Jesse Zhang - 打造Decagon - [最佳投资之道，第443期]

顶尖运营者始终专注于杠杆效应，寻求放大影响力的方法而非单纯更努力工作。但我在各财务团队中看到的却是：才华横溢的人深陷费用管理的琐事中。细想之下，人们成为财务领导者正是源于对战略工作的热爱——情景建模、优化资本配置、发掘真正推动业务发展的洞见。

The best operators have a relentless focus on leverage, finding ways to multiply their impact rather than just working harder. But here's what I see happening in finance teams everywhere. Brilliant people getting buried in expense management busy work. If you think about it, you become a finance leader because you love strategic work. Modeling scenarios, optimizing capital allocation, finding the insights that actually move the business forward.

然而现实却是，你们在追索发票和分类交易。这与杠杆效应背道而驰。这正是我对Ramp团队成果如此看好的原因。Karim和Eric明白，每一分钟的手动费用管理都是在窃取高杠杆工作的宝贵时间。所以他们实现了全流程自动化。

But instead, you're chasing receipts and categorizing transactions. It's the opposite of leverage. This is exactly why I'm so bullish on what the team at Ramp has built. Karim and Eric understood that every minute spent on manual expense management is a minute stolen from high leverage work. So they automated all of it.

自动分类、收据匹配、真正有效的支出管控。我钟爱由此产生的网络效应：当Shopify和Stripe等公司的财务团队将琐事自动化后，他们就能腾出精力思考更宏大的命题，提出更关键的问题，发现他人忽视的模式，做出区分卓越公司与普通公司的战略决策。这笔账很简单：夺回你的时间，聚焦要事。

Automatic categorization, receipt matching, spending controls that actually work. I love the network effect that this creates. When finance teams at companies like Shopify and Stripe automate the mundane stuff, they free up cycles to think bigger, to ask bigger questions, spot patterns others miss, and make the kind of strategic bets that separate great companies from good ones. The math is simple. Get your time back, focus on what matters.

访问ramp.com/invest，见证消除琐务后的蜕变。卡片由FDIC成员Sutton银行发行。适用条款与条件。在资产管理领域，增长往往取决于定制化服务。这是我们行业的本质，作为亲历过这个问题的主动型基金经理，我深知根据客户偏好量身定制产品服务才是关键竞争力。

Check out ramp.com/invest and see what happens when you eliminate the busy work. Cards issued by Sutton Bank, member FDIC. Terms and conditions apply. In asset management, growth often depends on customization. It's the nature of the beast in our industry, and I know having experienced the problem firsthand as an active manager, it's a competitive differentiator to tailor products and services to clients' preferences.

我们这些致力于业务增长的人总想对客户有求必应。这意味着提供定制化的投资组合、量身打造的报告或个性化的服务预期。应允带来增长，同时也带来定制化和巨大的权衡。规模越大，吸收的复杂性就越多；答应得越多，高效且持续地扩展就越困难。

Those of us growing our businesses always want to say yes to customers. It means delivering a tailored portfolio, a tailored report, or a tailored expectation for service. Saying yes leads to growth, and it also leads to customization and a big trade off. The more you grow, the more complexity you absorb. The more you say yes, the harder it is to scale efficiently and consistently.

这正是Ridgeline的用武之地。Ridgeline实现了定制化流程的自动化。它让资产管理公司能够大规模提供个性化服务，而无需增加人手、人工操作或运营风险。作为早期设计合作伙伴之一，我亲眼见证了从零开始构建这套我们期待已久的系统的力量——一个前后端整合的平台，将公司所有核心功能集成在单一数据集上。

That's where Ridgeline comes in. Ridgeline automates customization. It gives asset managers the ability to deliver personalized experiences at scale without adding headcount, manual work, or operational risk. Having been an early design partner myself, I saw firsthand the power of taking an entirely clean sheet of paper to building the system we've all been waiting for. A front to back platform that combines all of a firm's core functions on a single dataset.

领先企业正是借此不再在增长与效率间取舍，而是开始两者兼得。我相信最优秀的企业将以Ridgeline作为其操作系统来构建。我也预见这里将诞生记录系统与人工智能结合的典范案例。若您尚未深入了解，我强烈建议您探索Ridgeline能为业务带来的可能性。本节目长期听众都知道，AlphaSense是我多年来推崇的市场情报平台。

It's how leading firms stop choosing between growth and efficiency and start saying yes to both. I believe the best firms will be built on Ridgeline as their operating system. I also believe there'll be a leading case study in combining the power of systems of record and AI. If you haven't spent time with them yet, I urge you to see what Ridgeline might unlock for your business. Longtime listeners of this show will know that AlphaSense is the market intelligence platform I've admired for years.

它为机构投资者提供了访问超过5亿份优质资源的途径，涵盖公司文件、经纪商研究报告、新闻、行业期刊等。此外还包括20多万次专家访谈，覆盖全球最重要的公司和行业，所有这些都集中在一个平台上，让投资团队能够更快行动、更深入分析，并充满信心地做出高确信度决策。我很兴奋能参加AlphaSense于今年十月在布鲁克林举办的首届Alpha Summit 2025。我将与来自瑞银、富国银行、埃森哲、谷歌、Stripe集团、凯雷集团等机构的领袖同台，探讨人工智能如何重塑投资研究与决策。Alpha Summit旨在展示顶尖公司当前使用的真实工作流程与策略。

It gives institutional investors access to over 500,000,000 premium sources from company filings and broker research to news, trade journals, and more. Plus over 200,000 expert calls covering the world's most important companies and industries, all of it in one platform so investment teams can move faster, go deeper, and make high conviction decisions with confidence. I'm excited to join AlphaSense at their inaugural Alpha Summit twenty twenty five this October in Brooklyn. I'll be on stage alongside leaders from UBS, Wells Fargo, Accenture, Google, Stripe's group, the Carlyle Group, more to talk about how AI is reshaping investment research and decision making. Alpha Summit is about showing the real workflows and strategies that top firms are using today.

这场活动为期三天，汇聚了行业顶尖的主题演讲嘉宾阵容。您将聆听这些行业领袖的见解，与金融和企业战略领域的同行建立联系，参与其他地方无法听到的对话。请加入我参加2025年10月在多米诺炼油厂举行的Alpha Summit。如需注册并查看完整演讲嘉宾名单及议程，请访问alphasense.com/invest。

The event features an incredible lineup of industry leading keynote speakers over three days. You'll hear from these industry leaders, connect with peers across finance and corporate strategy, and be part of the conversations you won't find elsewhere. Join me at Alpha Summit twenty twenty five, October, at The Refinery At Domino. To register and to see a complete list of speakers and the full agenda, go to alphasense.com/invest.

大家好，欢迎收听。我是帕特里克·奥肖内西，这里是《像这样投资》节目。本节目是对市场、观点、故事和策略的开放式探索，旨在帮助您更好地投资时间和金钱。如果您喜欢这些对话并想深入了解，请查看我们的季刊《巨像评论》，其中深入剖析了塑造商业和投资格局的人物。您可以在joincolossus.com找到《巨像评论》及我们所有播客内容。

Hello, and welcome, everyone. I'm Patrick O'Shaughnessy, and this is Invest Like the This show is an open ended exploration of markets, ideas, stories, and strategies that will help you better invest both your time and your money. If you enjoy these conversations and wanna go deeper, check out Colossus Review, our quarterly publication with in-depth profiles of the people shaping business and investing. You can find Colossus Review along with all of our podcasts at joincolossus.com.

帕特里克·奥肖内西是Positive Sum的首席执行官。帕特里克及播客嘉宾表达的所有观点仅代表其个人意见，不代表Positive Sum的立场。本播客仅用于提供信息，不应作为投资决策依据。Positive Sum的客户可能持有本播客讨论的证券头寸。了解更多信息，请访问psum.vc。

Patrick O'Shaughnessy is the CEO of Positive Sum. All opinions expressed by Patrick and podcast guests are solely their own opinions and do not reflect the opinion of Positive Sum. This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. Clients of Positive Sum may maintain positions in the securities discussed in this podcast. To learn more, visit psum.vc.

今天的嘉宾是杰西·张。杰西是增长最快的AI客服公司之一Decagon的联合创始人兼CEO。Decagon提供集中式AI引擎，可随时以各种语言、跨所有渠道自动解决问题。杰西分享了通过直接询问潜在客户愿意为解决方案支付多少费用，来系统化寻找产品市场契合度的方法。我们探讨了为何客服与编程成为企业最明确的两大AI应用场景，以及推动Decagon发展的关键商业与技术因素。

My guest today is Jesse Zhang. Jesse is the co founder and CEO of Decagon, one of the fastest growing AI customer service companies. Decagon provides a centralized AI engine to auto resolve issues at any time, in every language, and across every channel. Jesse shares his systematic approach to finding product market fit by asking potential customers exactly how much they'd pay for solutions. We explore why customer service and coding have emerged as two of the clearest AI use cases for enterprise, and the key business and technical factors behind Decagon's momentum.

我们还讨论了当前构建AI的激烈竞争态势，关于建立专有模型的战略决策，以及在企业规模部署AI代理的实践。请欣赏这段与杰西·张的精彩对话。

We also discuss the intense competitive dynamics of building an AI today, strategic decisions around building proprietary models, and deploying AI agents at enterprise scale. Please enjoy this great conversation with Jesse Zhang.

或许可以从一个有趣的话题开始——我猜你没想到——请你谈谈办公室里墙上写的那句话，你之前提到过它。

Perhaps an interesting place to begin, I bet you don't expect this, is for you to tell us a little bit about the phrase that you've talked about that's written on your wall in the office.

没错。确实如此。在我们旧金山办公室的墙上，纽约办公室最近也挂上了同样的一句话。大意是：没有克服不了的挑战，也没有击败不了的敌人。我们非常喜欢这句话。

Yeah. For sure. So in our wall, in the SF office, and we just did it in the New York office as well, we have this quote. It basically goes along the lines of, there's no challenge that can't be overcome and there's no enemy that can't be defeated. And we just like it.

我认为它完美契合我们的文化。我们拥有一支极具竞争力的团队，每个人都渴望胜利。当我们全力投入建设时，整个团队充满能量——因为感觉就像，你知道的，这个行业正在发生巨变，而我们拥有如此强大的团队，只管去赢就好了。这个理念其实源自我父亲的教诲，虽然我不确定华为在中国是否真的验证过这一点。

I think it really fits our culture. People there are we have a very competitive team. Everyone wants to win. We have a lot of energy when we're trying to go out and build because it just feels like, you know, there's all this stuff happening in the industry, like, we have such a strong team, let's just go and win. Motivated by my dad told me this, I don't know if this is actually validated or not at Huawei in in China.

华为显然是家巨头企业，但他们以狼性文化著称。他们用中文将类似理念写成标语，用鲜红的大字悬挂在总部大厅后方。每次有人走进大厅，这些字都会像海岸线般醒目地跃入眼帘，非常有意思。

It's obviously a massive company, but they're known for just having, like, killer culture. And they have some version of this written in Chinese, of course, in just big red letters across the back of the big hall. So, yeah, really like that. It's kinda interesting. Like, any time that someone walks in, it it, like, stands out, the coast stands out.

我最初询问这个是因为，在这个可能是我们此生经历的最激动人心的技术时代，面对强大的竞争对手，像你们这样建设公司究竟是什么样的环境。像'击败'这样的词汇，最近我还听到有人用'暴力'形容企业文化，还有'侵略性'——这些词在三、四年前根本不会出现，当时使用这些词甚至会带来大麻烦。显然现在情况完全改变了。

I asked about it to begin because I'd love to spend some time on just what this environment is like building a company like yours against formidable competitors in probably the most exciting era of technology that any of us will ever live through. And words like this, defeated, I've heard the word violence recently about a company culture, aggression. Like these are not words that were being used three years ago or four years ago. In fact, if you use them, it was a big problem for you. And obviously that has completely shifted.

不仅创始人开始使用这些词汇，我认为顶尖人才和高管们也对此趋之若鹜。他们渴望加入能'击败敌人'、具有'暴力'或'侵略性'特质的文化。或许你可以展开讲讲，在当前这个与几年前截然不同的时代，在对话式AI等重要领域进行建设和竞争是怎样的体验？

And not only have founders started using these words, I think talent and senior people like rallied around them. Like they want to be in a culture that is defeating enemies or violent or aggressive. Maybe just riff for a while on what it's like to be building and competing in one of the big areas in your case, conversational AI, etcetera, at this moment in time because it's just so different than it was a couple years ago.

我们的核心观点是：任何值得进入的领域，任何热门大市场，必然充满竞争。这不局限于AI领域——想想Databricks、Snowflake或Ram for Sprex，每当出现重大增长机遇，人们自然会理性地涌入。所以这代人吸引的，是特定类型的建设者和创始人。虽然机遇令人兴奋，但所有人都清楚竞争激烈：如今创业门槛低、融资容易，你必须深刻理解自己的竞争优势，而企业文化可能就是其中之一。优秀文化既难以复制又持久传承——正如你所说，当整个团队都专注于成功、努力工作和保持某种强度时，这种文化会产生深远影响。

Our whole view is that any space that's worth going after, like any large hot market, it's gonna be competitive. This is not really specific to AI, right, if you think about Databricks or Snowflake or Ram for Sprex and anytime there's these big massive growth opportunities, people rationally wanna go after it. So I I think in this generation, it's just I think it almost attracts like a specific demographic of founder and people that want to build because it is exciting, but of course, everyone knows it is competitive because if there is market and everyone's trying to build it, it's quite easy to start a company these days, you can raise funding really easily and so when you go out there, you have to at a certain point like have a pretty deep understanding of like what your competitive advantages are and one of those could be the culture. And if you build a good culture, that's pretty hard to replicate, it also lasts quite a long time. And to your point, if you have a culture where everyone is really geared towards succeeding and working hard and having some level of intensity, it can go a long way.

关于很多公司采纳相似思维这点，根据我的观察，这代创业者确实吸引了我这类人群——在高度竞争环境中成长的人，无论是学术还是其他领域。很多创始人从小参加数学竞赛、编程比赛并表现出色。如今我同龄人中创业的很多都发展不错。我认为如果你从小就适应这种硬核生活方式或环境，确实能很好地适应当前局面，因为两者存在诸多相似之处。

So I think to your point of a lot of companies adopting similar mindsets, I do think from my observation, this generation of company building has attracted like a almost like my demographic of person where it's just like people that grew up in fairly competitive environments, academics, or whatever. And a lot of these founders have done well. Did a lot of math contests and coding contests growing up. A lot of the people around my age are doing startups now and people are doing quite well. And I think there is some element of if you really embrace this hardcore lifestyle or environment growing up, then, yeah, I think it lends itself pretty well to the current situation because there's a lot of parallels.

我最近和Cognition的Scott Wu在一起，他众所周知是数学冠军之一，你也是。谈谈那个环境。那种竞争市场是怎样的？那是什么样的？某种程度上让人们进入那个世界，因为显然，它对你有很大影响。

I was with Scott Wu from Cognition recently who is well known to be, like, one of these math champions, you as well. Talk about that environment. What was that competitive market like? What was it? Kinda let people into that world because, obviously, it shaped you a lot.

可以说是一种有趣的经历。我想，当我们现在都长大了回头看，那是一种相当激烈的成长方式。但我对童年有着非常美好的回忆。你得到了一个很好的社区，很多人在做同样的事情。我在科罗拉多州的博尔德长大。

It's kind of like a interesting experience, I would say. Like, I think when you look back now that we're all grown up, it is kind of like a pretty intense way to grow up. But I look back on my childhood with very fond memories. You get a really nice community, a lot of people that are doing the same thing. I grew up in Colorado in Boulder.

博尔德是一个学术氛围浓厚的城镇，但并没有很多人热衷于这些竞赛。我的很多朋友都来自你意想不到的地方，比如加利福尼亚、德克萨斯、纽约。所以它给了你某种社区感，好吧，外面有很多人和你做同样的事情，并结识他们中的许多人。现在我们长大了，这些关系持续了很长时间。但没错，那种环境有点像创业。

Boulder's a pretty academic town, but there's not very many people that are like super gunning for like these contests. A lot of my friends had just kind of grew up in not the places you would think, right, California, Texas, New York. And so it gives you some level of community of like, okay, there's yeah, there's lot of other people out there that are doing the same things as you and meet a lot of them. And then now that we're all grown up, those relationships have lasted for a long time. But yeah, then the environment is one where it's like similar to company building.

你的表现如何是相当客观的。需要投入很多，有很多准备，感觉像是一项长期的事情，但因为结果相当客观，所以有持续改进的动力。我觉得这很好。我的一个重要观点是，这群人或这个社区的人有很多未开发的潜力，实际上可以引导他们成为想从事商业或公司等事情的人。

You're like, how well you're doing is like fairly objective. There's a lot that has to go into it. There's a lot of preparation, a lot of having the feeling of it is like a long term thing, but because your results are fairly objective, there's constant motivation to improve. I think that's quite nice. One of my big thesis is that this audience of people or this sort of community of people, there's a lot of untapped potential there and to actually like turning them into people that wanna do business or companies and things like that.

