolution that we're going to go through in turbo speed instead of 100 years, we're going to go through in 10 or something. I don't know what the right time horizon is obviously, but is there any advice that you give or you think about for people that are kind of trying to navigate their way through an uncertain future?

这可能是我们行业以外的人问我的头号问题，因为很难知道哪些工作会持续存在。首先，我认为很多工作会持续存在。我们围绕我们生产的技术创造经济，而不是反过来。所以我们经历了农业革命、工业革命、全球化，这些都极大地改变了我们的经济。所有这些，如果你看看农民或工厂工人的比例，现在我们更像是服务经济。所以工作都会消失的想法根本说不通。我们会找到新的事情做。你能想象回到一个每天工作 15 小时生产食物的农民吗？不，这个播客之类的东西，让我们在 100 年后再说。没有人再考虑获取食物了。那就像打开电灯开关一样令人惊叹。所以我个人并不担心这一点。但那是社会层面。在个人层面，我认为你的工作可能不会那么频繁地改变，但职业，我们做事的方式，我认为会彻底改变。我认为理解你所在部门提供的价值非常重要。我以前用过会计在 Microsoft Excel 之前和之后的类比。你的工作不是加数字。你之前用 HP 计算器和实际的纸质表格来做。然后当 Microsoft Excel 出现时，会计的算术变成了商品，你最终得到了像数据透视表这样更高杠杆的东西，更少的人做更多的事情。

It's probably the number one question I get from people outside of our industry just because it's so hard to know what jobs will sort of endure. First, I'll say I think lots of jobs will endure. We create an economy around the technology we produce, not the other way around. And so we've gone through the agrarian revolution, the industrial revolution, we've gone through globalization, and it really shifted our economy a lot. All of those, if you just look at the percentage of people who are farmers or worked in factories, who now we're sort of a services economy. So the idea that the jobs are all going away just doesn't make any sense. We find new things. Can you imagine going back to a farmer that worked every day 15 hours a day to produce food? And it's like, no, this podcasting thing actually, let us tell us in 100 years. No one thinks about acquiring food anymore. That's just like turning on a light switch is mind-blowing. So I am not personally worried about that. But that's at a societal level. At an individual level, I think your jobs will probably change less frequently, like vocations, the way we do things I think will change dramatically. I think it's really important to understand the value your department provides. I've used the analogy before of accounting before and after Microsoft Excel. Your job wasn't to add up numbers. You did that with your HP calculator and actual physical spreadsheets before. And then when Microsoft Excel came out, the arithmetic of accounting became a commodity, and you end up with higher leverage things like pivot tables and fewer people doing more things.

这就是我对投资银行的看法。80 年代的银行家每周工作 100 小时，今天的银行家每周也工作 100 小时。

That's what I think about investment banking. Like bankers in the 80s worked 100 hours a week and bankers today work 100 hours a week.

没错。但他们做的事情多得多，模型也更复杂。这是非常不同的事情。

Exactly. But they're doing a lot more and their models are more sophisticated. It's very different things.

所以我喜欢这样想：如果你是 Excel 之前和之后的那个会计，你是那个先学会数据透视表并教给同事的人吗？你的老板会说，“哇，那个人真是个有干劲的人。他们学会了这个神奇的新东西，让他们效率高多了。”如果你是一名软件工程师，这意味着你周末在使用 Codex 和 Claude Code，并积极拥抱这些工具。这就是我会给出的建议：对于那些采用这些工具的人，他们的效率会比同行高 10 倍甚至 100 倍。如果你积极使用这些工具，它们会变化很大。比如一年前的 Cursor 和今天的 Codex、Claude Code 之间的差异是巨大的，对吧？所以我们说的是每月都在变化。如果你这样做，我认为你可以围绕它们建立职业生涯。我不确定它会是什么。它可能是你职业生涯中显而易见的东西，也可能是反直觉的。但我认为，特别是对年轻人来说，这是一种新的超能力，一种你可以获得的超级智能。让它成为你日常生活的一部分，并把它带入你的职业和事业中。然后无论你的职业发生什么，你都会处于最前沿。你是第一个建立网站的公司吗？你是第一个使用 Excel 的会计吗？你是第一个让编码智能体为你工作并产出成果的软件工程师吗？如果你处于那个位置，我认为你处于一个非常好的位置。这既是一种心态，也是一种对职业生涯的保障。

