Glean CEO：企业担心前沿 AI 模型会蚕食他们的业务

90% 或更多的用例现在可以由许多不同的模型（包括开源模型）完全处理。Arvind Jain，Glean 的创始人，是过去十年科技界的杰出人物之一。他之前创立了 Rubric，该公司成功上市，是一家出色的上市公司。现在他创立了 Glean，这是一家了不起的企业，已经从 Kleiner Perkins 和许多其他优秀投资者那里筹集了资金。对于几乎所有不进行前沿模型训练的 AI 公司来说，他们应该将模型公司视为巨大的资产。一旦转向消费模式，就没有固有的捆绑优势。你必须付出 10 倍的努力才能从客户那里获得相同的收入。准备好了。Arvind，我对此非常兴奋。我们有一个共同的朋友 Mimoon，他对你赞不绝口，我认为他是我们这个时代最伟大的大使之一，所以我真的很期待这次对话。感谢你接受我的邀请。

90% or greater of use cases can now be fully handled by many many different models including open source models. Arvind Jain, the incredible founder of Glean, is one of the technology luminaries of the last decade. He founded Rubric before which obviously IPOed very successfully and is a brilliant public company. Now he's gone on to found Glean, an incredible business today that's raised money from Kleiner Perkins and many other great investors. For almost all other AI companies that are not doing frontier model training, they should see the model companies as a huge asset. Once you move towards consumption, there's no inherent bundling advantage. You have to do 10 times the work to get the same amount of revenue from your customers. Ready to go. Arvind, I'm so excited for this. We have a mutual friend in Mimoon who says many, many wonderful things about you and I think he's one of the greatest ambassadors of our time and so I'm really excited for this. So, thank you for joining me.

感谢你的邀请。

Thank you for having me.

我认为对于创业者来说，要么你因胜利而兴奋，追逐胜利；要么你害怕失败，这种恐惧激励着你。你是哪一种？

Now, I think with entrepreneurs, you're either thrilled by winning and it's that chase to win or you're terrified of losing and it's that fear of losing that inspires you. Which one are you?

嗯，这是个好问题。我想我可能会说是后者。我总是担心可能出错的事情，这让我夜不能寐。

Well, that's a good question. I think I would probably say the latter. I'm always worried about what can go wrong and that keeps me up at night.

我喜欢这一点。只有偏执狂才能生存。你一直是这样吗？

I love that. It's only the paranoid survive. Has it always been that way?

是的，大部分时候是这样。

Yeah, mostly. Yeah.

即使你取得了成功，这也很耐人寻味。Rubric 取得了非凡的成功，如今是一家上市公司，而你是联合创始人之一。这种心态不会随时间改变。

Even with the success you've had, it's so interesting. Like Rubric was a phenomenal success, public company today, and you're one of the co-founders. It doesn't change with time.

不，因为我认为首先，每次你创办一家新公司或启动一个新项目，在某种程度上就像是重新开始。你从过去学到了一些好的经验，但这是一个新世界，一个新环境。想想 Glean，它在所有方面都与 Rubric 根本不同，尤其是在 AI 领域，你必须这样思考，因为每天都有颠覆。如果你开始更多地专注于在已有的基础上继续建设，那种想要加倍下注的赢家心态，我认为在 AI 新世界里是不够的。

No, because I think number one, every time you do a new company or build a new project, it's sort of like starting from scratch in my opinion. You have some good lessons from before but it's a new world. It's a new environment. Think about Glean, it's fundamentally different from Rubric in all ways possible, and especially in the world of AI you have to think that way because there is a disruption every single day. If you start to focus more on sort of keep building on what you've already built, like that's the winning mindset that you want to double down on, I think that's not sufficient in this new AI world.

能否为那些不了解的人，用 60 秒概括一下 Glean 是什么以及你们如何运作？

Can I ask for those that don't know, can you provide a 60-second summary on what Glean is and how you work?

Glean 是一家企业 AI 公司。我们最初是一家面向企业的搜索公司，帮助员工快速找到埋藏在公司 100 或 1000 个不同系统中的信息。这有点像我们起步的方式，一个工作生活的 Google。但随着时间的推移，随着 AI 模型变得更好，它演变成了一个 AI 平台。如今，可以把 Glean 看作是 ChatGPT、Claude、Gemini 的超集，所有这些都整合到一个产品体验中。它是你员工的同事，连接到公司所有的上下文，即公司内部的工作方式。

So, Glean is an enterprise AI company. We started as a search company for businesses, helping an employee quickly find information that's buried across one of 100 or thousand different systems inside their company. That was sort of like how we started, a Google for your work life. But then over time, as AI models got better, it evolved into an AI platform. Today, the way to think about Glean is that it's a superset of ChatGPT, Claude, Gemini, all of those combined into one product experience. It's a co-worker for your employees and it's connected to all of your company's context, how work happens inside your company.

Palantir 的 Alex K 上周在 CNBC 上说，全球最大的企业比以往任何时候都更怀疑前沿模型提供商。你与一些最大的企业合作，拥有令人难以置信的客户。你同意他的说法吗？他们比以往更怀疑吗？

Mr. Alex K from Palantir went on CNBC last week and he said that the largest enterprises in the world were more skeptical than ever of frontier model providers. You work with some of the largest, you have incredible customers. Do you agree with him? Are they more skeptical than ever?

两件事。第一，他们害怕这些提供商，就像每家软件公司都担心：我们还能生存吗？模型会吞噬一切吗？同样，企业领导者也担心他们的核心 IP、数据、信息以及他们的学习方式、做事方式，是否会过度依赖这些模型提供商。所以这种感觉肯定存在。但我认为他所说的是，AI 在企业中并没有发挥作用，然后每个人都害怕真正说出来，因为这不应该是件酷事……

Two things, you know. One, they're terrified of them in the sense that just like every software company is worried about, hey, will we be in business? Will the models eat it all? Similarly, enterprise leaders also worry that their core IP, their data, their information, as well as their way of learning, their way of doing things, will they be subject to too much technology dependence on these model providers. So that feeling is there for sure. But I think what he said was that AI is not working in the enterprises, then everybody's afraid to actually say so because it's not supposed to be a cool thing to...