他们中的许多人历史上进入了交易或学术界。当然，这些都是非常棒的工作。这里的很多人，这种背景与一种更倾向于规避风险、只追求好成绩和按部就班的轨迹相关。如果你能将其中一些人才引导到创业领域，我认为可以做很多事情。

A lot of them have historically gone into trading or academia. Now I mean, are perfectly awesome jobs. A lot of the folks here, this background is correlated to one where it's a little bit more like risk averse and kind of just get your good grades and like follow a track. And if you can kind of diverse some of that talent into like company building, I think there's just a lot that can be done.

有什么相似之处？是什么让这些人、这种训练或竞赛中的特定能力让你或其他人擅长创业？这似乎是一个显而易见的趋势，现在有足够多的样本表明，有这种背景的人在这种环境中表现得非常出色。也许这是最好的。如果你能以某种方式索引这群人，你现在会有惊人的表现。

What are the parallels? Like, what is it about the people or the training or the specific aptitude in the competitions that make you good at company building or and others good at company building? Like, this seems like an obvious true trend that there's enough sample size now of people that had this background that are doing extremely well in this environment. Maybe it's the best. If you could somehow index that group of people, you'd have, like, fantastic performance right now.

是什么？是什么相似之处让这成为现实？

What is it? Like, what are the parallels that make that true?

其一是我们刚讨论过的竞争性，其二则是解决问题的能力。我深信的一点是，在创业或做任何事情时，最有益的做法就是清楚认识自己的优势所在。对这个群体而言，优势更多体现在解决问题上——即便问题可能非常模糊，比如‘如何创立一家成功的公司？’

One is the competitive nature that we just talked about. The other is just like the problem solving. So I think one thing that I believe very strongly is that one of the best things you can do when you're building a company or anything is just have pretty good introspection on where your strengths are. And at least for this group, it's more on the problem solving side just like and the problem can be very vague. Problem could be like, how do I build a successful company?

你需要拆解问题，用批判性思维从第一性原理分析这些市场。当然这不是创业的唯一方式，有些人天生对PLG（产品驱动增长）类事务有敏锐直觉。但总体而言，这些竞赛真正教会你的是解决问题的能力。虽然竞赛题目并非现实难题，但思维方式是相通的。

And you kind of break down the problem, you can like think really critically and just from first principles a lot of these markets. Of course, that's not the only way to build a company. Other people I think some people just have like very good intuition, especially on these like PLG type things. But in general, I think just the problem solving capability, like what these contests really teach you is like you're just solving problems. And those problems of course are not real problems in real life, but the sort of thinking is the same.

现在的创业过程比参赛时更情绪化吗？情绪化？

Does it feel more emotional now company building than it did in the contests? Emotional?

可能因为年长几岁，现在反而没那么情绪化。此前创立的公司比现在艰难得多，这段经历让我心态更平和，也更懂得珍惜顺境。时机至关重要——当初我提前一年毕业创业，虽然结果幸运，但整个过程跌宕起伏。

I think I'm a bit older now, so it's not really that emotional. I started a company before this that was, I would say, way tougher than the current one. I think just like mentally makes you feel a lot more calm and appreciative of when things are going well. And timing matters a lot. I think when I first started, I graduated a year early to try to start it and I would say we had a very fortunate outcome, but the whole journey was like very bumpy.

那是家什么公司？做什么业务？

What was it? What did it do?

公司叫Low Key，主要开发游戏高性能视频捕捉软件。玩家可以轻松录制、编辑并分享游戏片段。产品目标就是最大化用户规模。我们在2021年适时退出，这个时机非常幸运。

So the company was called Low Key, we basically built high performance video capture software for video games. So when people are playing games, you can really easily capture video clips and edit them and share them. The whole goal of the product is just like you just get as many users as possible. And again, I think we exited at like a very fortunate time. It was 2021.

但整个历程充满坎坷：初期根本不清楚要做什么，尝试各种方向。作为应届生缺乏判断力，埋头苦干三个月后才发现市场根本不成立。约一年半后，两位来自学校的联合创始人——也是我的挚友——因过度 burnout 选择离开。

But throughout the journey, like, at the beginning, I didn't really know what we were building and then trying a bunch of things. I mean, you're like a new grad, you don't really have good intuition, like what idea is good or not. So you just worked really hard for, you know, three months and so then you realize that obviously you had no market. So that's tough. Then about a year and a half in, I had two of my good friends from school, one my co founders, they both got burnt out and then decided to leave.

所以之后就只有我一个人在琢磨该怎么办。我认为那比我们现在做的任何事情都要艰难得多。就像你说的那种情感上的挑战，你根本不知道未来会怎样，压力巨大，因为你不想让自己投入的第一件事就以失败告终。所以从心理层面来看，那段时期要艰难得多。而现在，我觉得困难主要在于工作量实在太大。

So was just me after that trying to figure out what to do. And so that's I think when you I think that's way tougher than anything we're doing right now. It's like a the way you put it like emotional thing, it's you don't really know if there's any future or not, there's a lot of pressure because you don't want like the first thing you work on to be a failure essentially. So I think that was much tougher from like a psychological point of view. Nowadays, I think it's tough just just like a sheer amount of stuff to do.

我们睡眠严重不足，但从大局来看，你必须心怀感激——能处于这样的位置，有做不完的工作，解决不完的有趣问题，还有每天一起共事的优秀团队。

We're not getting much sleep, but in the grand scheme of things, you gotta be really grateful even be in this position to have enough work to do, to have enough interesting problems, to have a team that you're really excited to work with every day.

你估计自己每天睡多久？

How much do you think you're sleeping?

这个真的说不准。这周在纽约大概就四五个小时吧。这样很不好，因为我这个人睡眠不足八小时脑子就转不动。

I mean, it really varies. This week in New York, it's probably like four to five hours, I'd say. And it's not good because like I'm not someone whose brain functions that well on less than eight hours.

如果对比你第一次和第二次创业，现在这家公司运作极好，发展速度惊人。在Decagon初创期，你从之前经历中吸取了哪些经验来推动这次成功？

So if you think about the difference between the first and the second company, like obviously now you have a company that's working extremely well, it's growing really really fast. What did you do at the beginning of Decagon that was informed by your prior experience to make this one go better?

我觉得创业公司都这样。第一阶段绝对是最难的，因为你要摸索方向。而确定方向本身就极具挑战性——这不像有个明确目标只要埋头苦干就能达成。寻找方向更像是探索未知领域。现在我和联合创始人Ashwin（他非常出色，也有相似背景）在这方面已经相当擅长了。

I think this is probably true of starting companies. I think the first stage is like by far the hardest because you're kind of just like finding direction. And finding direction is very difficult because by definition, it's not something where you can just like, you have like a goal and you're just like grinding towards it and you can get there. Your goal could be, yeah, like, by finding a direction, but it's more exploratory. So I think what we actually are quite good at now, and Ashwin, my cofounder, who's amazing, he also has similar background.

他之前创办的公司被收购了。首次创业时大家都会经历同样的困境——寻找方向举步维艰。所以这次我们规划得更加周密。归根结底还是要发挥自身优势：我们擅长解决问题，理性分析，执行力强。

He started a company before that got acquired. I think when you start a company at first time, you often share the same experiences, which is like finding directions very difficult. So I think this time we were a lot more thoughtful about it. And again, it goes back to what your strengths are. I think we view our strengths as like, hey, we're very good at problem solving, we're very just rational about things, we're just good at like execution.

既然如此，我认为我们应该尝试将创意过程系统化。归根结底，无论你从事什么工作，都必须确保人们愿意真正投入其中。如何判断这一点？你可以通过深入询问来了解，而人们通常会对这类深入问题感到些许尴尬或不自在。但当你与潜在客户交谈时，他们其实并不介意回答诸如'如果我们为你打造这个产品，你具体愿意支付多少钱？'这样的问题。

And so if that's the case, then I think we just try to systematize the ideation process. And it just comes down to, okay, know, whatever you work on, it has to be something that people will really invest in. And how do you tell if that's the case? You can just go really deep asking them and I think people are usually a little bit almost embarrassed or not comfortable going super deep in these questions. When you talk to a potential customer, they actually don't mind answering questions such as, okay, if we built this for you, like exactly how much would you pay for it?

比如，是否需要你的直属上司或更高层批准？整个组织会如何考量投资回报率？你会如何向领导层展示投资回报率以保护自己并赢得认可？我认为如果真正深入探讨这些问题，本质上就像在创始人阶段提出经典的销售资质审查问题——而由于你创始人的身份，这种深入探讨反而显得不那么像推销。

Like, would your boss need to approve it or your boss's boss? Like, who needs to approve it? How would the entire organization think about ROI? How would you present ROI to leadership to protect yourselves and also like make you look good? And I think if you really go deep there, it's almost like you're basically asking like classic sales qualification questions when in founder form and because you're a founder, it's like it just feels a lot less sales y for you to go deeper.

这个过程能让你获得更多有效信号。我们起步时很幸运能接触到许多大型企业（主要是数字原生企业），通过不断追问这些问题并深入挖掘。当时我们持有多元创意，不受限于任何特定想法，保持着开放探索态度——从数据分析到网络安全，从售前支持到运营优化都有涉猎。这个过程极具价值，它表明与客户的交流能带来的有效信号远超'打造客户想要的产品'这类泛泛建议。

That process is what gives you a lot more signal. When we first started, we fortunately were able to get in front of a lot of large companies, mostly digital native ones. And we just asked these questions and we kept digging in. We had a bunch of different ideas, At the time, we're not tied to any idea and kind of open exploration, we're looking at things ranging from like data analysis to like security to, you know, pre sales to ops stuff and that process was very helpful. It just shows you that there's a lot more signal to gain than just talking to customers, which is the general advice or building something that customers want, building something that people want.

确实，这个原则没错，但具体落实却很难。客户会告诉你他们想要什么，但往往那些需求并不实用。

Like, yes, that is true, but it's very hard to just know that. Yeah. Like they they'll just tell you what they want, but then it turns out that that's not useful.

具体操作流程是怎样的？你会向同一个人同时询问多个创意构想，还是针对性地向不同人探讨不同想法？

What was the literal process? Would you go to a single person and ask them about multiple of your ideas at once, or would you target it more like one idea to one person?

必须根据对方的职责范围来定位。如果是高层管理者（比如首席运营官），你可以探讨多个用例；如果是特定领域的副总裁，则应该聚焦单一用例。

You have to target it based on what that person owns. If that person's very senior, you can ask about multiple. So if you talk to like a COO, for example, you can talk about a bunch of different use cases. That gives you some signal too, and then if you're talking to more of like a VP of a certain area, you're probably focusing on one use case.

能否详细还原其中一次对话？这样其他人或许能从中受益。你如何安排提问顺序？怎样构建对话框架才能获取最大信息量？

So maybe go in detail through one of these conversations so that others maybe could benefit from what you've learned. So what is the order of the questions? Like how would you structure those conversations to get the most information possible?

那么，假设我们进入通话，你可以先从高层次的需求挖掘开始，比如：你们目前正在进行哪些项目？你的时间主要花在哪里？现阶段什么让你感到压力等等？这样你就能大致了解用例类型。然后很快就能即时形成假设，比如：什么样的产品在这里是合理的？

So, yeah, let's say we get into the call and you start by just doing very high level discovery like, hey, what are the sorts of projects that are ongoing right now? How do you spend your time? What is kind of like stressful for you right now, etcetera? And then you can kind of get a sense for the types of use cases. And then very quickly, you can just form hypotheses like literally on the fly of like, okay, what would a product be that makes sense here?

接着你可以解释说：如果有个AI助手能处理x y z，会有帮助吗？对方很可能会说'有'，因为通话中人们总觉得欠你一个积极回应。他们会说'对对，那太棒了'——这样你就初步验证了产品概念的可行性。

And so then you're kind of explaining like, okay, yeah, so what if something like an AI agent could do x y z? Would that be helpful? And most likely they will say yes, because there's this thing that happens where if someone's on a call with you, they almost feel like they owe you like A positive you can take away. So they're like, oh, yeah, yeah, that'd be great. Now you've kind of solidified at least the potential product ideas.

他们可能会稍作调整，比如'不，其实应该这样运作'。我记得和很多运营负责人聊过，比如Rippling的Matt McGinnis是Decagon的好友，他详细讲述过团队面临的各类问题——真的多到数不清。

They might adjust it a little bit. They'll be like, yeah, no. Actually, no. It should work like this and so on. I remember we talked to a lot of ops leaders like Matt McGinnis from Rippling is like a great friend of Decagon and telling us about all the different things that have it on his team because there's like so many.

其他接触的运营负责人也类似，比如Oura Ring等不同规模公司。深入交流后会问：'你愿意为此支付多少钱？'这能迫使他们思考，因为人们在讨论创意时很少考虑成本。一旦涉及付费问题，就能帮助评估市场规模。

And similar with other ops leaders we talked to, it was kind of a range of companies as well, people like Oura Ring and so on. And you get down to it and you're like, okay, great. Now we have these use cases, how much would you pay for it? And that kind of forces them to think because most people are not thinking about that as they're talking about ideas, there's like, oh yeah, this would be cool. As soon as you force them to think about how much you pay for something, it kind of is a forcing function for finding some order of magnitude, some level of scale.

这时他们可能说：'如果有五个全职做这事，AI若能做好，或许能减少一个人力，这样我们愿意付2万美元年费'。你心里就有数了——2万当然不算高，但如果能快速复制多个同类产品...

And so then they're like, okay, well, yeah, you have five people doing this full time. If the AI agent can do this well, maybe we'd be able to get rid of one of them, assign one other one, and then, you know, as a result, I'd pay you like, you know, 20 k a year or something like that. And then in your head, you're like, okay, cool. So now at least you have a general order of magnitude. 20 k, of course, isn't amazing.

但现实是优质创意很少。多数情况下做完这个练习你会庆幸：'还好没继续深入'。因为最终可能只是每月100美元的订阅，即便对大公司而言。

But if it's like, I could quickly churn out a ton of these, like, maybe that's interesting. But most of the time, that's not the case because there's not that many good ideas out there, honestly. Most of the time at the end of this exercise, you're like, okay, great. I'm glad I didn't pursue this further, you know, because Yeah. Yeah.

这个练习的妙处在于让客户与你思维同步，他们会透露更多信息。我们公司实际经历就是：讨论用例时对方突然说'对了，我们有个500人的支持团队，那里机会更大'——这时你就能继续深挖。

Like, that would have been a waste of time and at the end, it's like people are paying you like a $100 subscription per month, and it's like a big company. So you just do this exercise, and the nice thing about this exercise as well is that it puts the customer in the same frame of mind as you, and so then they can tell you other things. Essentially what happened with our company is like we were talking about all these use cases and they were like, okay, great. Yeah, if you did this, you know, with five people over here, but by the way, we have a 500 person support organization and there's a lot of opportunity there and we'd be like, okay, great, tell us more, right? And then you kind of dig into it.

我认为这恰恰说明作为创始人，你必须建立自己的信念，并亲自完成这个过程。因为当时，几乎所有告诉我们的人，包括我们认识的那些非常聪明的资深创始人在内，都说这个用例超级明显。很可能已经有现有公司正在将其附加到产品上。正因为它如此明显，现在还没有人做得很大，或者可能已经有人领先了，这其中必有原因。但事实上没人真正知道。

And I think this just goes to show that as a founder, you kind of have to build your own conviction and kind of do this process yourself. Because at the time, essentially what everyone told us, including like very smart older founders that we knew and so on was that this use case like super obvious. There's probably just gonna be incumbents that are just tacking onto the product. And because it's so obvious, there's probably a reason why like no one's super big right now or there's gonna be someone that's ahead. But no one really knows.

即使现在，当其他创始人跟我讨论其他领域时，我虽然有自己的看法，但并不真正了解那些领域的细节。唯一能真正了解的方法就是通过与客户交流获取信号。事后看来，在任何浪潮中，真正的好点子总是极少数。而你的工作就是在正确的时间找到其中一个。所以从定义上来说，这必然是不太明显且相当困难的。

And even for me right now and other founders talk to me about other spaces, it's like, yeah, like I have my own opinions, but I don't really know the details of the space. And the only way you can really know is by talking to customers and getting that signal. And in hindsight, it turns out that in any sort of wave at any time, I would say, like a very small number of good ideas. And your job is to ideally find one of those at the right time. And so, yeah, by the definition, it's gonna be like pretty non obvious and pretty difficult.

因此，如果你能很好地完成这个过程，它将给你最明确的信号。

And so if you can do this process well, it'll give you the most signal.

你还记得在这些创意讨论中，有人为某个东西愿意支付的最高金额是多少吗？

Do you remember the highest number anyone said for how much they'd be willing to pay for something in one of these ideation sessions?