So what I like to think about, and if you were that accountant before and after Excel, are you the one who learned pivot tables first and taught your colleagues at work, and your boss said, "Wow, that person's like a go-getter. They learned this amazing new thing that made them a lot more productive." And if you're a software engineer, that means are you using Codex and Claude Code on the weekends and leaning into the tools. That would be the advice I would give: this is for the people who adopt it, they'll be 10 or 100x more productive than their peers. And if you lean into using the tools, and they'll change a lot. Like the difference between Cursor one year ago and Codex and Claude Code today is stark, right? So we're talking on a monthly basis, these tools are changing. And if you do, I think you can create a career around them. And I'm not sure what it will be. It might be something that's obvious in your career. It might be something that's counterintuitive. But I think that for younger people in particular, this is a new superpower, a superintelligence that you have access to. Make it a part of your daily life and be the one who pulls it into your profession and into your career. And then whatever happens to your profession, you're going to be on the bleeding edge of it. Are you the first company to put up a website? Are you the first accountant to use Excel? Are you the first software engineer who's got coding agents working for yourself and producing your outcome? And if you're in that position, I think you're in a really good position. It's both a mindset and a kind of a safeguard for your career.

关于学什么，我仍然坚信计算机科学。我认为对一些计算机科学系的批评是它们太理论化了。如果你想想复杂性理论、算法如何工作以及分布式系统，我认为其中很多都会让你受益匪浅。如果你不了解 AI 智能体在做什么的背景，就很难告诉它该做什么，或者很容易犯非常严重的错误。所以我认为理解基础可能比以往任何时候都更重要。不过另一件事，也许作为临别想法，我好奇并在某种程度上希望这将有利于通才。

On what to study, I still am a big believer in computer science. I think one of the criticisms of some computer science departments is they've been too theoretical. If you think about complexity theory and how algorithms work and distributed systems, I think a lot of that will really benefit you. It's hard to tell an AI agent what to do if you don't have the context on what it's doing, or it's very easy to make really significant mistakes. So I think it's probably more important than ever to understand the fundamentals. And the other thing though, I just like maybe in a parting thought, I wonder and to some degree hope that this will benefit generalists.

我认为 Patrick 在科学领域写过很多这方面的内容，但随着世界变得越来越复杂，社会整体走向了更细分的专业化。结果就是，一个有好应用创意但编程能力不强的人，必须依赖很多不同的人才能实现它。换句话说，你可能有很好的品味，但如果你没有实际掌握实现这种品味所需的技艺工具，世界可能就无法从中受益。我总会想到 22 岁时的 Christopher Nolan。他拍了《记忆碎片》，但他拍不了《星际穿越》。拍那样的电影需要太多资金了。下一个 Christopher Nolan 能否在没有社会许可的情况下实现他的创意愿景？因为现在做这件事的成本低得多。你不需要成为视觉特效专家，你只需要是一个有伟大艺术视野的人。所以，我真正希望的一点——这跟职业建议有点沾边——就是那些对广泛领域都有较好理解的通才，能够借助 AI 和这项技术创造出非凡的作品。我认为这对社会非常有益。而走向狭窄专业化的趋势——大多数科学、技术、产品和艺术上的重大突破，都来自多个不同专业领域的交叉与融合。

I think Patrick's written a lot about this in the context of science, but the world has gone towards more specialization as the world has become more complex. As a consequence, someone with a great app idea who wasn't a great programmer had to depend on a lot of different people to make it. Put another way, you could have great taste, but if you didn't actually have the tools of the craft to act on that taste, the world might not benefit from it. I always think of Christopher Nolan at 22. He made Memento, but he couldn't have made Interstellar. You just need too much money to make a movie like that. Will the next Christopher Nolan be able to act on his creative vision without societal permission? Because the cost of doing so is so much easier. You don't need to be a visual effects expert. You can just be a person with a great artistic vision. So, one of the things I really hope, a little adjacent to career advice, is that generalists who understand a broad range of things fairly well will be able to leverage AI and this technology to produce exceptional things. I think that's really good for society. The move towards narrow specialization — most great breakthroughs in science, technology, products, and art have been the cross product and intersection of many different domains of expertise.

Brad，感谢你接受采访。

Brad, thank you for doing this.

我的荣幸。

My pleasure.