在我们谈到 AI 不起作用之前，因为我认为这可能是最重要的问题之一，但这是一个独立的部分。你认为他们担心前沿模型提供商抢走他们的饭碗是对的吗？

Before we get to AI not working, because I think it's probably one of the most important questions but it's a whole separate segment. Do you think they're right to be afraid of the frontier model providers eating their lunch or not?

嗯，是的，我的意思是取决于企业。是的，我认为如果我们谈论的是从根本上改变人们的工作方式，并且如果我们说今天我们所做的大部分工作将由一个完全由这些前沿模型公司驱动的智能体完成，那么从某种意义上说，你已经将大量运营转移给了这些技术提供商。这不仅仅是技术依赖。这实际上是对为你运行这些智能体的公司的真正运营依赖。这很有趣：如果你思考工作随时间如何发生，当你第一次执行任务时，你可能会记录一个流程，比如完成某项工作所需的 10 个步骤。然后随着时间的推移，人们开始优化和调整该流程。其中很多从未被记录下来，你只是基于反复做这项工作，积累了所有这些学习，并实时应用于未来的工作。所有这些机构学习实际上会积累在执行该工作的智能体中。因此，如果你无法自己控制该智能体的运行，如果你不拥有它多年来获得的学习，那么你基本上完全依赖这些 AI 公司来完成你的工作。所以我认为，企业今天面临一个根本性问题：他们如何实际使用这些 AI 技术，同时仍然保留控制权以及 AI 带来的所有复合学习？这些学习属于企业。

Well, yeah, I mean depending on the enterprise. Yes, I think if we are talking about fundamentally changing how people work and if we are saying that majority of the work that we do today is going to be done by an agent which is fully powered by one of these frontier model companies, then in some sense you've now transferred a lot of your operations to these technology providers. This is more than technology dependence. This is actually real operational dependence on the companies that are running those agents for you. It's actually interesting: if you think about how work happens over time, when you initially do a task for the first time, you might document a process, like the 10 steps you need to take to complete some piece of work. Then over time, people start to optimize and tweak that process. A lot of it never gets documented and you just, based on doing this work over and over again, you've now built all these learnings that you apply in real time to do this work in the future. All of that institutional learning is actually going to accumulate in that agent that is doing that work. So if you don't have any control on running that agent yourself, if you don't own the learning that it gains over the years, then you're basically fully dependent on these AI companies to get your work done. So it's absolutely, I think there's a fundamental question in front of enterprises today: how do they actually use these AI technologies but still retain control and all the compounding learnings that happen with AI? They belong to the enterprises.

你是否看到企业客户从前沿模型提供商转向开源？

Are you seeing enterprise customers move away from frontier model providers towards open source?

这正在发生。所以我认为我们正处于开源的一个真正转折点。部分原因是，你一直在等待开源模型真正变得更好。这种愿望已经存在了很多很多年。

That's something that's happening now. So I think we are at a real inflection point with open source. Part of it, you know, you're waiting on the open source models to actually get better. The desire has been there for many, many years.

没有一家我们聊过的企业会说，嘿，我用 OpenAI 或 Anthropic 就能搞定工作，我满足了。每个人都希望掌控自己的命运，能够使用多种模型。而且现在 AI 变得非常昂贵，你经常听到公司为 AI 制定年度预算，但一两个月就超支了。所以这加速了对开源的需求，再加上现在开源中已经有了非常好的模型。

Nobody there's no enterprise you know that we talk to which is okay with saying that hey look you know I can get my work done with open AI or with Anthropic and I'm good. Everybody wants to make sure that they are in control of their destiny that they get to use many of these models. Um and now given that AI has become so expensive right I mean like if you look at people hear stories all the time about companies you know coming up with a annual budget for AI and they run run past that like you know within a month or two. So in this >> CFOs >> yeah so so that sort of that has really uh accelerated that sort of um that desire you know for open source cuz you know and and that coupled with the fact that we now have really good models in open source.

他们关心什么？是成本吗？还是所有权，比如数据留在本地、模型他们能看得见？到底是什么？

What do they care about? Do they care about cost? Do they care about ownership in terms of their data staying on prem in models that they actually can have visibility on? What is it?

我认为当前开源驱动力来自成本。当然，有些企业确实需要将所有推理工作负载保留在私有数据中心。AI 刚出现时，公司更担心数据失控、模型公司用他们的数据训练。但这种恐惧已经不存在了，人们相信只要签了合适的合同，模型公司会负责任，不会用企业数据训练模型。所以现在的驱动力是成本。

I think right now the open-source drive is coming from the cost point of view. I mean there are certain businesses of course you know that have the requirements to to actually keep you know all the inferencing workload within their own private data centers. when AI just came, there was a companies were a lot more afraid of um getting their data outside of you know their own control and and model companies training with with you know with their data. But that sort of is a fear that it's no longer there like you know like people believe that the model companies are going to be responsible and not train uh their models on on enterprise data so long as you know I've signed up you know for the right kind of contract. So right now the the drive is coming from cost

就你在行业中的位置而言，我聊过的每位投资者都说我必须问这个问题，也就是那个显而易见的问题：你担心 Anthropic 会像他们对 Figma 那样，或者像他们在法律、健康领域所做的那样，进入你的领域并蚕食你的业务吗？首先，我认为要谨慎看待他们实际对 Figma 或法律、金融领域做了什么。他们确实在推出这类垂直包，但在我看来相当浅薄。我实际上没听说有人把工作负载完全从 Figma 或其他工具迁移到 Anthropic。这其实总是净新增的，或者说是在扩大市场。比如在设计领域，设计师仍然用 Figma，但非设计师开始用 Cloud Design。我们看到的是 AI 让事情变得更简单，所以如果人们不是某个工具的主要用户，他们可以用 Claude 来做一些工作。

in terms of where you sit in the landscape. Every one of your investors that I spoke to said that I had to ask this question which is the obvious question which is >> do you worry that Anthropic will do what they did to Figma say or what they've done with legal or what they're doing with health and move into your space and cannibalize your business. First of all, I think we should be careful like in terms of what they've actually done for Figma or legal space or finance space. Um, they are launching these sort of vertical packs, but I think they're quite um >> quite shallow in my opinion >> and and I don't actually know of people who are uh sort of moving their workload entirely from from Figma or for that matter from any other tool to Anthropic. it's actually sort of net new always like you know uh or like I think it's expanding the market like for example now in design the designers still use Figma uh but the non-designers actually you know are using cloud design right I mean so that's sort of what we're seeing is that AI is making things simpler um so if people are not experts you know on the primary users of that particular tool they can actually um they can actually start to do some of that work with claude

所以你不担心他们强调进入企业领域并成为那个上下文吗？

so you don't worry that they all put emphasis on moving into enterprise and being that context.