记得。金额大概在六位数的低到中段范围。

Yeah. So the time is probably on the order of low to mid 6 figures.

当你们接近流程尾声并确定Decagon的业务方向时——也许现在正是请你详细描述它的好时机——最后的决定性时刻是怎样的？那种'经过这番探索后，这显然就是我们要做的事'的信念是从何而来的？

As you neared the end of that process and settled on what Decagon does, which maybe probably is the right time to ask you to describe in detail what it is, what was like the final closing? Like, where did the conviction come from? Like, oh, this is clearly the thing after this discovery process?

它之所以显而易见，是因为即便只是把人们为每个创意愿意支付的金额相加，这个数字也比其他想法高出一个数量级。具体来说，Decagon提供的是AI客服代理服务，这是最简单的理解方式。我们打造的对话式AI几乎可以成为品牌的前端接口——无论是用户想联系品牌，还是品牌想联系用户，都能通过这些对话来实现。

It was clearly the thing because if you just kind of tallied up even just the amounts that people said and added them together per idea, this was probably like an order of magnitude more than anything else. So very specifically, what Decagon does is it's AI customer service agent. So that's the simplest way to think about it. And you're building a conversational AI that can just be almost like a front end for a brand. Anytime someone wants to talk to the brand or anytime the brand wants to talk to them, you can kind of initiate these conversations.

当然，长期来看这并不局限于客户服务，但我想再次强调，通过这次实践，我们认为客户服务是最紧迫的需求。事后看来，我可以分析我们为何如此认为，但基本上这就是我们的感受。让我们坚定信心的是，嘿，我们有一大群人排队等着。他们甚至愿意为一个随机的两人团队投入六位数的资金，尽管他们并不十分了解我们，因为这确实是他们最关心的项目。

And of course, long term this is not specific to customer service, but I think, again, going back to this exercise, customer service is where we felt like the most urgent need. In hindsight, I can dissect why we think that is, but that's basically what we felt. And what gave us conviction is that, hey, we had all these folks that were lined up. They were like very willing to invest 6 figures in a random two person team. They didn't even know that well because it was actually like a very top of mind initiative for them.

其他所有事情都像是在挣扎，比如‘你要为此付多少钱？’我不知道，这确实很令人兴奋，但你知道，我们现在的预算很紧，而且也很难衡量效果如何等等。所以，是的，这给了我们足够的信心。然后你就一步一步地推进下去。

Everything else, was just like a struggle to it's like, how much you pay for that? I don't know, like, this is like really exciting, but, you know, our budgets are tight right now and also, like, it'd be hard to measure how well this is doing and so on. So, yeah, that gave us enough conviction. And then you just kind of take a step by step from there.

再详细说说你刚才提到的‘事后看来，为什么这是关键问题’这一点？

Say more about the comment you made about in hindsight, it's clear why this was the key problem?

是的，我认为客户服务具有一系列难以事先推理的优良特性，这也是为什么我如此强调这种以客户为中心的探索过程。其中一个特性是，投资回报率（ROI）在内部很容易证明。这些数字已经被追踪了。比如，嘿，我们有这么多的对话量。现在，我们有一个简单的聊天机器人或一个简单的IVR电话树，可以说解决了其中的15%到20%。

Yeah, so I would say customer service has a bunch of nice properties that I think are very hard to reason through ahead of time, which is why I think I feel so strongly about this process of discovery, just really staying customer centric. One of the properties is that the ROI is really easy to justify internally. You have these numbers already tracked. It's like, hey, we have so much conversation volume. Right now, we have a simple chatbot or a simple IVR phone tree, it's resolving so to speak, like 15 to 20% of that.

如果你能把这个比例提高到50%、60%、70%、80%，那将是巨大的投资回报，而且很容易量化。就像，好吧，我要把总成本削减60%，这就是我节省的部分。另一个我认为有点被低估的特性是，它很容易上线。我认为现在很多生成式AI的应用案例都在为此挣扎，尤其是在企业层面，因为存在风险。人们不希望在他们发布产品时感觉可能会出问题。

If you're able to take that to fifty, sixty, seventy, eighty, like that's huge ROI and it's very easy to quantify. It's like, okay, well I'm gonna take the total cost and then chop off 60% of it and that's what I'm saving. The other property, which I think is a little underrated, is that it's very easy to go live. I think a lot of Gen AI use cases are struggling with that right now, especially at the enterprise level because there's risk involved. People don't want to feel like something could go wrong for them when they release your product.

领导会因此对他们发火。就像，你知道，你为什么要这么做？所以这对NAI来说是个大问题。我认为这也是很多应用案例难以真正起飞的原因之一。因为归根结底，总是会有风险，因为模型是非确定性的，所以总有可能发生意外。

Leadership's gonna get mad at them. It's like, you know, why did you do this? So that is a big deal with NAI. I think that is one of the reasons why there's been difficult for a lot of use cases to really take off. Because at the end of the day, there is always gonna be risk because the models are non deterministic and so something could always happen.

但客户服务的好处在于，产品的工作方式自然内置了升级路径。代理正在进行对话，如果出于任何原因需要退出，它就会升级到人工服务。这种基础设施已经建立，你已经有了呼叫中心，已经有了电话系统或其他什么，所以直接连接就行。我认为仅这一特性就让事情变得容易得多，因为我们合作的那些大企业会说，好吧，太好了，我们在内部测试，然后上线时，我们只选择这一个领域，发布给5%的用户，即使对这5%的用户，如果出现问题，也只是升级处理。这给了人们足够的信心去尝试。

But the nice thing with customer service is that you have a escalation path just naturally built into the way the product works. The agent's having the conversation, if for any reason it needs to exit, it'll just escalate to a human. And that infrastructure is already set up, you already have your call center, you already have your telephony stack or whatever, so you just connect to it. I think that property alone just makes things way easier because these big enterprises that we work with, they're like, okay, great, we test it internally and then to go live, we're just gonna choose this one surface area and release it to 5% of the user base and even for that 5% if anything goes wrong it just escalates. And so that gives people enough comfort to go for it.

我认为这两点是企业级应用中最具吸引力的用例的重要原因之一。而编码则是另一个领域。编码稍有不同，它更多是自下而上驱动的。

Those two things I think are one of the big reasons why it's probably the I would argue, the use case with the most traction at the enterprise. And coding is another one. Coding is a little different, it's a lot more bottoms up.

能否与编码进行对比？这两个领域显然对客户极具价值，也能围绕其构建伟大业务。看看收入曲线就明白——你们这边显然是Cursor、Cognition等公司。请比较编码与客户服务的异同？

Can you compare it to coding? It seems like these are the two areas where it's blindingly obvious that it's useful to customers and you can build great businesses around it. Just look at the revenue curves. Yours here is obviously like cursor, cognition, etcetera. Compare and contrast coding and customer service?

是的，它们差异很大。或许可以这样理解：归根结底，AI代理的存在本质上是为了替代人类劳动。这正是其令人兴奋之处，也是众人聚焦的原因。

Yeah. They're very different. Maybe one framework to think about this is that at the end of the day, what AI agents are there for is to essentially replace human labor. That's why it's exciting. That's why everyone's so focused on it.

我们可以绘制当前人类劳动成本的频谱图。客户服务通常已被外包，尤其是AI正在处理的一二级咨询，这些岗位薪酬普遍不高；而频谱另一端是高薪的工程师。AI应用会从频谱两端开始吞噬——工程师因薪酬最高，他们具备充分利用AI的成熟能力，AI能极大提升其工作效率。

One thing you can do then is like you can just map out the spectrum of like how much that human labor currently costs. So with customer service, it's generally outsourced already, especially for the tier one, tier two type inquiries that AI is now handling. It's generally not folks that are super highly paid and then on the other end, it's engineers, which are the most highly paid people. The one way to think about it is that like AI use case will start eating the spectrum from both ends. And the reason why is that because engineers are the highest paid, they have the sophistication to like really leverage it well and like AI just gives them so much leverage.

当然还有其他因素。比如编码恰好可被标记化，且模型非常擅长。目前没有公司会说'有了编码代理就要裁减工程师'，因为工程需求是无限的，AI只是增强工具。

I mean, there's other factors as well. Like, it just happens that coding is tokenizable and the models are really good at it. That's one way to think about it. I don't know any company that's like, hey, I would like to let go of a bunch of my engineers because now I have coding agents. There's infinite engineering work to do, so you're just augmenting them.

而在另一端，替代效应更明显。维持大型业务流程外包(BPO)成本高昂且运营复杂——持续招聘、高流失率、培训、质量监控等。AI在此极具价值，因其能完全替代相对简单的工作。我认为'AI取代岗位'的讨论有些夸大，就连我们接触的BPO也不太担忧，毕竟这些岗位本就存在高流动性。

On the other end, it is more of the replacement sense. You have a large BPO and that's costing you a ton of money and it's also like a really high operational thing to maintain because you have to hire people all time, there's a ton of churn, you know, train them, know, QA them, you know, make sure that nothing goes wrong. So AI is really valuable there as because the work is easier for the AI to do, it can fully replace. And I think that goes into the conversation of I think it's a little bit overblown of like, oh AI is replacing jobs and so on. Even the BPOs we talked to, they're not really that concerned because what typically happens anyways, is that there's already very high turnover in these BPOs.

人员自然流动会逐渐减少这类岗位，他们转而从事AI尚无法胜任的更高级工作，比如数据标注等。这是我们观察到的BPO现状。这个频谱确实存在，接下来的问题就是：未来会如何演变？

People are just hopping around and doing all sorts of different things, of just naturally let it decrease and then they just go on to do sort of the next level of tasks that they I can't do yet, right? And so maybe that's like data labeling or something. That's generally what we're seeing from the BPOs. But yeah, anyways, I think this spectrum is pretty real. And so then the question is like, what is the next thing that happens?

所以这是一个非常有趣的结论，即究竟是增强顶尖人才的能力，还是取代光谱中最易被替代的那一端？这真是个引人深思的结论。我想谈谈你们在如何做好第二件事上初步获得的经验。如果其他人想要创办或投资一家公司，专门从事这种‘从光谱末端逐步蚕食’的业务，就像你描述的那样，你们学到哪些关键要素能提高成功率？比如在与特定公司建立合作时，如何通过流程设置来大量取代那些低垂的果实——比如某些类型的客服电话，或是过去需要人与人互动、现在可由AI处理的那些光谱末端的工作？

So it's a really interesting conclusion, which is try to augment the very highest talent or replace the most replaceable end of the spectrum? That's a really interesting conclusion. I wanna talk about what you've begun to learn about how to do that second thing well. So if others out there wanted to start a company or invest in a company that was doing sort of that end of the spectrum eating its way in, as you described, What have you learned are the key things to like the setup process with a given company to increase the likelihood that you can replace a lot of the low hanging fruit for types of customer service calls or types of what used to be human to human interaction and can now be handled by AI on one end of the spectrum?

是的，我认为我们最大的收获是——虽然如果能完美解决这个问题会让一切变得简单得多——但往往最耗时的是就‘优秀标准’达成共识。你可能以为在我们这个领域这很容易，毕竟就是整理一堆问答对而已。但不幸的是，实际情况要微妙得多。‘优秀’可能体现在语气、品牌指南或对话自然度上。就连具体答案本身也是如此——我们合作的许多企业业务范围非常广泛。

Yeah, I would say the biggest learning we had is that oftentimes a long pole in the tent and again, if by never if you can solve this well, it just makes things go a lot easier, is aligning on what does good look like. And you would think that in our space it's pretty easy because it's like, okay, maybe you just have a bunch of questions and answers and that's what good looks like. But unfortunately, it's a lot more nuanced than that. What good looks like could be in the sense of like tone and brand guidelines or how conversational you are. And even for the actual answers, like we work with a lot of enterprises where obviously their scope is broad.

因此前期必须确立的标准就是‘何为优秀’。我们在产品中内置了测试模拟套件，比如构建一万个测试用例，每个用例会持续运行五次，这样就能评估系统表现。这其实相当困难，我认为这对任何尝试构建此类公司的人都适用。如果要取代人力劳动，你必须先知道优秀的人力劳动是什么样。所以我们发现：能否直接让人告诉我们所有问题的标准答案？

And so one of the things you need to set up beforehand is what's good look like. So we have in our product essentially like a testing or simulation suite of like, hey, we're gonna build out 10,000 tests and each one is gonna be constantly running like five times and then you can get a sense of how well things are performing. And that's actually pretty difficult and I do think that is broadly true for anyone that's trying to build in this style of company. You're If gonna be replacing human labor, you need to know what good human labor is. So what we found is like, okay, well, can someone just tell us what are the answers to all these questions?

但大多数人其实并不知道，因为这些大型机构结构复杂，没有人能掌握所有问题的答案。因此必须设计一个流程，能轻松从所有知情者那里提取答案——可能是客户体验负责人或各产品线主管。你需要把他们聚集起来，让他们共同确定评估标准。如果这一步做得好，后续就会顺利很多，因为你可以专注建设，同时拥有量化评分（比如AI表现分数）。当建设完成且评分达标时，就能正式上线。

And most people don't actually know because these are large complex organizations, no one is like the person where like, hey, know how to answer all these questions. And so you have to design a process where it's very easy to extract these answers from all the people that do know. So maybe it's all the CX leaders or people lead different areas of the product and so you have to get them all together and get them to align on like, okay here's what the eval is essentially. You can do that well, then it makes everything a lot easier because now you're just building, building, building, you have this like quantifiable score that's like, hey, here's how well the AI is performing. And then once you're done building and the score is high, then you can go live.

是否可以这样理解：你们相当于在组织内部创建了一个封闭的强化学习系统？这是简化的说法吗？

Is the right way to think about this, you just created like a captive reinforcement learning process within an organization. Is that like the simplified version?

没错，这个视角很有趣。虽然不完全是训练模型那种纯粹意义上的强化学习，但确实是通过强化学习让智能体持续改进。

Yeah. Yeah. That's a interesting way to think about it. And it doesn't have the reinforcement learning in the pure sense of training a model. It can just be reinforcement learning and make an agent improve.

这种改进可能来自积累更多评估案例，也可能是制定更完善的边界准则来明确其行为界限。

And that could be compiling more evals, that could be compiling just like more guardrails guidelines around what it can and can't do.

这种推广速度有多快？假设我是客户，拥有一个500人的客服呼叫中心，我其实很好奇。我不清楚具体规模，比如这样的中心为一家公司处理多少通话量或互动量。但如果我把5%的工作量交给AI，并且对其表现满意，比如它运行良好且问题不多。人们愿意以多快的速度从5%提升到10%、15%乃至20%？

How fast does this spread happen? If I'm a customer and I've got the 500 person customer service call center or whatever, actually curious. I don't know what the volumes are, like how much call volume or interaction volume a center like that handles for a given company. But if I give you 5% of my workload and I'm satisfied with AI's performance, like it performs well and there's not lots of problems. How fast are people willing to go from five to 10 to 15 to 20%?

非常快。

Very fast.

我认为即便对大企业而言也只需数周。大家都想全面上线，但分阶段实施是为了确保不出差错。通过监测指标，几乎能立即发现问题。所以一周内，你就能评估500人规模或我估计年对话量达中高六位数的中心（可能更高些）的运行情况。关键是确认一切顺利，比如一周内就能看到解决率如何？

I would say even for large enterprises within weeks. Everyone wants to just go live to everything, but the reason why you stage it out is so you can make sure nothing's going wrong. And you can tell us something's going wrong like almost immediately because you have these metrics. So even within a week, you have 500 person or I would probably estimate mid to high six figures of conversations a year, maybe slightly higher. What you're doing there is making sure that things are going well, so within a week you can see like, okay, what is the resolution rates?

是否符合预期？很好。客户满意度如何？这些评分也有现成数据。此外他们可能还有基于人工审核的准确率指标。

Is that what we expect? Okay, great. What is the customer satisfaction? People have those scores as well. And then they'll probably have some sort of accuracy metric based on like human review.

如果这些都达标，就没有理由不全面推广。何况商业价值显而易见，对吧？既能大幅提升运营效率，客户也更满意。那就直接全面铺开吧。

If this all check out, there's really no reason why you shouldn't roll it out and again the business case is so obvious there, right? So hey, we're both generating a ton of operational efficiency and our customers are happier. So yeah, let's just send it out to everything.

当出现问题时最常见的情况是什么？随着产品改进这类情况越来越少，但早期阶段客户与AI互动中容易出什么差错？

What goes most wrong when something bad happens? I'm sure this is happening less and less as the products gotten better, but even in the early days, what sort of thing would go wrong in one of the customer to AI interactions?

问题类型千差万别。有时很明显——我们和客户双方都非常重视。但也有很多意外情况。早期有个大型票务平台客户，曾出现用户找不到票据的情况。

It ranges all sorts of different things. Sometimes it's obviously like both sides like us and the customer, we take everything very seriously. But there's a lot of things you wouldn't expect. In the early days, we have a customer that is essentially like a large ticketing platform. And one of the things that would happen was someone came in, they couldn't find their ticket.