嗯，他们已经在做了，不管做没做，我们每天都面临与企业客户的竞争。人们经常问我们，Claude 也可以通过 MCP 连接企业系统，那有什么不同？Cleo 能做什么而 Claude 不能？所以我们得解释上下文到底是什么，以及构建它为什么复杂。我们实际上在竞争，而且我认为他们可能比其他公司更早开始与我们竞争。因为如果你把 Claude Co-Work 看作一个应用或 Claude Desktop，它的主要用例一直是问答，而今天世界上最大的 AI 应用或用例就是信息检索和问答。

Well, they're already doing it or or whether they're doing it or not, we actually face that competition every day u with enterprise customers. People will often ask us well I mean cla can also connect with enterprise systems through MCP. Um so what's different like you know what can clean do which you know cloud cannot. So we have to go and explain um like what like you know context really is and why it is actually complicated to actually build it. So, so we so we are we are competing in fact actually I would say that they probably started to compete with us before others like you know they like cuz you know if you think about Claude Cowork as an application or Claude Desktop um the the primary use case for that has always been question answering right that's the like the largest application or use case for AI in the world today is in fact information seeking and question answering

你认为先发优势有多重要？实际上非常有优势，但它只能帮你，不能带你走到底。对我们来说，作为全球第一家企业 AI 公司，第一个将 RAG 引入企业，第一个构建概念语义搜索，这给了我们品牌和在这个市场竞争的权利，尽管现在相比 OpenAI 和 Anthropic 这些巨头我们小得多。所以这是一个巨大的资产，但既不是必要条件，也不是救世主。

how important do you think being first to market is it's actually very advantageous uh but it's actually only it's it's it's a it's a thing that it helps you but it's a thing that's not going to carry you. So for us, you know, we actually get a lot of credit for being the first enterprise AI company in the world, the first ones to actually uh bring rag into the enterprise, the first ones to build conceptual semantic search. And so that actually gives us that brand and the right to compete in this market even though now like you know we're much smaller compared to the giants that uh OpenAI and Anthropic have become. So it's a it's a huge it's a huge sort of asset but neither is it a requirement nor is it a is it a savior.

你如何建议那些夜不能寐、担心前沿模型提供商进入他们领域的创始人？

How do you advise founders who are losing sleep at night worried that the frontier model providers will come into their space.

哦，我会说绝对不要担心。作为创始人，你必须解决问题而不是担心，这是第一位的。当然，你需要预测他们会做什么，看清他们当前的能力。但我认为，对于几乎所有不从事前沿模型训练的 AI 公司来说，应该把模型公司视为巨大资产，而不是竞争。我们相信，Anthropic、OpenAI、Google 所做的一切，以及开源中的所有创新，对我们来说都是好消息。我们不担心，也不认为那是竞争。事实上，没有他们的帮助，我们永远无法交付现在的产品。

Oh right I would say like absolutely don't worry about that. I mean like like I think as as a founder you have to you have to actually solve problems not worry number one right. So you have yes you know like you have to always you know anticipate you know what they're going to do. you have to see their current capabilities. Um but I think like for almost all other AI companies that are not doing frontier model uh training uh they should see the model companies as a as a huge asset um not a competition in my opinion like you know we like you know we actually believe that uh everything that Anthropic is doing everything that openai and Google is doing as well as all the innovation that's happening in open source that's great news for us like we don't worry about that as and we don't think of that as competition in fact like you know they've allowed us to deliver a product that we could never you know without that help.

当谈到 Anthropic、OpenAI 以及模型层的崛起，尤其是开源中国提供商快速推出新模型时，你不认为我们正在看到模型层的终极商品化吗？我们不是在见证模型层的终极商品化吗？有一点很清楚，我们来谈谈企业用例。

Do you not think we're seeing the ultimate commoditization of the model layer when you speak about Anthropic, OpenAI and the rise of the model layer and the speed with which new models are coming out especially bunding from open source Chinese providers >> are we not seeing the ultimate commoditization of the model layer? So one thing is clear um that let's let's talk about enterprise use cases.

90% 或更多的用例现在可以由许多不同的模型完全处理，包括开源模型。所以从这个角度看，确实存在商品化。事实上，在我们 Glean，这是我们为客户提供的核心价值之一，即成本控制。我们会告诉他们，当用户在平台上完成任务时，我们会为你选择合适的模型。如果你愿意使用开源模型，我们会在认为合适且能生成高质量答案时使用它们。

90% uh or greater of use cases can now be fully handled by um many many different models including open source models. So there's definitely um uh there's definitely commoditization from that perspective. In fact like you know we at lean uh that's that's actually one of our core value ads to our customers you know which is cost control. We will actually tell them that hey look you know um as people actually complete their tasks on our platform we actually pick the right model for you and if you're okay with using open source models uh we'll use them you know when we think it's appropriate when it's going to generate a high quality answer.