我查看了他们的账户，发现里面根本没有票。然后他们就说，好吧，我打算直接去活动现场，从城里找八个无家可归的人一起带进去。客服人员当时都惊呆了，说你能为社区做点好事真是太棒了。但这种事情吧，基本属于可遇不可求。所以这些都需要你慢慢调整适应。

I looked into their account, it's like, hey, there's no tickets here. And then they were just like, okay, well what I'm gonna do is I'm gonna show up to the events and I'm gonna find eight homeless people from the city and bring them with me. And the agent was like, oh my god, that's so awesome that you're thinking about doing something nice for the community. Things like that where it's like, okay, do not expect that to happen. So those are just like tuning you have to do over time.

核心精神在于找到监管力度与灵活性的平衡点。至少在我们这个领域，这就是游戏规则——也是我们设计产品的理念，更是我们目前取得成功的关键原因。生成式AI真正释放的是极致灵活与高度个性化的能力。你可以把它想象成古早的决策树对话系统，但传统系统的问题在于：当用户提问偏离预设分支时，就会被强行拉回既定路径。而ALM技术将大量决策逻辑转化成了语言模型的神经元连接——这才是真正的颠覆性突破。

And generally it's in the spirit of that where you're trying to find the right level of guardrail and flexibility. I think that's the name of the game with our space at least and that's sort of the way we design our product and why probably the number one reason we've been successful so far, is that what Gen AI really unlocks is super flexible, super personalized. One way you can think about it is like in the old days to map out a conversation you just build like gigantic tree of decisions and that's very hard because no one likes that experience and you ask something that's not quite one of the branches and just forces you down that branch and there's no way to go back. And what ALM does is it extracts a lot of that tree into the neurons of the language model. So that's really powerful.

光谱的一端追求极致的灵活性和原生对话能力，让AI听起来完全像人类。但在企业场景中，很多情况恰恰不需要这种特性——比如受监管的用例就必须绝对严谨，三个步骤必须严格按顺序执行，前序条件未达成绝不能跳到第三步。所以必须设计能适配整个光谱需求的系统。

On one end of the spectrum, you're just looking for flexibility and root power and being able to sound really human like. But with the enterprise, there are lot of things we don't necessarily want that as much. And you want full like rigor, right, if it's a regulated use case, like you cannot afford for it to ever deviate. These three steps always have to be followed in this order and you can't go to step three until something has happened already. So you need to design a system that can be anywhere along that spectrum.

回到关于可能出错的问题，最糟糕的情况就是AI说了不该说的话。因此必须构建具备高度鲁棒性的AI系统——你可以根据用例需求选择严格模式。比如账户查询这类简单场景，我们就不需要设置太多限制，这种自由度正是提升客户满意度的关键。

And back to your question of what could go wrong, well, the worst thing that has happened would be it just says something that's not supposed to say. And so you need to design an AI that's really robust to that and you can choose like, okay, for this use case we really need it to be over here, that's a lot more robust. But for other use cases where if you're just asking a basic question on their account, we don't want it to be like that. We don't have to be super free form, and that's how we get the customer satisfaction up.

由于非确定性的特性，多数人仍聚焦在可能出错的地方。那么光谱的另一端呢？有哪些远超预期的成功案例？智能体在哪些方面的表现完全突破了你们最初的设想？

I think most people are still focused on things that could go wrong because it's nondeterministic. What about the total other end of the spectrum? What have been the things that have gone way more right than you expected? Where has the potential of agents outperformed your expectation in terms of what they can handle or what they can do?

本质上是体验升级。有个常被忽视但对我们至关重要的指标：用户直接要求转人工的频次。就像我们打电话给客服时狂按0键那样——这源于人们对糟糕体验的条件反射。

It's really just elevating the experience. One sort of metric, which is usually a secondary metric that folks think about later, but is is quite important to us is just how often do people come in and just say like agent agent agent, like, give me to a representative, like, I wanna talk to you. I've done that. You probably have as well, where you're just like calling into some sort of customer service and you're just like pressing zero the whole time. And that's because people are used to bad experiences.

用户早已对这些系统失去信任。令人惊讶的是，只要从一开始就明确这是全新体验，人们就愿意尝试，结果截然不同。比如我们某个可穿戴戒指客户，在使用JAN AI系统前，每三个咨询者就有一个会不停要求转人工；现在这个比例降到了二十分之一。

So they've already lost the trust of these systems. I think what surprised us was that if you just make it really clear off the bat that this is a different experience, people are willing to give it a chance and then the outcomes are just way different. One of our customers or a ring like the wearable ring. We did a case study with them where before having any sort of JAN AI system, one in three customer that came in would just not bother saying anything, just keep jamming agent until they got to one. Now it's one in 20.

因为我们投入了大量时间让流程的初始阶段显得与众不同，人们也愿意给它一个机会。所以我觉得这很令人振奋。

Because we just like spent a lot of time making the beginning of the process just feel very different, and folks are willing to give it a chance. So I think that's been exciting.

你认为这能发展到什么程度？随着产品后续迭代以及模型底层能力的提升，体验还能在哪些尚未完善的方面变得更好？

Where do you think that can go? How good can the experience get in ways that it's not yet that good with subsequent evolution of your product, but also of the underlying capabilities of the model?

目前最大的前沿领域是语音模型。我知道有很多有趣的初创公司也在研究语音技术。令人兴奋的是，这个问题显然还未完全解决，仍有大量工作要做，标准也非常高。想想看，人类整整十五万年的文明史中，每个个体的交互界面都是语言——口头语言。

The biggest frontier right now is voice, voice models and so, I know there's a lot of interesting startups as well working on voice. It's exciting because it's still I would say definitely not solved. There's a lot to be done there, the bar is very high. So if you just think about how humans communicate for literally the entirety of humanity, I don't know, a hundred fifty thousand years or something, the UI for every human is language. Spoken language.

听着，说话是我们大脑进化形成的本能，是最自然的交流方式。在这漫长的历史中，只有最近大约六十年我们才用键盘和手机打字交流。我认为任何与人类交互的智能体，语音都必须是关键因素，因为这就是我们的沟通方式。由于大脑对此高度适应，我们很容易察觉哪里不对劲。

Listen, you speak, and that's how our brains have evolved. That's the most natural way for us to communicate. Only in the last what, like sixty years of that entire time did we have keyboards and like phones and communicating through typing. I would say fundamentally in any sort of agent that communicates with humans, voice has to be a critical factor because that's just how we communicate. Because our brains are so evolved for this, it's very easy to tell when something feels not quite right.

因此标准极高，恐怖谷效应也很明显。这就是为什么大家都在努力提升语音体验。比如Chatty Bitty或Sesame这类语音模型已经开始让人印象深刻，但长时间对话后你肯定能发现它不是真人。这是目前存在的局限。

And so the bar is very high, uncanny valley is quite large. That's why there's lot of effort going into making the voice experiences good. I would say Chatty Bitty voice for example, or the Sesame or like these voice to voice models, they're starting to feel very impressive. But you talk to it long enough, can actually it's like you can definitely tell it's not a human. So there's that element of it.

但对于我们这样的企业用例，仍有许多障碍需要克服。这些模型虽然不错，但幻觉率很高，无法直接应用于现有系统。所以现在常见的做法是将语音转文本再转回语音，通过多重校验确保准确性。如何在保持拟人化的同时确保精准度，以及如何整合各个环节，这些都是当前重点探索的方向。

But then, you know, for enterprise use cases like us, there's still a ton of hurdles to cross because those models even though they're good, the hallucination rates really high. So you can't really use them necessarily as is in the current systems. And so a lot of people what they do now is they go from voice into text and then back to voice and then you can run a lot more checks there to make sure that things are accurate. And so a lot of cool ideas there to explore on how do you make it both human like but accurate and how you tie everything together. So that's where most of the work is going to these days.

虽然可能涉及技术细节，但为什么语音直连值得探索？相比总是转回文本处理的方式，它的独特价值是什么？

At the risk of getting too technical, why is voice to voice interesting and worth pursuing versus just always going back to text and being able to manage it that way?

所以根本区别在于，如果你只是发送文本，那么无论如何，最终的音频都只是对文本的朗读。语音转语音之所以强大，是因为它考虑了你所说的整个音频，因此它能感知节奏、你可能有多沮丧、语气等一切因素。延迟也少得多，因为你是直接从语音到语音。而我们在交谈时，延迟至关重要。就像我们现在交谈时，大脑会不断思考：他什么时候说完？

So the fundamental difference is if you're just going to text, then no matter what, the final audio is just a narration of the text. The voice to voice is powerful because it takes into the entire audio of what you said, so it knows cadence and maybe how upset you are and the tone and everything. Latency is a lot less as well because you're going straight from voice to voice. And latency matters so much when we're talking. When we're talking right now, our brains are constantly going like, okay, when's he done talking?

我该什么时候开始说话？如果有人打断对方，礼貌的方式下人们会非常自然地调整。这就是语音转语音的最大优势。最终我认为主流观点是：首先，无论最终体验如何，若想让它与人类交流难以区分，就必须采用语音转语音，或至少考虑语音输入。但语音转语音的根本问题在于，由于语音维度更多，每句话生成的标记量远高于生成文本时的数量。

When should I start talking? If someone interrupts someone else, it's like in a polite way, people adjust very naturally. So that is the biggest proponent of voice to voice. And ultimately, I think the prevailing view is that first, whatever the final experience is, if you really want to make it indistinctual from a human, you have to do voice to voice or you have to at least take into account the voice. The issue with voice to voice though is that also fundamentally, because voice has a lot more dimensions to it, the amount of tokens you generate per sentence is just a lot higher than when you generate text.

标记越多，出错的可能性就越大。因此迄今为止，幻觉率仍然高得多。

The more tokens you have, the easier it is for something to go wrong. And so the hallucination rate has so far just been a lot higher.

高多少？给我们一个概念，距离这项技术真正成熟还有多远？

How much higher? Give us a sense of how far we are from these being really good.

我想大概是八倍左右吧。确实高不少。当然你想利用这项技术，所以现在或许有创新方法将两者结合。比如让文本模型生成内容，但同时考虑之前的音频输入，这样能生成非常逼真的结果。

I think probably like eight x higher or something like that. Wow. It is quite a bit higher. Of course, you want to leverage that technology, so now maybe there's creative ways to make a hybrid of the two. Maybe you can have a text model generate the content, but you take into accounts the audio from before as well and like that makes something very realistic.

但延迟仍是难题，因为在企业场景中，你在开始回应前需要做大量工作：要分析对方询问的内容、收集所需材料、是否需要调用API获取数据等。必须以感觉自然的方式处理。有时想想人类的做法，你可能得说'给我点时间查一下'，因为API确实需要十秒返回数据。这些都是值得深思的有趣问题。

But at the same time latency is still the hard problem, because at the enterprise what's happening is you are doing a lot before you can start responding. You have to figure out like what are they asking about, like what materials do I need to collect, do I need to hit any APIs and get that data back. And so you have to do it in a way that feels very natural. And sometimes if you think about how human does it, you might have to say something like, give me a sec to look that up because it actually genuinely takes ten seconds for the API to come back. These are all interesting problems to think through.

那么请概述当前情况——如果汇总所有交互，包括时长、类型（语音、文本或其他形式），如今Decagon智能体与客户之间的整个交互语料库是怎样的面貌？

So give us a sense today, if you add up all the interactions, some idea of how long they are, what type they are, voice versus text versus some other modality, what is the entire corpus of interactions between a Decagon agent and a customer look like today?

我认为目前聊天和语音之间的比例相当平衡。从原始客户基数来看，至少对我们而言，使用聊天功能的人更多。但如果你只考虑像财富一百强这样的大型企业，它们存在已久，人们习惯直接打电话联系。因此在这些企业中，语音服务占比异常高。很多企业95%是语音，5%是聊天。

I would say pretty balanced at this point between chat and voice. On a raw customer basis, there's more people on chat, at least for us. But if you just think about the large enterprises like the Fortune one hundred, they've just been around for so long and everyone just calls them. So voice is just disproportionately higher there. A lot of them are 95% voice and 5% chat.

这就是我们观察到的现象。至于对话类型，通常从较简单的问题开始，这些都是基于问答的形式。第一层级就像，你根据静态知识回答他们的问题。比如关于会员制度如何运作的问题，或者如果我买了某样东西还能退款吗这类问题。第二层级虽然仍是问答形式，但会大量利用实时数据。

So that's what we're seeing. Then in terms of the types of conversations, it's generally ones that are fairly you start with the sort of easier ones of course, and so these are things where they're question answer based. So that's the tier one is like, you answered their question based on what you statically know. So that could be questions about how your loyalty system works or questions about if I bought something, would I still be able to refund it? Then the next level is is still question answer based, but you're leveraging a lot of real time data.

例如可能有这样的问题：我这笔交易应该获得5倍积分但只得到2倍，为什么？然后系统会实际查看你的账户并分析原因：好的，我看到顾问提到这种情况，让我查找相关文档——哦，原来是因为你通过旅行社预订。如果直接向航空公司预订就能获得5倍积分，但这个优惠不适用于旅行社渠道。第三层级则是实际执行操作。

So that could be, you know, I got two x points on this transaction, but I should have gotten five. Why is that? And then it'll actually go and look into your account and reason through things, okay, I see the counselor said this type, let me go find all the documentation on this type and like, okay, actually it's because you booked through a travel agency. If you have booked directly with the airline, you would have gotten 5x, but this thing doesn't apply to a travel agency or whatever. And then the third tier is you're actually taking action.

比如'我信用卡丢了需要补办'，这需要走一个相当复杂的流程。这正是AI代理表现出色的地方，因为你原本不会指望它能处理这种事。它可以进入一个复杂系统：首先需要确认你的地址是否正确，然后询问是否需要冻结旧卡，可能还需要检查是否存在欺诈行为——防止有人不断申请新卡。所有这些环节串联起来，才真正体现出它的代理能力，也是语言模型带来阶跃式进步的原因。

So I lost my credit card, I need a new one. And it's actually walking through a pretty large flow. And that's where AI agents have been really excellent, because you actually wouldn't expect it to be able to do that. And so it's able to go in, it can be a pretty complicated system where it's like, okay, well, first, I need to figure out what your address is and confirm if the address is correct. And then I need to look and see, hey, do want me to lock the old card?

就像这样：好的，我会处理。可能需要检查是否存在欺诈，确保这个人不是频繁申请新卡。正是这些功能的有机结合，才使其具有代理属性，这也是语言模型带来如此显著进步的原因。

Like, okay, great. I'll do that. I might need to check for fraud to make sure this person is not just constantly asking for new cards. And it's just like all these things stitched together. That's really what makes it agentic, and that's why there's been such a step function improvement with LMs.

顺着这个问题思路，公司从这些历史闲置的交互数据中能发掘哪些机遇？基于一对一互动中对客户更深入的了解，以及整体层面上对客户行为模式的把握，他们能为客户提供哪些前所未有的新服务？

In the spirit of that question, what could go right? What could go right for the company based on the data they're gathering from these interactions that they probably have been doing nothing with historically? Like, what new things can they do for their customer because of on a one to one basis, like, they're just learning more about a person. And on the aggregate basis, they like understand the behavior patterns of their customer base or something.

哦，这确实是个重大课题。我们认为这些数据极具价值——因为它们直接反映客户心声——但由于历来是非结构化数据，其利用率一直很低。这正是我们产品的重要部分。

Oh, yeah. That is a huge topic. I think that's a huge part of our product. We have a viewpoint that this data, of course, is super valuable because it's literally what your customers are saying. But it's very underutilized because historically, it's a very unstructured data.

通常人们的做法是，每月处理百万次对话，组建20人的全职团队，抽样检查对话内容，依据评分标准进行核查，汇总主题等。但这效果有限。如今，你可以让语言模型逐条阅读对话，提取所需信息。这样就能发现那些因组织庞大而连高层都未曾察觉的问题——比如系统会自动标记‘2%的对话进展不顺，源于我们对某主题缺乏背景认知’这类情况。

So what people typically would do is like, okay, well, every month we have a million conversations, we'll have a full time team of 20 people, and they're just like sampling these conversations and trying to like check on a rubric and try to compile topics and things like that. And that won't get you so far. But now what you can do is you can literally have a language model that reads every conversation and extract whatever info you want from it. And so that allows you to do things like, okay, well over time there are these topics that people probably didn't even know about because these organizations are big. So the people in leadership positions, they can only have such granular insight into what's happening, but it'll literally just flag like, hey, there's this 2% of conversations where things are not really going that well and it's because we don't have context on this topic.

因此我们可以标记这些问题。根据客服人员的处理方式或现有流程，起草改进建议。比如‘这是针对客服人员的调整方案’。通过这种方式，客服能力就能持续自动提升，这非常关键。

And so let's flag that. Let's draft a suggestion for what could go better here based on what how the human agents are handling or based on the other procedures that we have. And here's a suggestion for how you should adjust the agent. And that allows the agent to improve automatically over time. And that's really critical.