有多少客户不接受开源模型？这其实非常新。我认为开源真正达到前沿能力，在 3 个月内发生，这实际上就在一个月前，甚至不到一个月。比如，GLM 5.2 是我们团队第一次觉得可以放心地在那个模型上运行大部分工作负载。所以我们还不知道人们会怎么说。从开源和使用模型的角度来看，每个人都会接受。问题在于：他们是否接受中国模型？这才是唯一的问题。不是开源与闭源之争。当你拥有所有权，能够本地部署时，为什么他们不接受中国模型？你没有向中国回传任何东西。为什么不接受呢？

What percent of customers are not okay with open source models? This is actually so new. I would say that open source truly coming to frontier capabilities within 3 months has just happened literally like a month back or not even a month. I would say GLM 5.2 is the very first time where our own team, for example, feels comfortable that now we can run majority of our workloads on that model. So we are yet to find out what people are going to tell us. From a point of view of open source and using the model, everybody's going to be fine. The question is going to be: are they okay with the Chinese model or not? That's the only question here. It's not open source versus closed source. Why would they not be okay with a Chinese model when you look at the ownership that you have, the ability to have it on prem? You're not sharing anything back to China. Why would you not be?

我认为这只是舒适度问题。只是担心万一出问题怎么办？总会有偏执和恐惧。万一有后门呢？某种我们甚至不了解的魔法后门。所以确实存在一些担忧。还有，如果你使用这些模型，一旦这件事被公开，可能会被竞争对手利用等等。所以有多种因素。但归根结底，这取决于谁愿意大胆尝试，因为这是新事物。大型企业必须迈出这一步，先行者会先行动，然后这就会变得更正常。我做这个节目一直想了解的是，90% 的企业工作流是否可以用开源模型完成。

I think it's just comfort. It's just what if something goes wrong? There's always paranoia and fear. What if there's a back door? Some magic back door that we don't even understand. So there are some concerns. There's also if you use these models and if it becomes a known thing, then it could be used against you by your competitors and things like that. So a variety of factors. But ultimately, it boils down to who's willing to be bold because this is a new thing. Large enterprises have to make this move, and the early movers will make the move first, and then it'll become a more normal thing. I'm always doing this show to learn if 90% of enterprise workflows can be done with open models.

我们是否完全错误地定价了前沿模型格局？这是一个非常不同的时代。

Have we completely mispriced the frontier model landscape? It's a very different time.

我确实觉得，模型业务本身，暂且不提开源，即使在实验室内部也有大量竞争，而且越来越多的公司进入这个领域。在如此激烈的竞争中，即使是三方竞赛，我认为也会带来相当大的定价压力。现在，当然，有了开源，价格实际上低了一个数量级。我确实听到传言说 OpenAI 将大幅降低模型价格，以应对这些发展，比如竞争和开源。所以我认为模型业务本身可能并不像大家想象的那么赚钱。但这些公司现在有了更多东西。它们不再是单纯的模型公司了。

I do feel that the model business on its own, regardless of open source for a minute, there is plenty of competition even within the labs, and more and more companies are coming into that space. In that fierce competition, even in a three-way race, I think you can actually get a good amount of pricing pressure. And now, of course, with open source, it actually is an order of magnitude cheaper prices. I actually heard rumors that OpenAI was going to drastically reduce their model prices in response to these developments, like competition and open source. So I think the model business on its own is actually probably not as lucrative as everybody believes. But these companies now have a lot more things. They are no longer model companies only.

完全理解。但如果他们在那些相邻领域只做浅层的事情，他们不太可能产生一万亿美元的收入。就像 Dario 说的那样。

Totally get that. But if they're doing shallow things in those adjacencies, they're not exactly going to generate a trillion dollars of revenue. Like Dario said.

首先，这两个实验室是非常根本不同的业务。OpenAI 当然有出色的消费产品，而 Anthropic 实际上有一个有趣的事情正在发生：人们正在他们的平台上构建。当你现在想到 Anthropic 时，有很多人实际上正在开发自动化和技能，每个人都在创建这些 MCP 服务器连接到他们的内部系统，并连接到云端。所以实际上正在形成一个生态系统。所以你应该很大程度上将它们视为应用层公司，而不仅仅是模型公司。

First of all, these two labs are very fundamentally different businesses. OpenAI of course has amazing consumer product, and Anthropic actually has an interesting thing happening: people are building on top of their platform. When you think about Anthropic right now, there are a lot of folks who are actually developing automations and skills, and everybody's sort of creating these MCP servers to their internal systems, getting connected and connecting it all to cloud. So there's an ecosystem actually being developed. So you should very much consider them an application level company, not just a model company.

如果你猜一下，3 年后，你认为你的工作流中有多少百分比是通过开源实现的？

If you were to make a guess in 3 years time, what percent of your workflows do you think are through open source?

我们一直在告诉客户，我相信 3 年后，大多数企业工作负载实际上会运行在开源模型上，这是肯定的。

We've been telling customers that I believe the majority of enterprise workloads will actually be on open source models in 3 years for sure.

你面临的另一个竞争因素是获取模型提供商。实际上，微软，你知道，微软在创造一个 70% 好的产品的基础上，将其捆绑成企业套餐，然后贴上漂亮的标签出售，从而建立了非凡的业务。你如何看待微软 Copilot 的捆绑压力作为竞争威胁？

Another competitive element that you face for getting kind of the model providers. It's like actually Microsoft, you know, Microsoft have made a phenomenal business on the back of creating a 70% as good product but bundling it into a bundle for enterprises and then selling it with a nice sticker on it. How do you think about the bundling pressure from a Microsoft Copilot as a competitive threat?

对我们来说，他们是我们最重要的竞争对手之一。捆绑策略确实有效，你必须与之抗争。幸运的是，总会有同类最佳软件的空间。我们的客户正是这样看待我们的。如果你试图带来一个出色的搜索产品，如果你试图构建一个横向的综合 AI 平台，他们知道我们做得更好。所以公司愿意在此基础上投资，作为微软捆绑产品套件的一部分。但另一件可能使捆绑不再那么有效的事情是，AI 正在转向基于消费的模型。一旦转向消费，就没有固有的捆绑优势。作为一家企业，我可以获得六个工具，让用户选择在哪里工作。无论他们在哪里工作，我都需要为那个特定工作单元付费。所以消费最终可以打破这种捆绑策略。

For us, they are one of our most significant competitors. And the bundling strategy actually works, and you have to fight against that. Luckily, there has always been room for best of breed software. Our customers think of us exactly like that. If you are trying to bring a great search product, if you're trying to build a horizontal comprehensive AI platform, they know that we do it better. So companies are willing to invest on top of that, as part of the bundled product suite from Microsoft. But the other thing that is maybe making bundling not as effective of a strategy anymore is the fact that AI is moving towards consumption based models. Once you move towards consumption, there's no inherent bundling advantage. As a business, I can get six tools and let the users choose where they want to do their work. Wherever they do their work, I have to pay for that particular unit of work. So consumption can ultimately break that bundling strategy.

恕我直言，我不知道如果你与企业合作，这是否会成立，因为他们会让你符合企业捆绑的要求。所以你将经历全球最大企业内部的审批、流程、签核流程，比如大众、福特、通用或泰森鸡肉。我总是用它们作为例子，你知道，我认识一些随机公司。但它们会批准微软作为单一供应商。如果它们突然要批准 15 个供应商，先不谈定价和交易，这会产生它们以前没有的供应商管理问题。

And respectfully, I don't know if it does if you're working with enterprise because they will make you compliant as an enterprise bundle. And so you'll go through approvals, processes, sign off processes internally for the largest enterprises in the world, your VWs or your Fords or your Gs or Tyson chickens. I always use them as like, you know, I know random companies. But they'll approve Microsoft as one vendor. If they're suddenly having to approve 15 vendors, forgetting the pricing and the transactions, it creates a vendor management problem that they didn't have before.

这是真的。但我想说，如果你去和那些经历过微软冲击的公司交谈，大多数都会说定价是主要杀手，因为我认为很难与免费竞争。

That is true. But I would say that if you go and talk to companies that have been on the other side of the Microsoft onslaught, most of them will talk about pricing as the main killer, because I think it's hard to compete with free.

谁更是一个激烈的竞争对手，微软还是前沿模型？

Who's a fiercer competitor, Microsoft or the frontier models?

好问题。我认为现在说这个还为时过早。

Good question. I think it's early to tell that.

但微软确实很强大。我们在拓展客户时，经常听到这样的回答：‘我们是微软的客户，我们已经有了 Copilot，所以没必要考虑你们。’这种回答比‘我已经用了某个实验室的产品，所以不会再做别的’更常见。你一开始提到了 Alex Karp 的采访，我打断你说先别谈价值。我认为 2026 年下半年和 2027 年，每个人都会问：‘等等，这笔投入到底有没有产出或 ROI？’我们该如何看待企业获得的投资回报？Alex Karp 说每个人都在问‘我的回报在哪里’，他说得对吗？

But Microsoft is formidable. When we prospect, we often hear, 'We're a Microsoft customer, we're getting Copilot, so it doesn't make sense to consider you.' We hear that more often than someone saying, 'I've embraced a lab product, so there's nothing else I'll do.' You mentioned Alex Karp's interview at the beginning, and I interrupted you saying, 'Let's not dive into value yet.' I think 2026 H2 and 2027 is when everyone goes, 'Wait a minute. Is this spend generating output or ROI?' How do we think about the return on investment that enterprises are getting? Is Alex Karp right that everyone is asking, 'Where's my return?'

我认为目前确实有一些领域实现了价值。比如客户支持这个垂直领域，很容易衡量生产力。你可以说，公司里一个支持人员原来每天处理 10 个工单，现在因为 AI 能处理 12 个。所以你能看到非常具体的生产力提升。这是一个 AI 确实擅长的用例，因为支持团队的大量时间花在阅读知识库和向客户总结信息上。所以确实有一些领域实现了明确的价值，企业也感觉良好。其他一些领域则更复杂。比如，我认为目前大部分 AI 投入都花在编程上。编程实践已经改变了；大多数开发者现在都用 AI 写代码，不再手写。所以从某些方面来说，你可以说 AI 产生了很大影响。但他们是否真的在更快地交付产品？这正是大多数公司说‘没有’的地方——产品的实际交付速度并没有提高，尽管编程速度显著提升，因为编程只是整个产品交付的一小部分。

I would say there are pockets of value realization today. For example, take customer support as a vertical. It's easy to measure productivity. You can say that in your company, a support agent resolves 10 cases a day, and now they can do 12 because of AI. So you see a very concrete measure of productivity increase. That's a use case where AI is actually pretty good, because a lot of the time spent by support teams is about reading knowledge and summarizing it to customers. So there are definitely areas where there is clear value realization, and enterprises are feeling good. Some other areas are more complex. For example, I think the majority of AI spend right now is on coding. Coding as a practice has changed; most developers now use AI to write code. They're not writing it by hand anymore. So in some ways, you can say AI made a big impact. But are they shipping products faster? That's where most companies say no, the actual shipping speed of products has not increased, even though coding speed increased significantly, because coding is only a small part of overall product shipping.

你们自己的交付速度提高了吗？

Has your shipping speed increased?

我认为实际上很难衡量。这就是挑战所在，因为工程生产力是最难衡量的东西之一；它是所有工作中最模糊的。如果你看代码行数这样的指标，我们当然写了更多代码。但如果你看我们是否以更快的速度交付功能，是的，我们确实更快了。但这同时也是因为我们有了更大的团队，而且团队经验更丰富。所以有时很难区分。但尽管如此，作为一家公司，我们决定继续投入。

I would say it's hard to actually measure. That's the challenge, because engineering productivity is one of the most difficult things to measure; it's the fuzziest of jobs. If you look at metrics like lines of code written, of course we're writing way more lines of code now. But if you look at whether we are shipping features at a greater pace, yes we are. But it's also a result of having a larger team and a team that is more tenured than before. So sometimes it's hard to tease apart. But with that, as a company, we are saying we are just going to keep investing.

你认为 Glean 现在有多少比例的代码是由 AI 编写的？

What percent of Glean code do you think is written by AI now?