在智能代理领域构建竞争壁垒时，核心在于：与客户合作一年后，你的代理是否通过数据学习实现了持续进步？这不同于单纯的数据训练，而是进化到其他代理难以企及的绩效水平。

So when you think about moats in the agentic world, a lot of it is around if you've been working with a client for a year, has your agent just continuously gotten better by learning from the data? And that's a different concept than just training on the data. But has it continuously gotten better to the point where it's just very difficult for another agent to come in and perform at the same level?

当前有个有趣现象：CEO们迫切希望在业务中应用AI，却不知具体方向。他们缺乏评估框架，虽然了解自身业务，却陷入‘不部署AI就会错失良机’的焦虑中。

There's this funny situation today where I think especially CEOs really want AI in their businesses. They want it now, but they don't know what they want. They don't have a good framework for thinking about, okay, I understand my business. I don't know where to go. It seems like I would be missing a major boat if I don't deploy this in my business.

他们面临三重困惑：不知从何着手，没有优先方向，也找不到专业咨询。除了客服自动化等具体应用，你们是否为企业领导者开发过评估框架？我指的是能系统性识别哪些问题适合由智能代理或大语言模型解决的思维框架。

I don't know what to do. I don't know where to go first. I don't know who to call. Have you developed any framework for those company leaders that desperately want to use this technology in their business, but they simply don't know apart from automating customer service or something? I don't mean specific use I mean like a framework for thinking about what kinds of problems might be addressable by agents or by LLMs.

有意思。其实有个类似我们之前讨论的二元框架。多数企业领导者更关注‘自下而上’的维度——即找出那些重复性高、人力成本效益低的环节。这是我们观察到的主要切入点。与高管交流时有两个显著现象：

Oh, interesting. I mean, one framework is similar to the sort of framework we talked about before of two ends of the spectrum. And I would say most leaders we talked to are focused on the more bottoms up end of the spectrum, which is like where are the areas that we should just not have humans doing because it's so mundane and repeatable and there's tons of cost efficiencies there. So I would say that's where folks are typically thinking of. So when we talk to leaders, I think there's a couple observations.

首先，当前所有AI计划都是自上而下推动的，因为这属于董事会级战略。C级高管极度关注AI部署方向，这意味着在大企业推进项目必须获得顶层支持。其次，正如你所说，他们评估应用场景的核心标准始终是投资回报率。

One, pretty much all AI initiatives are very top down at this point, because it is such a board level mandate. So that C suite is very, very invested in like, okay, where do we deploy AI? That almost means that if you wanna get something going at a larger organization, you have to have buy in from the top level because it's gonna get up there anyways, and they have to make decision at the end of the day. So that's one. Two, the way they think about the use cases, to your point, is back to ROI.

这就像是在问，我们能在哪里大幅节省成本或创造大量新收入？如果你连半句话都解释不清楚，那现在就行不通。他们压力很大，需要展示快速见效的成果。如果这不能快速见效，比如我说的1000万美元，那就不会被优先考虑。

It's like where can we either save much money or make a lot of new revenue? And if you cannot in basic half a sentence explain that, then it's just not gonna work right now. They are under a lot of pressure, right, they needed to show like quick wins. If this is not gonna be a quick win, they can point to like I say $10,000,000, then it's not gonna be something that's prioritized.

你认为编程能很好地回答这个问题吗？你觉得编程的投资回报率明确吗？

Do think coding answers that well? Do you think the ROI is clear in coding?

是的。我很了解编程代理，他们通常的做法是：首先很容易向工程师们测试展示，然后只需对工程师做个调查，比如‘嘿，你觉得你的效率提高了多少？’工程师往往是这些组织中最宝贵的资源，所以他们的回答会被高度重视。如果工程师告诉你效率提高了50%，那就够了。

It is. And I know the coding agents quite well and the way they do it generally is, one is very easy to test to play out to the engineers, and then you just do a poll on the engineers of like, hey, how much more productive do you think you are? And engineers are often the most valuable resource in these organizations, and so those answers are treated with high importance. Yeah. If insurance tells you that you're a bit, like, 50% more productive, it's like, okay.

很好。

Great.

你确信这是真的吗？他们能自我报告并保持准确吗？之前有那个计量研究或类似的东西出来，实际上我不知道研究是否可靠，但生产力是下降或持平的。自我报告的生产力与实际测量的生产力存在差异之类的。

Are you confident that that's true? That they can self report and be accurate? There was that meter study or whatever that came out that actually I don't know if the study is good or not, but productivity was down or flat or something like this. The self reported productivity was at odds with actual measured productivity or something like this.

哦，是的。我对那个了解不多，但我想说的是这其实不重要。如果整个工程团队都说‘嘿，我们喜欢这个，它让我们的效率提高了50%。’

Oh, yeah. Don't know nearly enough about that, but I'm just saying that it doesn't really matter. If their entire engineering team is like, hey, we love this. This is making us 50% more productive.

是啊。就像超级简单。构建你的东西。

Yeah. It's like super easy. Building your thing.

没错。正是如此。然后你从CEO的角度想想，如果他们向董事会汇报这类事情，就会说：看，我们整个工程团队声称效率提高了50%。这绝对是值得的投资，因为这些人的薪资水平摆在那里。现在我们可以加速产品开发，加速一切进程。

Yeah. Exactly. And then you think about from the CEO and if they're like reporting this to the board or something, it's like, hey, my entire engineering org said that they're 50% more productive. This is like worthy investment because these people are all paid this much. Now we can accelerate the product, we can accelerate everything.

对。你认为未来会是——我想聊聊品牌，以及企业如何希望其代理体现品牌特质，包括文化、风格、语气等。但你觉得最终状态会不会是每家公司都有个具名化的人格代表，让用户习惯与之互动？不仅处理客服问题，还包括销售咨询。

Yeah. Do you think that the future is I wanna talk about brands and how a given company might want its agent to feel and sound that an agent might be a way to express brand, culture, style, tone, whatever. I wanna talk about that. But do you think the end state here is that each company sort of has almost like a named personified representative that you just come to expect to interact with? And it's not just customer service issues, but it's sales issues.

你可以向它咨询该买哪双鞋之类的建议，所有这些功能都集成在一起。或者公司不同部门会有不同代理？其实我想问的是，五年后的自然终局里，企业代理的未来形态会是什么样子？

You ask it for advice on what shoe to buy or whatever it might be, and that that's all integrated, or that you'll have different agents for different parts of the company. Like, I guess what I'm trying to ask is what the future of a company's agent or agents looks like in the natural end state five years from now or something?

是的，我认为最终状态会更趋统一，原因正如你所说——企业需要统一的品牌形象。品牌对他们至关重要，虽然不同企业重视程度不同。但最终这会成为企业的前端界面，既是获取新客户的渠道，也是维系现有客户的手段。极端情况下，比如银行、航空公司或电信公司，代理可能成为大多数用户唯一的接触点——他们甚至不需要打开你的App或访问网站。

Yeah, I would say that in the natural end state, it is more unified for the exact reason you listed, which is people want a unified brand out there. Brand is very important to them, and for some businesses it's more important than others, but eventually this becomes the front end for the business. And so that means it's both how you gain new customers, but also how you support the existing ones and make them retain more and so on. It's almost like in the limit, if you're working with a bank or airline or telecom company or whatever, the agent could be the only thing that most users interact with. They don't even have to touch your mobile app, they don't even have to go down your website ever.

用户只需通过这个经过身份验证的代理，它了解用户的一切信息，记得所有对话上下文，能直接解决问题。它可以代你执行操作——订机票、升舱、解答各类疑问。这就是我们正在构建的愿景，某种程度上我们称之为'数字礼宾'，一个能包办一切的服务存在。当然双方都要务实，从明确的使用场景起步。但这就是行业的发展方向。

They just have this agent where they're authenticated, it knows everything about them, and has all the context of your previous conversations, it has memory, and it can just solve your issue. It can take actions for you, you need to book a flight, you upgrade a seat, you have questions about this or that. So I think that's the exciting vision that we're building towards and in some ways we often call this like a concierge, it's just a digital concierge that can do everything for you. I mean, you also have to be pragmatic on both sides, you know, start with a clear use case. But I think that is where folks are building towards.

短期来看，确实会有不同团队各自开发代理，因为大公司的现实就是不同团队有独立预算和决策权。但长期而言，要么需要统一框架整合这些代理，要么干脆就变成同一个代理。

And so in the near term, I do think that's different teams because the reality is that at these large companies, different teams have different budgets and they make different decisions and so they might have different agents. But long term, you either need to have a unified system that ties them together like unified framework, or they could just be literally the same agent.

这个类比是否恰当：企业会像对待官网那样对待代理？投入大量精力设计，让它呈现特定风格，成为企业与世界统一的交互界面。这个类比准确吗？

Is the right analogy here at a company's website, They'll think about their agent like they think about their website. A lot of work will go into it. It'll look and feel a certain way, like it'll be kind of a unified interface with the world. Like, that a is that a clean analogy?

我认为这是个很好的比喻。它就像一个前端，像一个用户界面。只不过不是视觉上的UI，而是对话式的UI。

I think that's good analogy. It's like a front end. It's like a UI. But instead of visual UI, it's a conversational UI.

你觉得品牌方在要求他们的AI代理展现个性时最在意什么？他们希望它友善、简洁还是幽默？从你开始做这行以来，这个维度是如何演变的？

How do you feel brands pulling personality requests in their agent out of you? What do they care about? They want it to be nice, they want it to be concise, they want it to be funny. How has that dimension evolved since you started?

企业通常已经具备品牌指南，因为他们需要用它来培训人类客服。好处在于这些现成的人类培训流程，理论上可以同样适用于AI。比如他们会要求'必须保持自信'，有些客户不希望AI道歉，有些则特别看重歉意表达——这些不同偏好都需要高效地教会AI。不过这种沟通方式效率相对较低。

People already almost always already have brand guidelines because they need to show brand guidelines to their human agents to train them. The nice part is that they already have all this training process for the humans, and you should ideally be able to apply it in the same ways with AI. And so they'll have like, hey, you need to do this, you need to always be confident. Sometimes folks don't want the agent to apologize, sometimes people really want to be apologetic, so people have different preferences, and that needs to be taught to the AI in an efficient way. That's also kind of a lower throughput way of communicating.

另一种训练方式是提供大量范例。比如展示顶尖客服的优秀对话样本，让AI从中学习。

I mean, the other way that you can show the agent is just give a lot of examples. So here's examples of what great look like from our top agents, and just learn from that.

Decagon可能实现的、最宏大到几乎不敢说出口的愿景是什么？

What is the very biggest, you're almost afraid to admit it because it feels so big, version of what Decagon could be?

我们如此兴奋并认为市场巨大的原因是——我们最终构建的将是产品的新型交互界面。想想大公司在手机应用和网站上的投入吧，这就是人们与品牌交互的主要方式。未来任何用户与品牌的交互都可能通过AI代理完成，涵盖各种使用场景。我们的客户尤其看重AI的上下文理解能力，因为它能无缝衔接不同流程，比如在客服结束时切入销售场景，反之亦然。

The reason why we're excited and we feel like the market is so massive is that, yeah, at the end of the day, what we are building towards is this concept of becomes like a new UI for the product. Just think about how much these large companies invest in their mobile app or their website, and this is literally how everyone communicates with them. So this could be a way where eventually, the way any user interacts with any brand is through an AI agent and it's for all sorts of different use cases. The benefits our customers for the AI to have all this context because it can seamlessly flow between things. A lot of them wanted to do sales type use cases at the end of a support flow or vice versa.

这很令人兴奋。

That's exciting.

最佳客户拥有哪些内部背景或资源使体验更佳？你提到品牌指南，比如可能有关于品牌指南的书面说明。哪些内部资产是企业拥有或缺乏的，若拥有它们，能让Decagon体验大幅提升？

What internal context or things do the best customers have that make this better? So you mentioned brand guidelines, like maybe there's a write up on what the brand guidelines are or whatever. What are like the internal assets that companies have or don't have, that if they have them, it's made the Decagon experience like way way better?

首要因素是供AI使用的API接口。包括执行操作的API、查询数据的API、数据重构的API。通常如果有这些，第一个月就能预见出色体验。若尚未具备，我们应尽快构建，让AI真正实现高阶体验。这是关键所在。

Number one thing is just APIs for the AI to use. So APIs to take action, APIs look up data, APIs to reformat things. That is often like, hey, if you have those, you already know that within the first month, it's already gonna be a great experience. If you don't yet, then we should try to build towards that as soon as possible so that the AI can actually achieve that elevated experience. That's the main thing.

多数企业已有文档，可能未及时更新，但我们可以协助。我想多数企业也已有标准操作流程(SOP)。我们能利用这些生成专属格式——我们称之为智能体操作流程(AOP)。

Most people already have documentation. It might not be up to date, but we can help with that. Most people already have SOPs, I guess. And then we are able to use that and generate our format. We call them AOPs, agent operating procedures.

这些是AI专用的SOP。如我所说，品牌指南通常企业也已具备，直接导入即可。

They're SOPs but for AI. And then brand guidelines, like I said, people usually have those as well, so you just ingest them.

上次我们讨论过，任何企业（尤其是你们行业）的三个关键利益方：团队、需招募的人才（我想详谈这点）、资本/投资者、客户。供应商或许是第四类，但最关注前三个。上次见面时我问你最困扰哪个环节，你笑着回答'绝对不是投资者'。请谈谈投资者对你们这类公司的投资热情及其感受。

We were talking last time about in any business, but certainly in yours, the three key stakeholders being your team, talent that you have to recruit, and I wanna talk about that in detail, capital, investors, and customers. Maybe suppliers too is a fourth category, but especially interested in the first three. And the last time we were together, asked you, like, which one do have trouble with? And we laugh because you said, definitely not investors. Talk a little bit about the demand from investors to invest in companies like yours and how that feels.

他们如何争取投资机会、进入股东名册？竞争有多激烈？当前感受如何？因为确实有少数像你们这样的公司，处于风口领域、拥有关键市场表现和优秀团队，几乎每个投资者都想参与。这种体验是怎样的？

What are they doing to try to give you more money, get on the cap table? How competitive does it feel? What does that feel like right now? Because it does seem like there's a handful of companies like yours that are in one of these white hot areas that have key traction, that have great teams, and basically every investor wants to be involved in those companies. What does that felt like?

目前AI领域确实存在过度狂热。融资似乎过于容易。市场上很多公司...对我们来说非常幸运。但我们不会视为理所当然。

It definitely feels like there's maybe a little bit too much excitement right now on the AI side. It just seems way too easy to raise money. So many companies out there I mean, for us, we've very fortunate. Right? We don't take it for granted.

我是说，我认为在我的第一家公司，我们也很幸运。那时融资很容易，但原因不同，比如2021年那样的大环境。如今，真正在营收方面有实际吸引力的AI公司并不多，尤其是在企业领域。所以我觉得这对投资者很有吸引力，因为他们想把资金投入到AI这个最大的趋势中。因此，我认为我们与所有投资者都保持着良好的关系。

I mean, I think in my first company, we're also fortunate. It was easy to raise, but it was for different reasons, like 2021. Nowadays, it's just there aren't that many AI companies that have real real traction on the revenue side, especially in the enterprise. And so I think that's attractive to investors because they wanna deploy their capital into AI and that's like the biggest trend. So I think we've had a great relationship with all our investors.

我觉得我们真正选择的是那些在个人层面上相处融洽，并且对我们市场拓展非常有帮助的人。是的，这是个挺有趣的过程。基本上每次我们完成一轮融资后，几乎立刻就会收到抢先投资的提议，这本身就不太合理。从第一性原理思考，投资决策应该基于业务表现，上一轮的估值不该成为决定性因素。

I think we just really select it for folks that we get along with at a personal level and we feel like will be very helpful for our go to market normally. Yeah, that's kind of an interesting process. So yeah, pretty much after every single time we've raised a round, we just like almost immediately gotten preempted and that alone can't be right. Just thinking first principle is to make an investment. The previous valuation should not be like a super big factor in that.

应该看企业运营得如何？我相信的潜力有多大？所以感觉确实有点狂热，但没错，我们绝对更关注另外两个要素——人才和客户。

It should be like how well is the business doing? How what do I believe the potential is? So it feels like there's a little bit of mania, but yeah, we're definitely I would say more indexed on the other two things like talents and customers.

投资者为了投资你们做过的最疯狂的事是什么？

What's the craziest thing that an investor has done to try to invest in the business?