可能几乎接近 100%。现在基本上没人再手写初始代码了。也许偶尔有个别‘工匠’还在手写。所以是的，几乎所有代码都是用 AI 写的。但我们强制要求人工审查，所以你不能生成大量 AI 代码然后直接提交到仓库。我们可能比大多数公司更保守。公司内部确实有过讨论：既然 AI 能写这么多代码，真正的瓶颈已经从写代码的人转移到了审查代码的人。所以有人提议取消代码审查，直接让代码提交到仓库。很多公司已经在这么做了。

It's probably about almost 100%. Nobody is actually writing the initial code by hand anymore. Maybe sometimes you have an artisan in the corner. So yeah, almost all the code is being written with AI. But we actually enforce human reviews, so you cannot generate tons of AI code and then just check it into the repos. We're probably more conservative than most other companies. There was in fact a discussion inside the company that now AI can write so much code, the real bottleneck has shifted from the person who writes the code to the person who has to review it. So there was a proposal to actually eliminate code reviews and just let the code be directly submitted into the repos. Many companies are doing that.

嗯，如果你必须有一个严格的代码审查流程，那几乎就抵消了加快代码开发速度的意义。

Well, if you have to have a stringent code review process, it almost removes the point of having a faster code development process.

是的，确实如此。所以这就是原因。但我认为目前的情况是，我们仍处于学习如何正确有效地使用 AI 的阶段，思考其长期影响。因为当你用 AI 写代码时，你可以写一百万行代码，但随着时间的推移，维护、理解和管理这些代码会变得极其困难。

Yeah, it's true. So that's why. But I think what it's doing right now is that we're still in the learning phase of using AI properly and effectively, thinking about long-term ramifications. Because when you write code with AI, you can write a million lines of code, but it becomes incredibly hard to actually maintain it, understand it, and manage it over time.

但 AI 不正是做这些的吗？有 AI 做重构，AI 做安全，AI 做……

Is that not what AI does though? You have AI that does refactoring, AI that does security, AI that does...

是的。唯一的问题是它现在还不够完美。所以我认为我们宁愿承担审查代码的成本。我们仍然比以前快，因为编写部分现在快得多，而且写代码的人自己会做第一轮审查。

Yeah. The only thing is that it's not that perfect right now. So I think we would rather pay the cost of reviewing the code. We're still faster than before because the writing part is much faster now, and the person who writes is the one who does the first review.

那么当你说 AI 的 ROI 实际上是一个吞吐量问题时，是什么意思？

So when you say AI ROI is really a throughput problem, what does that mean?

我们首先要做的是确保能够为这些 AI 智能体提供正确的上下文。如果你看看今天大多数企业，他们部署 AI 的方式就是把它扔进系统，用 MCP 服务器以粗糙的方式将 AI 与所有企业系统连接起来，然后让你试图用 AI 完成的任何工作都让模型通过暴力方式去尝试找出并组装完成任务所需的原始材料。在这种模式下，AI 非常慢。仅仅组装它所需的基本信息就要花很多时间。而且成本也非常高，因为大部分 token 都消耗在试图为给定任务组装正确的上下文上，而且你还在用 AI 做那些它并不擅长或不需要的事情。所以相反，我们讨论的是，要让 AI 真正发挥作用并交付成果，你必须围绕它进行投入。你必须确保提供正确的上下文，这样它才能以更低的成本更快地工作。

The first thing we have to do is make sure that we are able to bring the right context to these AI agents. If you think about most enterprises today, the way they're rolling out AI is that they just throw it into the system and connect AI with all of your enterprise systems in a rudimentary manner using MCP servers, and then let any piece of work you are trying to do with AI let the models brute force their way into trying to figure out and assemble the right raw materials they need to complete the task. In this mode, AI is super slow. It takes a lot of time to just assemble the basic information it needs to do the work. It also becomes very costly because most of the tokens are being burnt just trying to assemble the right context for that given task, and you're trying to use AI for things where it's not even good at or needed. So instead, what we talk about is to make AI really perform and deliver, you have to invest around it. You have to make sure you provide it the right context so that it can actually work faster at a lower cost.

围绕它投资意味着什么？作为 CEO，我们鼓励所有团队成员尝试用 AI 取代自己，即使这意味着浪费 token，这是错的吗？

What does it mean invest around it? And are we wrong as CEOs to be urging all of our team members to be trying to replace themselves with AI even if it means that we're wasting tokens?

我认为说“用 AI 取代自己”是一个错误的目标。首先，我觉得你太高估 AI 了，而且它现在还没准备好。你举个例子，说一个能用 AI 取代的工作。你觉得它能取代你的行政助理吗？

I think it's a wrong goal in my opinion to say that hey like replace yourself with AI. First of all, I think you're giving too much credit to AI and you know when you say that it's just not ready right now. You give me a name of one job that you can replace with AI for example. Do you think it can replace your EA?

我的？不能。但我比较挑剔。

Mine? No. But I'm a diva.

对大多数人来说，我认为它能完成大部分工作。是的，我这么认为。关键是：它确实能处理很多任务，但无法取代最后那点无形的东西。不过，这可能是一个临界点——对很多人来说，如果它能做到 90%，那就够了。你知道吗？你会亲自给妻子买生日礼物，因为一年只有一次，不常发生，而且 Claude 还不够个性化，不知道你妻子喜欢的香水。所以这个角色会被蚕食。

For most people, I think it can do the majority. Yeah, I do. That and that's the thing: it can actually take care of a lot of things for any given role, but it cannot replace the final intangible. No, but that can be a tipping point where actually for a lot of people if it does 90%. Fine. You know what? You'll do that birthday present for your wife because it's once a year. It's not very often and Claude isn't quite personal enough to know your wife's preferences of perfume. And so it's very but that role will get cannibalized.

我不确定。我告诉你为什么。我认为无论做什么，你都希望做到最好，而你不会接受一个 90% 的解决方案。

I'm not sure. And I'll tell you why. I think you want to be performing the best in whatever you do, and I don't think you're going to take a 90% solution.

我不是因为成本限制才这么刻薄。

I'm not cost constrained being a dick.

嗯，我的意思是，这不是你不在乎成本的问题。而是你必须在工作中与他人竞争。记住，他们也有你拥有的所有 AI 工具，但如果他们还有一个人力在上面，你怎么跟他们竞争？

Well, I mean like look, it's not about you not being cost constrained. It's about you have to be competitive in your work with others. Like remember they also have all the AI tools that you have, but if they also have a human on top, how are you going to compete with them?

所以你不觉得……团队……你们公司现在有多少人？我们公司现在超过一千人。

So do you not think team how many people would you have now in our company? We're over a thousand people now.

超过一千人。你认为五年后会有多少人？

Over a thousand people. How many do you think you'll have in 5 years time?

嗯，希望是 5000 人。

Well hopefully you know 5,000.

哇。所以你不……

Wow. So you don't...

我们会增长。

We're going to grow.

但这很不典型。我和全球最大的 CEO 们坐在一起，他们每个人都在缩减团队，每个人都在说……

But that is very atypical. I sit with the biggest CEOs in the world and every single one of them is shrinking teams and every single one of them is saying...

我完全不相信。

I absolutely don't believe in it.

为什么接近？为什么？

Why close? Why?

嗯，我的意思是，首先逻辑上想想。拿两家公司：可口可乐和百事可乐，或者两家相互竞争的公司。一家决定缩减规模，另一家还有更多人。两家都完全可以使用相同的 AI 工具和技术。那么问题是：如果你试图做同样的工作量，并且相信可以用更少的人完成，因此你缩减规模，你的竞争对手也可以这样做，但他们选择不做同样的工作量。他们选择提升，打造 10 倍更好的产品或生产 10 倍更多的商品，因为他们有更多人。他们会变得更大。他们会打败你。

Well, I mean I think like first think logically. Take two companies: take Coca-Cola and Pepsi, or two companies that compete with each other. One company decides to shrink and the other one still has a lot more people. Both of them have full access to the same AI tools and technology. And so now the question is: if you were trying to do the same amount of work and you believe you can do it with fewer people and therefore you shrink, your competition can also do the same but they chose actually not to do the same amount of work. They chose to actually elevate and build a 10x better product or produce 10 times more goods because they have more people. They're going to be larger. They're going to beat you.

但我不认为更多人就能做出更好的产品。

But I don't think more people makes for better products.

那是另一回事。

That's a different thing.

如果我能削减人员，然后负担得起最好的前沿模型、最好的技术给我的 100 倍效率的工程师，因为我减少了人员。

If I can cut headcount and then afford the best frontier models, the best technology for my 100x engineers because I've reduced headcount.

但我认为那个论点……

But I think that argument...

而且我认为更多人会拖慢一切。

And I think more people slow down everything.

但这不是 AI 的论点。那个论点一直成立。

But that's not an AI argument. That argument has always been true.

当然。但结合 AI 元素，你能够交付更多。如果你能交付更多，我向你保证，当你有更多人时，他们只会设置障碍，阻碍产品发布。

Sure. But combined with the AI element of you're able to ship more. If you're able to ship more, I promise you when you have more people, they'll just put up the barriers to get in the way of that product going out.

嗯，看，即使在现在的 AI 讨论中，但在此之前，疫情后，许多公司觉得自己臃肿，裁掉了 15% 到 20% 的员工，每个 CEO 都出来说他们因此实际上快了 20%。所以很多公司都谈过这个。所以那个论点一直存在：在某个时候，团队变大，开始互相拖慢。人类就是这样。我也相信这一点。但最终，人也是你的资产，你必须正确地将他们部署到合适的项目上。我不认为世界上最伟大的公司会是只有 100 人的公司。看看那些模型公司，它们也一样。为什么它们招聘这么积极？

Well, look, even in the AI discussions right now, but before that, just post-COVID, many companies felt they were bloated, they cut down 15% to 20% of their staff, and every CEO came out and said they're actually moving 20% faster as a result. So a lot of companies talked about that. So that argument is always there: at some point, teams get large, they start to slow each other down. Humans do that. I also believe in that. But ultimately, people are also your asset, and you have to be able to deploy them correctly in the right set of projects. I don't think the world's greatest companies are going to be companies with 100 people. And look at the model companies, like same for them. Why are they hiring so aggressively?

你不认为最优秀的人会想用最好的技术，我们会看到全球最大公司的技术支出从现在的 8% 到 12% 增加到 16% 到 20%，然后实际上你会看到人员减少但技术支出增加，最优秀的人会想去拥有最好工具和设备的地方吗？

Do you not think that the best people will want to work with the best technology and we'll see an increase in technology spend by the biggest companies in the world from 8% to 12% where it is today to maybe 16% to 20%, and then actually you'll see a reduction in headcount but an increase in technology spend and the best people will want to go where they have the best tools and equipment.

我也不确定，因为我认为技术实际上不应该增加成本。首先，你承认目前这项技术就其提供的价值而言定价荒谬吗？我认为这完全取决于它做什么，所以不，我完全不认为 Cursor 或任何开发工具定价过高。我认为它们仍然被严重低估。当你看到 Marc Benioff 在 Anthropic 上花费 3 亿美元，那只是开发人员工资的 3.7%。我认为这相对较小。我会说它贵得离谱。我给你举个例子。我们有一个很酷的工程分类智能体。我们有一个 15 人的值班团队，他们的工作是分类每一个生产问题，比如任何系统警报、出问题的事情。我们构建了这个智能体，现在自动处理 95% 的问题。但即使这样，它的成本也值得怀疑：它真的比人类更高效吗？我们每个月在那个智能体上花费 100 万美元。这实际上比……的成本还要高。

I'm not sure about that either, because I think technology is actually not supposed to increase in cost. First of all, I think do you admit that currently this technology is priced absurdly for what it delivers? I think it totally depends on what it's doing for so no I don't at all for Cursor or for any of the dev tools. I think it's still dramatically underpriced. When you look at Mark Benioff spending 300 million on Anthropic, it's 3.7% of developer salaries. I think that's relatively small. I would say it's absurdly expensive. I'll give an example. We had this really cool triage agent for engineering. We have a 15-person on-call team whose work was to triage every single production issue that happens, like any system alerts, things that are going bad. And we built this agent that now takes care of 95% of those issues automatically. But even there, it's actually doing at a cost which is questionable: is it actually more efficient than humans? We were spending a million dollars a month on that particular agent. And that was actually more than the cost of...