我觉得没什么特别疯狂的事。我们主要采取的做法——实际上我鼓励更多创始人这么做——就是在投资者有意向但尚未投资的阶段，那时他们最愿意提供帮助。这其实是检验他们后续帮助程度的绝佳方式。因为如果他们在迫切想投资时都不够积极，那投资后肯定更不会帮忙。当然他们仍会保持友好，希望不至于拖后腿，但这个阶段确实是测试对方帮助意愿的最佳时机。

I don't think there's anything like super crazy. The main thing that we do and I would actually encourage more founders to do this, is that during the stage where people want to invest, but they haven't yet, that's when they're most willing to be helpful. And so it's actually like a great way for you to use that to proxy how helpful they'll be afterwards. Because if they're not that helpful in that stage where they really, really want to invest, they're willing to do anything, they're for sure not gonna be helpful afterwards. I mean, they'll still be friendly and like, hopefully, they'll not be detrimental, but that's like your opportunity to really test folks and see how helpful someone will be.

我们当然很了解很多投资者，我想他们对此都没意见。他们明白自己参与的是竞争性活动，本质上需要赢得投资顶尖公司的资格。所以从他们的角度看，他们也乐意为此付出努力。

And we obviously know a lot of investors very well, I think they like no one has issue with that. They know that they're going back to competition. They're also in a competitive sport. They know that they need to earn basically the ability to invest in the best companies. And so I think from their perspective, they're happy to work for it.

所以只要给他们机会，他们就会行动。

So if you just give them opportunity to, they will.

我想从创始人和投资者两个角度来探讨这个问题。在那个融资窗口期，你会给投资者什么建议？比如，当融资正在进行时，这是一个开放的过程或窗口期。你见过最优秀的投资者是如何表现的？不仅仅是简单地帮助你。

I'm gonna ask about this from both founder and investor perspective. What advice would you give investors during that window? Like, when there's a fundraising that's happening, you know, this is sort of an open process or a window or whatever. What have you seen the best of them do well? Not just, I'm sure, helping you.

比如给你介绍10个客户，这当然很好。但在尽职调查方面，他们如何在短时间内确保对你的业务有极其深入的理解？那些最顶尖的投资者在这个窗口期做了什么？

Here's 10 customers you can talk to. I'm sure that's great. But also on the underwriting side, them making sure they understand your business extremely well, I'm expecting in a short period of time. What have the very best done in that window?

首先，我认为最优秀的投资者深谙此道。我们确实接触过很多知名投资者，他们往往不愿意在投资前提供太多帮助。从他们的立场来看这没错——毕竟要优先服务现有被投企业。

Number one, I think the best investors get this dynamic. We've definitely talked to a lot of high profile investors where they're just like not willing to help much until they're invested. Totally, they're right. Right? It's like, hey, we have a lot of our current investments.

他们不愿为新投资者或新项目消耗社交资本。但创始人会想：如果这样，我如何判断你是否真能提供价值？这和那些实际上帮不上忙的人说同样的话有什么区别？最优秀的投资者会向创始人释放强烈信号：我真心愿意且有能力帮忙，我们拥有强大的网络，特别擅长某些市场拓展环节——这些都能在融资前几个月就展现出来。

We don't wanna like use our social capital or whatever to help within it, like a new investor or a new investment. But I think from the founder's perspective, okay, if that's the case, then it's hard to tell if you're actually useful or not. There's no difference between you saying that and then someone who can't really help by just saying that. I think the best ones just are able to give a lot of signal to the founder that like, hey, I'm really willing to help. I have the ability to help and we have a very strong network or we're very good at certain elements of go to market and we're just able to show that in that, let's say, it doesn't have to be that long a period of time, like a couple months before the round actually happens.

这是一方面。另一方面，无论是员工、投资者还是顾问，我们最看重的是认知能力和原始思维吞吐量。就像评估员工一样，通过与投资者相处，你能感受到他们是否从你公司的第一性原理思考问题。比如当前AI公司比传统SaaS成长更快，情况已然不同。你不想要那些只会说'按X/Y/Z套路操作'的投资者。

That's one thing. I think the other thing is that if you just think about anyone you bring into the world, this could be employees or investors or advisors or anything. What we really index a lot on is just cognitive ability, raw intellectual throughput, and you can almost feel that out in an investor in the same way you would feel it out in an employee, just by spending time with them and just like actually seeing if they're thinking about things from first principles of your company. For example, right, a lot of AI companies are growing much faster than traditional SaaS companies right now, Because that's the case, a lot of things are different. You generally don't want investors that are just like, so many reps, you gotta do things like x y z way.

你需要的是保持智力好奇心，能与你共同思考并解决问题的人。

You want people that are just intellectually curious and will think about things along with you and can help problem solve.

那么在纯粹的尽职调查方面呢？假设投资者很聪明，他们想尽快全面了解你的业务——包括优势、劣势和隐患。你见过最优秀的投资者在这个环节如何操作？比如他们的决策速度、审慎程度等具体表现？

What about on just like the pure underwriting side? So an investor comes in, they're really smart, let's take that for granted. They just want to understand your business, the good, the bad, the ugly, as fast as possible. What have you seen the best do in that side of the investment process, including things like how fast they move or how deliberate they move or anything like that?

我认为最优秀的专家，可以说他们欠我个人情，他们对我们的客户有着极为深刻的理解。这其实很有趣。我们的客户可能通过专家咨询赚了很多钱，因为太多投资者在联系他们。而最优秀的专家甚至在与你交谈前，就已经做了大量研究，对你的客户有了相当全面的了解。遗憾的是，我们发现其中存在很多噪音，因为许多人在客户访谈中撒谎——我们遇到过自称使用过我们服务却从未听说过我们的人。但总的来说，只要做足功课且方法得当，你就能通过这种方式评估业务，因为这能提供最有效的信号。

I think the best ones, I would say, they owe me one, they get a very deep understanding of our customers. It's actually funny. Our customers have made probably so much money on expert calls because so many investors are hitting them up and I think the best ones can before they even talk to you, they probably already done quite a bit of research and have like a pretty full view on your customers. Unfortunately, I think what we've seen is that there is a lot of noise there because like a lot of people just lie on customer calls and we have people who had said that they've used us and they literally never heard of them. But generally, if you do enough research and you're good at it, then you can underwrite the business that way because that gives you the most signal.

另一个关键因素是重视文化的人，因为我认为文化至关重要。许多优秀投资者深知其重要性，所以如果他们感觉你正在成为顶尖人才聚集地，人们就会更看重这一点。

And the other one is people that index on culture because I think culture is quite important and a lot of good investors know how important that is and so if they feel like you are becoming a place where good talent is congregating, then folks will index on that more.

我们来深入探讨这个问题。从文化开始谈谈招聘和文化吧。你会如何描述你们的文化？我们之前提到过你墙上的标语是第一个问题——这当然是文化的一部分。

Let's talk about that. Let's talk about recruiting and culture, starting with culture. How would you describe it? We talked about the quote on your wall is the first question. So that's part of it, course.

极度竞争意识，极强的行动偏好，使命必达。你们文化的其他核心要素是什么？比如当你面对新入职员工时，会如何描述这里的工作环境？

Extremely competitive, extreme bias to action, get things done. What are the other key components of your culture? Like when you're sitting with a new recruit, what do you tell them about what kind of place it is?

我认为保持一定程度的紧张感很重要。你必须事先明确告知求职者这一点，让他们自我筛选是否适合这种文化。所有加入十边形（Decagon）的人都会说他们想努力工作，希望与真正聪明且志同道合的人共事。他们视这里为职业生涯的黄金时期——'我会全力以赴，但因此我将与杰出伙伴建立终身情谊，获得丰厚回报，并实现职业阶层的跨越式发展'。正是这种特质吸引着这类人才。

I would say there's definitely a level of intensity that's important. And, you definitely wanna tell people that upfront because you want them to self select into this culture. So everyone that has joined Decagon, would say they want to work hard, they want to be around other people that are really smart and are like them. And they're motivated by they view this as maybe the prime of their career where, hey, I'm gonna work hard, but we know that because of that, I'm gonna build lifelong relationships with like other amazing people, I'm going to have good financial outcomes, I'm gonna be able to leapfrog steps in my career because the growth is just happening so quickly. So it's you're attracting people like that.

单是这点就为文化奠定了坚实基础——你拥有的都是来真干事的人。许多员工就住在办公室附近，这正是因为我们选择了喜欢在办公室协作的人。我们大部分时间都在办公室，这些是根基。在此基础上需要格外注意的是：我们必须确保办公室成为员工乐于每日前来工作的地方。

I think that alone creates like a pretty strong foundation for the culture because you have people that are there to work. We have a lot of people that live right next to the office and part of that is because we've selected for people that like being in the office with other people, right? We're in the office a lot. Those are the foundations. And then I think what you have to be careful on top of that is we really want our office to be a place where people enjoy coming to work every day.

毕竟我们在这里度过大量时间，你希望人们乐于在此工作，感受到全员的支持。跨部门间要让员工觉得目标高度一致，大家都在为共同使命奋斗。这是个持续优化的课题——我不认为我们已经'解决'了文化问题，这是我们长期深入思考并持续投入的领域，要确保人们觉得长期在此工作能获得高度满足感。

Because we spend a lot of time in there, it's like you want people to be happy to be there, they feel like everyone there supporting them. Between the organizations, you want people to feel like they're pretty aligned and folks are working towards the same goal. And that's an ongoing problem. Like I don't think like we've solved culture. It's something we just put a lot of thought into and we wanna make sure that people feel like they will be very fulfilled by staying here for a long time.

整个商业演进及AI领域中最引人入胜的支线剧情之一就是人才争夺战。没错，最极端的案例显然发生在Meta、Anthropic和OpenAI等公司的模型层之间，关于人们为招揽顶尖人才愿意付出何种代价的精彩故事层出不穷。能否请您从内部视角谈谈这些人才争夺战的现状？毕竟您肯定也在与其他优秀初创公司激烈争夺顶尖人才。

One of the most interesting subplots of this entire business evolution and in and around AI is talent wars. Yeah. Obviously, it's happening in the most extreme cases at the model layer between Meta and Anthropic and OpenAI, there's great riveting stories to hear about the lengths people will go to to secure a great and the amount they'll pay to secure a great engineer or someone really key to the business. Can you give us your perspective on these talent wars, what it's like to be in them? Obviously, I'm sure you are too fighting for the best talent versus lots of other great companies that are being formed.

是的，请从内部人士角度描述下这种竞争环境的真实感受。

Yeah. Tell us from the inside what this environment feels like.

这确实需要团队全力协作。想要招揽任何人才，整个团队都必须全力以赴才能赢得那些炙手可热的目标。就像销售工作一样，你需要说服对方'这里才是理想归宿'，每个环节都必须做到完美。

Definitely feels like talent is a big team effort. For anyone you want to hire, you need the whole team to swarm around them to win, to win highly sought after talent. And you have to go and do all the things right. There are some parallels to sales, of course. You're trying to like convince people that's, hey, this is the the place they want to be.

因此往往需要了解其家庭状况、伴侣需求，真正把握他们的职业诉求，并量身定制合适岗位。不过在应用层领域，竞争还没达到Meta和OpenAI那种白热化程度——顶级研究者毕竟有限。我们团队拥有实力强劲的研究者，但更多从事应用型研究。

So that involves oftentimes getting to know their families, getting to know their partners, really figuring out what they want out of their own careers and making sure that you can design a role for them that is like that. Unfortunately, in the application layer, it is not as crazy as the metas and the OpenAI's. There's only so many top level researchers. We have some very strong researchers on our team. It's a lot more of like applied type research.

对我们而言同样如此。团队中许多人来自哈佛、MIT和斯坦福，但旧金山地区愿意坐班的这类人才始终有限，竞争异常激烈。值得庆幸的是，如今我们的雇主品牌影响力已远超创业初期，招聘确实变得更容易了——尽管所需人才数量也大幅增加了。

For us, it's the same thing that happens. We're going after, you know, a lot of people on our team, Harvard and MIT and Stanford. There's only so many of these folks that are in the market anytime who are like in SF or going to the office and so it is a competitive place to be. I think we're kind of fortunate and now our talent brand has gotten a lot larger than when we were first starting, so it definitely has gotten easier to hire. But at the same time, our like, the amount of people we need to hire has also gone up.

我们持续探索新的人才渠道。比如刚成立的纽约办公室，部分原因正是看中当地人才储备资源。

Constantly just finding ideas of how to get new people. I mean, we just opened up our New York office as you know, and part of the reason for that is that, hey, there's another talent pool over here, we should leverage that.

关于贵司这类企业，我认为大众最关注的问题在于：是直接使用现有大语言模型，还是基于/替代它们开发自有模型？你们积累的独家数据资产无需与他人共享，而优质数据正是模型发展的命脉。展望未来五年，您认为会演变为完全自主模型、混合模式，还是在最佳模型上构建自有数据上下文？这个维度您如何看待其发展轨迹？

Of the questions that I think so many people are interested in for companies like yours is the use of whatever core underlying LLM versus the development of your own models on top of or replacing those underlying LLMs. You are gathering all this incredible data that's just yours, you don't have to share with anybody else. These models thrive on good underlying data. How do you think about that aspect of all of this, where five years from now it's going be either your own model or your own model plus something else or just your own data and context on top of the best model. How do you think that will evolve?

我对此既好奇它如何影响产品，也好奇它如何影响你们的商业模式、业务中的权力分配——你们依赖或不依赖GPT的程度等等。

I'm both curious about this in terms of how it impacts the product, but also how it impacts your business mode, your power in your business where you rely or don't rely on GPT, whatever.

我们刚开始时大约是两年前，人们还在探索应用场景。几乎没人做微调。事实上，当时任何尝试微调的人都会看到大量文章说微调'效果不佳'，因为它带来的提升有限。另一个当时不做微调的重要原因是模型迭代太快，你还在摸索用例阶段，何必投入大量时间做不可逆的微调？下次新模型发布时就得全部重来。现在开源模型已经发展到虽然不足以处理所有事情，但在许多特定场景下确实不需要那么高的智能水平。

When we first started, this was about two years ago, people were still figuring out applications. Almost no one was doing fine tuning. In fact, anyone that was doing fine tuning, there's a lot of writing at that time was like fine tuning quote unquote doesn't work because it doesn't really get you that many gains. Another big reason to not do fine tuning at that time was that the models are changing so fast, you're still kind of figuring out your use case, why invest much time in the fine tuning, it's not reversible, you're gonna have to throw it out next time there's there's a new model release. I think now the open source models for example have gotten to the point where definitely not smart enough to do everything, but there's a lot of specific use cases where you just don't need that much intelligence.

以智能体为例，它首先要做的可能只是根据用户当前和历史对话决定执行路径和所需数据。这种场景完全可以用微调模型实现。这个模型不需要擅长数学或编程，只需要针对特定任务的小型微调模型。现在这类应用越来越普遍，因为整体生态已经更成熟了。

Example would be even the agent, let's say the first thing the agent does is it just needs to think about like, okay, based on what the user said and everything before, what path do I go down? What data do I need? You can make that fine tune model. That model doesn't have to be good at math or coding or anything, it's like you just need a smaller model that's just fine tuned on that. Nowadays we're seeing much more of that because a lot of the applications have gotten more mature.

当你清楚智能体的架构后，就能在需要模型的环节部署小型微调模型，这能从性能和延迟等方面提升整个系统。我认为未来这种趋势会持续增强。当然，OpenAI和Anthropic这类顶级模型永远会有巨大需求，毕竟某些场景需要最高智能。所以未来会是平衡状态，但短期内小型微调模型的应用会越来越多。

So you know how your agent is structured, you know the places where you need models to run, and you can take, you know, smaller fine tuned models and that improves the entire system both in terms of performance, but also latency and so on. So I think over time there's gonna be more and more of that. I do think there's always, there's still always gonna be a huge usage of the OpenAI's anthropics of the world because you just need intelligence, you need the best models. So I think there's gonna be a balance, but at least in the short term, there's getting more and more of the fine tuning small models that happen.

你对那些科技巨头有何看法？整个市场理所当然地聚焦在七八家最大的科技公司上。除OpenAI和Anthropic外，像微软、亚马逊、苹果这类公司中，哪些对你而言最为重要？你目前与它们的关系和思考是怎样的？

What is your perception of the very biggest companies? The entire market has been focused on, rightly so, the seven, eight, nine, ten biggest technology companies. Which of them feel most important to you? And I don't mean open AI and anthropic, I mean like Microsoft and Amazon and Apple and these sorts of companies. What is your relation and thinking about them today?

它们以绝对优势主导着股票市场，是举足轻重的企业。你如何定位与它们的关系？

They've been the driver of equity markets by a huge margin. They're important companies. How do you relate to them?

从工作角度关联不大，但我个人很看好谷歌。原因在于AI应用场景中，拥有直接面向消费者的入口至关重要——这才是数据源头所在。而谷歌在这方面优势明显。

I mean, work wise, it's not super relevant, but of course, we have our own opinions. I'm very bullish on Google actually. Why? I just think that's with AI use cases, having individual like consumers is so important because that's where all the data comes from anyways. And Google is like much stronger there.

是的。我的意思是，可以说Meta、Facebook也有这种特质，所以，也许他们的新超级智能实验室能让它运作起来。但像Anthropic这样的公司，比如他们在消费者端表现不如Chatty BT那么好。我认为长期来看确实需要消费者的支持，因为那才是新数据的来源。没错。

Yeah. I mean, you could say that Meta, Facebook also has that element, and so, yeah, maybe their new super intelligence lab will be able to to make it work. But something like Anthropic, for example, where they haven't done as well on the consumer side compared to like a Chatty BT. I think long term you do need that consumer buy in because that's where all the new data is gonna come from. Yeah.

谷歌，我是说，谷歌拥有一支了不起的团队，虽然我们有过几次坎坷的开始，但希望他们能成功。当然，所有大型科技七巨头基本上都非常强大，所以我们对其没有特别强烈的看法。

Google and I mean, Google just has an amazing team and we've had a couple of rocky starts, but you know, hopefully, they'll make it work out. Of course, all the large max seven basically are obviously super strong, so we don't really have strong views on them.

如果我让你明天构建一个投资组合，五个位置，每个20%投资于五家围绕AI领域的私营公司，你会怎么构建？不包括Decagon？

If I let you build a portfolio tomorrow where you got five slots, 20% each in five private companies building in and around AI, what portfolio would you build? Decagon excluded?

对，不包括Decagon。让我想想。倾向于人才聚集的地方。我之前提到过和Cognition团队关系密切。

Yeah. Decagon excluded. Let's see. Just kinda gravitate towards where most of the talent is forming. I mean, I mentioned close with Cognition guys.

Cognition肯定会在其中。Cursor也是。我其实对它们未来可能的交集挺感兴趣的。

Cognition would be in there for sure. Cursor also. I'm actually kinda interested in like where those might run into each other in

未来。让我想想。还有三个位置。

the future. Let's see. Three more slots.

对我来说，肯定会想在硬件层下注，尽管风险更高。上次我们讨论过Etched这类公司，我觉得应该会选一家放进去。虽然还处于早期，但潜力巨大。

For me, definitely would wanna take a bet on the hardware layer even though it's like a much higher variance. So last time we were talking about etched, like companies like that, I think you probably put one of those in there. Still on the earlier side, but obviously very high potential.

还有两个。目前为止我很喜欢这个投资组合。

Two more. I like the portfolio so far.

我另一个朋友正在创办一家叫Pika的公司，开发视频模型。我对那个团队评价也很高，所以他可能把他们列入了名单。最后一个，可能需要在底层模型方面下些赌注，尽管现在市面上已有许多大型语言模型。但我另一位非常看好的朋友正在构建模型，不是语言模型那种类型，而是针对比如医疗保健等领域的基础模型。我朋友乔什正在开发Chai。

Another friend of mine is is building a company called Pika, building video models. So I just think very highly of that team as well, so he probably put them in there. And the last one, probably would want some sort of bets on the underlying model side, even though all the large language models are out there. But another friend of mine, I think, very highly of, they're building models, but not like the types of language models, but still foundation models for like health care, for example. So a friend of mine, Josh, is building Chai.

他们正在构建一个基础模型。这类项目我觉得相当有意思。甚至我会把洛克希德公司的实体项目也加进去——这些都非常令人兴奋。只不过目前还很难预料它们最终会如何发展。但如果我要构建一个投资组合，肯定会把这类项目纳入其中。

They're building a foundation model. So stuff like that, I think, is quite interesting. Or even I would put Lockheed's company, physical I think those are very exciting. It's just, obviously, it's early to see how those will turn out. But if I'm building a portfolio, I definitely want one of those in there.

你对这个领域的两极分化怎么看？我同时关注处于不同位置的人——那些既不在你的位置，又兼具技术背景与商业嗅觉，身处这波浪潮中心的人们。他们对于当今AI能力的哪些方面存在高估或低估？哪些进展比大众认知更超前？哪些又相对滞后？

What do you think are I'm interested in both ends of the spectrum. The people that aren't in your position who are both technical and commercially at the center of this wave, what do they overestimate and underestimate about the capabilities of AI today? Where is it further along than the world thinks? Where is it further behind?

在解锁企业级应用场景方面，我认为AI实际上比人们想象的更滞后些，主要因为非确定性问题。需要解决两个关键点：首先必须改变人们对智能体的认知方式。这里存在一个悖论——明明AI驾驶员的错误率远低于人类司机，但由于这是新技术，人们对它的容错标准反而更高，要求近乎完美。

It's a little bit more behind in being able to unlock a lot of the enterprise use cases, I would say, because of the non determinism. So two things that need to happen. One, you need to reframe the way that people think about agents. There's the way more effect that often happens where it'll be objectively just way better than human drivers and human drivers make a lot of mistakes. But because we're investing in new technology, the bar's a lot higher, so it has to be near perfect.

这种认知偏差需要调整：评估AI时不应只盯着错误，而要看整体成功率。如果能建立这种认知框架，就会发现AI的成功率确实远超人类——毕竟人类本身也不完美。但目前企业界尚未完全接受这种思维，这是需要发生的转变。另一方面，AI模型在许多领域仍需改进，对吧？

So there's that dynamic that kinda needs to adjust in some folks mind where instead of evaluating AI in a way where you're just trying to find mistakes, you're evaluating holistically looking at the sort of success rate. And then if you can frame it that way, well, success rate is gonna be way higher than humans because, again, humans are not perfect. So I think that needs to happen, and I don't think that's fully happened yet in the enterprise. So that trend needs to happen. And then on the AI use case side, I think the models still need to get better in lot of areas, right?

我们刚才讨论过语音交互技术。当前模型的幻觉率还是太高。随着模型进步，更多企业应用场景会被解锁。但公众看到炫酷演示时可能会误以为语音技术已经成熟，导致企业高管们趋之若鹜。可实际落地时，就会发现部署难度远超预期。

We were just talking about voice to voice earlier. Hallucinations are too high there as those models get better. I think more enterprise use cases will be unlocked, but I do think the general public will just see a really cool demo and be like, okay, oh wow, voice is solved now. And as a results, you know, the enterprise should be adopting it left and right. And C suites will see that too, but then when they actually get into it, it's just harder to go live.

我想说的是，这部分尚未完全成熟，因此进展比人们预期的要慢一些。而另一方面被低估的是，事物正以指数级速度增长和改进。但没人真正理解指数增长意味着什么。这可能体现在性能或成本等方面。

That's I would say the piece where it's not quite fully there yet. And so it's a little bit slower than people think. I think the thing that's on the flip side of that where it's underestimated is just things are growing exponentially, improving exponentially. And no one is good at conceptualizing what exponential means and things like this. And so that could be from like a performance like cost perspective.

目前我认为，开发应用时利润率不该是首要考虑。人们常批评编程代理工具在烧钱，但别忘了技术改进是指数级的，成本同样会呈指数下降。所以零利润率其实无关紧要。

Right now, I would argue that if you're building an application, your margin shouldn't really matter that much. People will often critique the coding agents like they're hemorrhaging money. Yeah. But again, things are improving exponentially and like the cost will go down exponentially as well. So zero margin doesn't really matter.

真正关键的是抢占市场份额和用户心智。现阶段利润率不佳完全没问题，因为一切进步速度都远超预期。企业级市场稍有不同——原则相同，但企业合作周期长，即使成本下降，客户预期也可能变化，所以不宜持续大额亏损。

What really matters is you need get market share and you need to get mind share of users. Perfectly fine not to have good margins right now and it's just everything just improves way faster than people think. It's slightly different at the enterprise. I mean, the same principle applies, but generally don't want to be hemorrhaging cash with like an enterprise deal because those are just much longer term. And even though the cost will go down, their expectations might also change and so on.

因此企业级业务需要保持相对稳健。这才是当前最被低估的要点。

So you generally wanna be fairly healthy there. I think that's what's underestimated right now.

你如何确信成本会大幅下降？我对这个利润率问题很感兴趣——如果能确定成本将下降98%（比如服务编程代理的成本），那理论上应该追求极致负毛利率，全力获取用户基数，建立产品优势与用户粘性。但这完全取决于你对成本下降的信心。你的判断依据是什么？

How do you know that costs will get way lower? Like, I'm very curious about this margin question because if you knew for sure, then we actually wanna run super negative gross margins. If you knew that it was gonna get 98% cheaper or something like this, cost of goods to serve a coding agent or something like that, I think the argument would be get the install base, just like have the best product and get the users and build the affinity with that user base and don't care at all. But that hinges a lot on the confidence that you have that the cost will fall. So how do you know?

你如何看待'抢占用户基数'与'当下展示良好单位经济效益'之间的平衡？

How do you think about that equation of win install base versus demonstrate good unit economics now?

很简单，我们目前所处阶段绝不可能是技术巅峰。行业投入巨大，效率是核心指标之一。即便技术停滞，仍有多种系统重构方案可降低成本。只是现阶段把时间用于获取新客户更划算，因为未来必然变革。有些人把这看作拿风投资金补贴芯片公司的把戏，但本质不同。

Well, just think it's quite unlikely that where we're currently at is like the best that things will be. There's so much effort getting put into it and like one of the main metrics is efficiency. The other piece is that even if things don't get better, there's a lot of ways you can rearchitect your system so that it is more cost efficient. It's just that it's not worth putting time into that right now where you can put that same amount of time into getting new customers because you know that things will change in the future. So I think that's just one thing where people are like, oh, it's kind of like a a scheme where you take VC dollars and then like the VC dollars go to the chips.

这些公司正在亏损。如果他们愿意，或许只需花费一个月甚至更短时间就能大幅提升利润率。但目前做这些工作并不值得。你们现在真正优化的是质量和增长。只要能做到这点，后续的优化自然水到渠成。

These companies are losing money. If they wanted to, probably they could just like spend a month or even less and just massively improve their margins. But it's just not worth doing that work right now. What you're really optimizing for right now is just quality and growth. So if you can do that, then the optimization will always come later on.

你如何看待利润率问题？比如就Decagon而言，你们是否在意？是否设定目标？你们对可接受范围或目标值有哪些底线或参数？

How do you think about your margins? Like just putting it on Decagon, like, do you care at all? Do you set them? What's your guardrails or parameters for what's acceptable or what you're targeting?

是的。我们唯一的原则性标准是避免负利润率。总体而言我们利润率相当健康，可以这样理解：以任何商品的供应链为例，比如你在机场买可颂面包。最终解决用户需求的环节通常能获取最大利润。从面粉加工到黄油生产等中间环节，每个步骤都是在成本基础上赚取利润。

Yeah. I would say the only thing that we have a principle for is not to have negative margins. In general, we have fairly healthy margins because one way you can think about it is if you just think about supply chain for any good, let's say you're buying like a croissant at the airport or something. That last step where you're actually solving someone's problem, that's where you generally can capture the most margin. Every step along the way like whoever's enriching the flour or like making the butter or whatever, you're generally making a margin on top of the costs of whatever your goods are.

这就是应用层的优势所在——你们构建的产品实际解决商业问题。因此能获取更多利润，因为客户的考量逻辑是：我们在投资Decagon，并不太关心你们的成本（甚至完全不在意），真正重要的是商业投资回报率——比如运营规模缩减多少，AI如何通过用户互动带来收入增长。

And that's why it's nice to be in the application layers, what you're building is actually solving the business. And as a result, you can capture more of that because our customers the way they're thinking about it is, great, we're investing in Decagon. We don't care that much about what Decagon's costs are, in fact, we probably don't care about at all. What we care about is what is the business ROI that we're getting. It's like we're downsizing our operations by this much, we're actually generating more revenue now because the AI can engage people and keep them retained.

这就是我们最具活力的领域。坦白说这并非新颖观点。早期会有很多ChatGPT外壳产品，如果功能太单薄就难有价值。但若围绕模型构建足够软件生态，反而能捕获最大价值。

So that is where we probably see the most dynamic here. And I do think this is generally not really a hot take. In the early days, you're like, oh, there's just like Chateappity wrappers and so on. And, yeah, a lot of wrappers, if it's too thin, it's like gonna not gonna be valuable. But if you have enough software built around the models, then that's where you can actually almost capture the most value.

所以我认为OpenAI这类公司终将转向应用领域。长期靠API盈利很困难——实验室竞争激烈，用户切换成本极低。从AWS迁移到GCP尚且不易，但更换OpenAI到Anthropic模型只需修改一行代码。

That's why I think the open AIs of the world will continue to move towards applications. Because it's quite hard for them to make money long term on like their API, for example. Because there's like such high competition, all the labs are building. It's very easy for people to swap. It's not like moving from AWS to GCP is like very hard, but moving from a OpenAI model to Anthropic model, you just change like one line of code.

能否详细说说这个ChatGPT外壳概念？我的理解是，不仅你们的产品，很多公司工程师的工作并非AI研发，而是传统软件开发——将系统对接企业客户，这些与OpenAI或Anthropic提供的服务截然不同的产品与基础设施建设。

Can you say a little bit more about this ChatGPT wrapper concept? My sense is, certainly for what you built, but probably for other companies that a lot of the work that your engineers are doing is not AI work. It's traditional software work. It's the ability to hook this system up to enterprise customers. It's good old fashioned product and infrastructure building that is very different from what OpenAI or Anthropic are providing you.

这确实是项艰巨的工作。就像构建任何软件系统一样，都是辛苦活。人们总痴迷于一种幻想，仿佛五年后我

And that's hard work. Like, just like building any software system is hard work. Everyone's very enamored of this idea that, like, in five years, I

只需向你展示一个

can just show you a

软件片段，然后告诉编程代理：复制这个软件。这样就能降低软件成本。我不认为你是这么想的。或许可以描述下你对这个Chatty Bitty说唱式担忧的看法或批评？

piece of software and just tell a coding agent, just, replicate this piece of software. And that's gonna mean lower software motes. I don't think you think of it that way. So maybe describe your thinking or critique of this Chatty Bitty rapper concern that people have.

是的。通常人们用'说唱'这个词带有贬义，就像说'嘿，你不过是个说唱歌手'。这并非非黑即白。确实有很多应用只是简单套壳，无法成为真正的业务，因为它们创造的价值有限。

Yeah. I think generally people say rapper in a derogatory way. It's like, hey, you're just a rapper. It's not black and white. Yes, there are a lot of apps that are just rappers that are not gonna become real businesses because there's just not that much value.

比如文案写作领域——虽然我不太了解这个行业——但从外部看似乎很困难，因为用户可以直接登录ChatGPT说'帮我写这个'。如果某个工具缺乏足够的附加功能使其真正有价值，那么直接使用基础模型可能更简单。但大多数情况下并非如此，特别是涉及智能代理时。代理不仅仅是模型，需要设计架构、设置防护机制、教导新技能。

I mean, one argument, don't know too much about this space, but at least from the outside, it it has seemed like copywriting, for example, has been difficult because someone can just log in to ChatGPT and just like, hey, write this for me, and I'll just write it. There are things where if there's not enough tooling or functionality on top of something for it to be really valuable and needed, then maybe it is easier to just leverage the models. But most of the time, that's not the case, especially when you get into agents. An agent is not just a model, right? You have to like design it, you have to be able to put in guardrails, you have to be able to teach you how to do new things.

这正是模型之上软件层的价值所在。如果这个软件层足够重要，那么模型更新就很难直接淘汰你。另一方面，现在实验室对构建应用很感兴趣，他们开发各种编程应用如Cloud Code等，这些将成为Cognition或Cursor的竞争对手。或许正因如此，编程代理开始自主训练才是明智之举。

And that's where the software layer on top of the models come in. And if that's valuable, then it's just much harder for you to just like be made obsolete by a model update. The other side of it is like, okay, well now the labs are quite interested in building applications. So they're building a bunch of like coding applications and cloud code and so on. And so those will end up being competitors with Cognition or Cursor and and maybe for that reason, it is wise for the coding agents to start training stuff as well.

就我们领域而言，目前我们产品自上而下的功能体量非常庞大——这与AI完全无关。比如：如何监控对话过程？如何预警异常流量？如何建立对话质量评估和单元测试机制确保上线前的可靠性？这些纯粹的功能需求与AI无关，但都需要大量开发工作。

I think in our space, I would say for us right now, at least the sheer amount of functionality of the build because it is a very top down product is quite large, that has nothing to do with AI. It's just stuff like, okay, how do you have like observability into like what the conversations are? How do you alert the team if something spikes? And how are you able to QA and have unit tests for the conversation so that before you push it out to end users, feel it's like there's just all this like functionality that's there that doesn't really have anything to do with AI really. And so it's just like a lot of stuff that needs to be built.

对于正在考虑目标客户群体的其他创始人，你会给出什么建议？你们最佳客户最有趣的特质是什么？在筛选客户时，由于时间有限，你们只能服务特定数量的人群。我知道你们发展非常迅速，但在任何特定时期能服务的客户数量终究有限。你们如何确定想合作和不想合作的客户？

How would you advise other founders thinking about the ideal customers to go after? What are the most interesting qualities of your best customers? When you're qualifying them, are they gonna be you have limited time that you can only serve so many people. I know you're growing really fast, but you can only serve so many customers at any given time. How do you qualify who you wanna work with and don't?

比如，你观察到哪些特质是最关键的？

Like, what are the attributes that you've seen matter the most?

是的，我们希望合作对象在领导层具备知识层面的素养，对技术充满真诚的好奇与热情。实际上在企业客户中，这种特质的表现差异很大。在我看来，最优秀的领导者——也是我们最期待合作的类型——他们往往直接表示'我们希望全力推进AI应用'，对我们的系统运作原理表现出浓厚兴趣，并愿意主动破除官僚障碍推动项目落地。

Yeah, we want people that are intellectually just at the leadership level, really curious and excited about technology. You actually see a huge spectrum of that in the enterprise. And some of, in my opinion, the best leaders and like probably the folks that we are most excited to work with, They're just genuinely like, hey, we want to move on AI as fast as possible. We're very interested and just curious about how all your systems work. And as a result, I'm gonna help cut through all the crafts and like bureaucracy to get something going.

这种特质通常在初次对话中就能明显感知。你能分辨出对方是真心实意要全力推进项目、同时会提供高质量反馈的类型，还是仅仅把AI当作董事会强制要求的待办事项来应付。

And I think you can tell that pretty clearly in the first conversation. You can tell if someone's if they're like legit about this is something where I'm gonna both push aggressively, but also give you a ton of feedback and like the feedback is gonna be good. Versus someone where they just know it's like a board mandate and it's just like AI is like a thing on their to do list.

你遇到过特别能体现这种态度的领导者典范吗？

Is there any leader that you've come across that most exemplifies that posture?

有的。在数字原生企业领域，比如Chime团队就令人印象深刻——从高层到基层员工都堪称典范，值得作为案例研究。他们在企业文化塑造上做得非常出色，让全员对AI代理的每个数据决策都保持高度专业，严格追踪所有指标，对系统更新方式也极其审慎。

Yeah. So I will say in both the sort of digital native segments, you have folks like Chime, for example, super impressed with the entire team top down. Everyone that is in the org, that's like a case study to study at some point. Think they've done a really good job on culture, really making it so that people are very sophisticated about everything. Everything the AI agent does, it's like super data centric, they track everything, they're really thoughtful about how to to make updates and so on.

他们还为我们提供了大量优质的产品反馈。

And they've given us a ton of good product feedback.

当你与同样在这个领域创业的朋友们交谈时，你觉得在哪些方面你的世界观与他们最为不同？你对事物的看法或对某些事物的相对兴奋点在哪里与其他人差异最大？

As you're talking to your friends that are also building companies in this space, where do you feel that your worldview is the most different from them? Where's your view of things or your relative excitement about something the most divergent from from others?

是的。我想说，我的朋友们都是智商极高的人。但我觉得自己通常更倾向于...不知道这是否算性格使然，我往往更关注每个想法中的商业元素。

Yeah. I mean, I would say my friends are very high intellectual horsepower. I think I generally lean a lot more. I don't know why this isn't, it's like personality. I think I generally lean a lot more towards commercial elements of every idea.

再次强调，这不是创业的唯一方式，但在我看来，如果你试图优化成功的最大可能性，我认为你应该真正重视商业层面。因为我们见过太多超级聪明、求知欲强的人，他们只是在构建非常酷的项目。虽然可以说很多巨大成果需要这样做，因为有些事必须有人去做——毕竟这些事的商业元素在哪里呢？但我觉得Ashwin也是这种风格，我们配合得很好。

And again, it's not the only way to build the company, but in my opinion, you were just trying to optimize for like the highest likelihood of success, I think you should really index on the commercial side. Cause we've lost super smart, intellectually curious people that are just building very cool projects. And you can make the argument for a lot of huge outcomes, you need to do that because you just need to do stuff where like, no one's gonna work on this because where are the commercial elements of this. But I think I think Ashwin's this way too. It just made a good fit.

我认为我们就是非常专注于一点：你必须极其务实。

I think we're just like very locked in on, okay, you just gotta be like super practical.

你愿意为此支付多少钱？没错。这确实很有意思。你们是如何标记业务里程碑的？

How much would you pay for it? Exactly. Yeah. It's really interesting. How do you mark milestones in the business?

你们如何激励团队？在凝聚团队围绕特定目标方面，你学到了什么？我知道当你有一个想要争取的客户时，你会非常激进——这不仅仅是老套的销售流程，而是全员出动。派出工程师，不惜一切代价。

How do you motivate the team? What have you learned about how to rally around a given thing? I know you're super aggressive when you have a customer that you wanna get that it's not just like a old school sales process. It's like an all hands on deck. Send engineers, do whatever it takes.

在激励里程碑、凝聚团队、围绕共同目标组织方面，你学到了哪些经验？

What have you learned about motivating milestones, rallying the team, organizing around common goals?

是的。对我们来说，我们总会做的一件事就是确保视线范围内有个旗杆般的存在，能把大家凝聚起来。这种凝聚力载体非常有用。前几天我还在思考这个问题，其实竞争也是另一种形式的凝聚力。

Yeah. For us, I mean, one thing we do always is we always have like a flagpole that's within sight that can just kind of rally everyone around it. Having things to rally around are quite helpful. I was kind of thinking about this other day. Think another type of rallying is around competition.

对吧？当人们感觉身处战场且有明确对手时，这种竞争就很有意义。当然不能让竞争演变成真正的敌意，但适度竞争非常健康。它能让团队更团结，因为有了共同聚焦的目标。里程碑也是如此，如果你给团队设定清晰可见的里程碑——甚至可以是非常重大的那种。

Right? I think when people feel like they're in a battle and there's like clear enemies, then it makes sense. Again, you don't want it to get to the point where there's like active animosity, but it's a very healthy level of competition. It it ties the team together because there's like something to focus on. Same thing with milestones, like if you give someone a very more clear milestone, and this can be like pretty significant.

比如去年我们设定营收里程碑时，承诺给全员购置高端夹克——最后买了Decagon Arcterix的，所有人都为此兴奋不已。想想看，夹克的成本相比员工薪资简直微不足道，但它创造了'我们正在为这些夹克奋斗'的集体动力。

Like last year we had for our revenue milestone, we told everyone we get them like super nice jackets, we got like Decagon Arcterix jackets, and everyone was like super excited about that. And if you just think about like The cost. Yeah. The cost of the jacket and just how much people get paid is like trivial. But it just creates like this, hey, we're working towards these jackets.

这能让团队紧密联结，因为大家感觉是在为共同目标努力。寻找下一个里程碑已成为我们企业文化的重要部分。

It brings the teams together because now it just feels like everyone's working together towards this common goal. That's been a big part of our culture, just finding what is the next milestone.

关于商业领域或AI应用世界，还有什么我们没谈到但您特别热衷的话题吗？我觉得我们已经讨论得很全面了。

Anything that we haven't covered about either the business or this exciting AI applications world that you feel especially passionate about? I feel like we've done a good job of covering.

这次对话非常棒。我觉得现在AI初创公司圈有两个被过度炒作的现象：一是全员现场办公，搞什么996工作制——我个人认为996并不健康。

It's been a great conversation. I think what has become almost a meme or hyped up a lot in AI startups right now are like a couple things. One, obviously, everyone's in person. It's like, you know, nine nine six or whatever. Don't actually think nine nine six is that healthy.

虽然在中国这种现象很普遍，人们也显得特别拼，但部分原因是那边就业机会稀缺，雇主自然占据优势。在美国更应该保持工作生活平衡，高强度工作后需要适当放松。另一个被热议的是'前线部署工程师'概念，现在所有人都在讨论这个。

It happens in China and everyone's like super hardcore, but one of the reasons is that no one has jobs over there, so it's very easy for employers to have leverage. I think here, you generally wanna maintain a good balance because if you're working super high intensity, you need time to relax a little bit. But that is one element. The other element is the forward deployed engineers. So everyone's talking about forward point engineers.

而且，我觉得这有点好笑，因为我的联合创始人来自Palantir，他们实际上有前线工程师。在Palantir，前线工程师意味着你在处理一笔10.25亿美元的交易。所以你基本上就是全职为一两个或少数几个客户工作，专门为他们定制开发。我觉得人们有点把这个和初创公司的做法混淆了，初创公司只是非常亲力亲为，做一些无法规模化的事情。但要真正采用FTE模式，你需要有大客户。

And, yeah, I think it's kinda funny because my cofounder came from Palantir, so they actually have forward point engineers. What a forward point engineer at Palantir means is you're working on a $1,025,000,000 dollar deal. And so you're actually just almost full time working with either one or a small number of customers and like building very specifically for that. And I think people are a little bit conflating that with what startups do, which is startups are just very hands on and do things that don't scale. But for you to actually have a FTE model, you need to have massive clients.

大多数人并没有大客户。所以我其实很感兴趣想看看这会如何发展，因为我认为现在对前线部署工程模式有点过度追捧，比如明明交易规模只有5万美元，却要配一个前线工程师。这也是我们经常思考的问题。我们与客户合作非常紧密，但我觉得必须全局考虑问题，必须思考如何快速扩展规模。

And most people do not have massive clients. So I'm actually I'm kind of interested to see how that plays out because I do think there's like a over indexing on this forward deployed engineering model right now, where, yeah, I have a four point engineer and the deal sizes are like 50 k. And that's something we think about a lot as well. We are very hands on with our customers, but I think it's like you have to think about things holistically. You have to think about like, okay, well, how do we scale quickly?

我们没有5万美元级别的客户，但如果你为每个5万客户配备一个专职人员，那根本不可能规模化。所以需要在中间找到平衡点。我认为完整的前线部署模式只适用于Palantir那种方式。

We don't have 50 k clients, but for every 50 k client you have like someone that's like fully staffed to them, that's like impossible to scale. So you need to find that line in the middle. And I think the full four deployed model only works with the Palantir approach.

想必在你们这边，你们非常在意利润率。如果只是像LLM成本这类开支，你们会更公开透明。但如果涉及完整的人力成本，那就会成为问题。

Presumably on this side, you do care a lot about your margins. Margins you're much more open about if it's just like LLM cost or something like this. But if it's fully baked people costs, like that's gonna be a problem.

是的。这甚至不单纯是利润率的问题，而是会阻碍规模化发展。没人能那么快地招到优秀人才，找到合适的人本来就很难。

Yeah. It's not even a problem necessarily from the pure dollar margins. It's just prevents you from scaling. No one can hire good people that fast. It's just hard to hire good people.

所以如果你的业务完全受制于人才短缺，那也不是好事。

So if your business is fully constrained on good people, then that's also not a good thing.

你认为要证明前线部署工程师模式的合理性，客户规模和收入的最低门槛应该是多少？

What do think the minimum customer size and revenue is to justify a forward deployed engineer model?

哦，大概有一百万次吧。

Oh, like probably a million.

是啊是啊，真有意思。我想你知道我惯常用来结束对话的问题——别人为你做过最善良的事是什么？

Yeah. Yeah. Yeah. Fascinating. Well, I think you know my traditional closing question for everybody, what is the kindest thing that anyone's ever done for you?

如我所说，我们的共同朋友斯科特提前提醒了我，所以我认真思考过。小时候，大约五到十三岁，小学初中那会儿，我基本上是个懒散的孩子。我觉得大多数孩子都这样，很少有人天生就自我驱动力极强。

As I mentioned, our mutual friend Scott gave me a heads up here. So I did put a lot of thought into it. So when I was little, call it ages five to 13, like elementary, middle school, pretty like lazy kid in general. I think most kids are. There's very few people that are just like intrinsically self motivated.

那时我只想整天打游戏、和朋友闲逛或运动。我父母对我和妹妹的教养方式很特别——小时候对我们实施极端严格的管教，每天练钢琴三四个小时，参加四项比赛，甚至直接让我停学专攻。

I wanted to just like play games all the time or just like hang out with friends or play sports. And, yeah, my parents had like a very interesting way of raising us, me and my sister. What they did was basically when we were really little, it was like an extreme level of discipline. When I was little, played a lot of piano, essentially like three, four hours a day and then four competitions. They just pulled me out of school and just go hard at it.

后来父母认为数学可能更适合我，而我恰好有些天赋。于是全家全力投入：家里没有电视游戏机，假期也基本不过，所有事情都为专注一件事让路。

And then I guess fortunately for me, my parents decided, okay, math was probably a better way to go. And I was just like quite talented at math when I was little. Even for that, it's just like full force, like, you're just committing everything. We did not have TV in the house, we didn't have video games, we didn't really take vacation growing up. You're kind of in this mindset of you're sacrificing most things to focus on one thing.

孩童时期因为没有参照系，其实不觉得苦，毕竟标准都是父母定的。现在回想起来，我的童年很快乐。那种纪律性和竞争意识，成年后很难再培养——童年时大脑还在发育，这些会塑造人格根基。

When you're a kid actually, like, you have no frame of reference, so it doesn't feel hard necessarily. It's because your parents are kinda setting up the criteria for you. And in hindsight, I had a very happy childhood. But I think that level of discipline and also just competitiveness, it's very hard to gain that after your childhood is over. Because when you're in your childhood, your brain's still forming, so that kind of forms your personality.

对此我非常感激。这也解释了为什么我们这代移民子女普遍出色——很多父母是读研时来美的，现在都和我同龄。独特的是，我父母虽严格却从不专横：不会阻止我们想做的事，也不会强迫我们接受他们的安排。

So one, I'm very grateful for that, and I think that's also why when I talk about my generation of there are a lot of immigrant parents from my generation that came over grad school, and they're all around my age, and I think this, like, our crop of folks just are doing very well, probably because of that, probably because of the upbringing. And then what my parents did, I think, which is the more unique side is that a lot of times what happens, especially it's like the stereotypical Asian parent, is that it just like continues. And you just have like overbearing parents. I would say even though my parents were very intense about things, they never had any like semblance of like overbearingness. They wouldn't like prevent us from doing things we wanted to do, force us to like, hey, you should like take this or whatever and so on.

事情发生在初中毕业到高中那段时间，那时我们的性格基本已经定型。我父母硬是把一个懒散贪玩的孩子，改造成了一个极其上进的人。值得称赞的是，之后他们就完全放手了。他们对我们的职业选择或人生道路没有任何干涉，表现得非常支持。

What happened was towards the end of middle school into high school, think we had already kind of established these personalities. Basically, my parents just pounded a sort of lazy, wanting to play around kid into someone that was just very, very driven. Then to their credit, they just completely laid off. They don't have any opinions on what we do for careers or what we should do. They're, like, very supportive.

我认为这种付出非常难得，因为这是金钱无法替代的。你需要的是愿意花费大量时间精心培育你童年的父母，才能塑造出这样的品格。我现在拥有的很多东西都源于此。这大概是最珍贵的馈赠。我姐姐也是如此，成长过程中她承受了很多压力。对于移民又怀揣野心的父母来说，他们在异国他乡很难自己取得成功，于是就把很多期望寄托在孩子身上。

So I think that's very kind because you cannot replace you can't even pay for that. You basically just need parents that are willing to spend a ton of time crafting this childhood for you to develop this. And I think a lot of the things I have in life right now are from that. So that's probably the kindest thing. And my sister as well, like, even growing up, she was I think there was a lot of pressure on me and in some way for, like, parents that immigrate who are also ambitious folks, it's very hard for them to succeed themselves in a new environment, in new countries, so they put a lot of expectations on their children.

是的。我姐姐总是为我想要达成的目标做出牺牲。所以现在我会尽量多花时间陪伴父母。

Yeah. My sister was also like, she was just always trying to sacrifice things for me when I was trying to achieve things. And so, yeah, try to spend as much time with her parents as possible.

你数学生涯的高光时刻是什么？

What was the climax of your math career?

绝对是高中时期。那时我们参加数学竞赛和奥数比赛。美国有个重要的赛事叫美国数学奥林匹克，顶尖选手可以参加集训营。我去过几次，那应该就是巅峰了。

Definitely peaks in high school. So in high school, we did math contests, Math Olympiads. So there's, like, a major contest in the The USA Math Olympiad in The US, and there's, like, essentially a camp for, like, the top folks. And so I went there a few years. That was probably the peak.

这次对话实在太精彩了。你们在我们大楼里创建的事业令人着迷，感谢你详细解说，带我们从多个角度深入核心。谢谢你的时间。感谢邀请。

Well, this has been so much fun. So fascinated by the business that you've built in our building. Thanks for explaining it to us and bringing us sort of right to that white hot center in so many different ways. Thanks for your time. Thanks for having me.

若您喜欢本期节目，请访问joincolossus.com，获取本播客所有集数及精心编辑的文字稿。您还可订阅《Colossus评论》——我们每季度出版的印刷/数字/私密音频刊物，深度报道我们最欣赏的创始人、投资者和企业。详情请见joincolossus.com/subscribe。

If you enjoyed this episode, visit joincolossus.com where you'll find every episode of this podcast complete with hand edited transcripts. You can also subscribe to Colossus Review, our quarterly print, digital, and private audio publication featuring in-depth profiles of the founders, investors, and companies that we admire most. Learn more at joincolossus.com slash subscribe.