一个月 100 万。

A million a month.

是的。

Yeah.

你在买 C 罗吗？你在干什么？

Are you buying Cristiano Ronaldo? What are you doing?

不。嗯，我是说，那样的成本，确实很贵。

No. Like well I mean yeah cost like that I mean it is quite expensive.

但抱歉，我能回到你刚才说的吗？因为我在节目里讨论过很多次，所以你让我更聪明了。你认为把开发人员工资的 3.8% 花在这些工具上很多。如果你认为那很多，那么这些模型提供商就彻底完蛋了。

But sorry can I go back you said you said because I discussed this a lot on the show so you're making me much smarter. You think that spending 3.8% of developer salaries on these tools is a lot. If you think that's a lot then these model providers are absolutely screwed.

嗯，我觉得我要表达的观点是，3.8% 这个数字实际上看起来并不高，当你这样看的时候。

Well, I mean, I think the point I'm making is that the 3.8% number actually doesn't seem high at all when you look at it that way.

但我也知道，在开源领域，你已经可以用十分之一的成本完成同样的工作量，对吧？这是第一点。第二点，从历史上看，据我所知，我们从未把技术成本和劳动力成本放在同一个句子里。这是第一次听到有人说，嘿，我宁愿要更少的人类和更多的 token。这是第一次。我只是觉得这不是技术运作的方式。模型应该越来越便宜，技术会越来越可负担。

But I also know that already you see with open source that you can do the same amount of work for a tenth of the cost, right? That's number one. Number two, historically, for as far as I can remember, we've never put technology cost and labor cost in the same sentence before. This is the first time we're hearing that hey, I would rather have fewer humans and more tokens. The first time. And I just feel like this is not how technology works. The models are supposed to get cheaper and cheaper. The tech is going to be more and more affordable.

但这是——我很抱歉。这对我来说太有趣了，因为你是 Glean 和 Rubric 的联合创始人，而我只是个播客主持人？但这正是技术的用途。这就是智能体主动出击、有主见、做决策。它们绝对应该被纳入或与劳动力放在同一个句子里，因为它们正在取代我们过去花钱购买的劳动力。

But it's—I'm so sorry. This is so funny for me because you're the co-founder of Glean and Rubric, so who am I but a podcaster? But this is exactly what technology is for. This is agents being proactive, having an opinion, making a decision. They should absolutely be included or put in the same sentence as labor because they are replacing the labor that we used to spend money on.

我认为好的技术会找到让技术变得非常非常便宜的方法，这里也会如此。这是我的信念。你会看到推理成本下降几个数量级。我认为在过去 6 到 9 个月里，我们实际上看到了一些奇怪的事情：每个模型都提高了每 token 的价格。如果你回到 15 个月前，每个人都认为每 token 价格会像以前一样持续下降。所以我们不知道发生了什么。这也算是独特的。

I think good technologies figure out how to make technology really, really cheap, and it's going to happen here too. That's my belief. You're going to see inference costs come down by orders of magnitude. I think we saw something bizarre actually in the last 6 to 9 months: every model actually increased their per token price. If you go back 15 months, everybody thought that the per token price was going to keep falling like it was before. So we don't know what happened here. This is also sort of unique.

他们需要在上市前证明自己是好生意。这就是发生的事情。

They needed to prove that they were good businesses before they went public. That's what happened.

我能说一些你不能说的话。

I can say things that you can't.

是的。是的。但我打赌 AI 会比今天便宜得多。如果 AI 变得比今天便宜得多，这些已经亏损的企业——它们目前支撑着我们整个全球经济——将受到严重威胁。

Yeah. Yeah. But my bet is on AI getting much, much cheaper than what it is today. If AI gets much, much cheaper than it is today, these already loss-making businesses which prop up our entire global economy pretty much at this point are very threatened.

是的，我的看法保持不变。

Yeah, I mean, my take remains the same.

这真的很有趣。所以你不认为未来工程团队会变小。

So it's really interesting. So you don't expect an engineering team to get smaller in the future.

我认为人均生产力会飙升。但需求也会增加。为了获得同样的收入，你未来必须生产出 10 倍更好的产品。不幸的是。

I think per person productivity is going to shoot up. But so will the demands. To make the same amount of revenue, you have to produce a 10x better product in the future. Unfortunately.

当你考虑内部的 token 支出时，你和你的团队以及 CFO 是如何坐下来思考的？你们是如何做出关于 token 预算的决策的？

When you think about token spend internally, how did you sit down and think about it as a team with your CFO? How did you go through the decision of how to think about token budgeting?

嗯，我想我们可能做了大多数公司做的事，那就是什么都没做。我们处于让人们自己摸索能用这项技术做什么的阶段。

Well, I think we did probably what most companies did, which is we didn't do anything. We're in this phase of letting people figure out what they can do with this tech.

你看到了什么？人们疯狂了？还是没人采用？发生了什么？

And what did you see? People went crazy? People didn't adopt it? What happened?

在我们公司以及所有客户那里都存在一种幂律分布。你会看到有些人每月花 10,000 或 15,000 美元在 token 上，而另一些人只花 20 美元。但有一件事很有趣：每个人都在某种程度上接受了 AI。每个人都在使用基本的——正如我之前提到的，当今世界 AI 的第一大应用或用例是信息检索和问答。每个人都在做这个。所以整个团队以及我们的客户都在做。每个人都在提问，获取一些基本的摘要和信息整合。但高级用例仅限于大约 5% 的员工。

There's a power law in our company and also at all of our customers. You will see some people who spend $10,000 or $15,000 in tokens every month. And then you have others who are spending $20. One thing is interesting though: everybody has embraced AI to some degree. Everybody's using the basic—as I mentioned before, the number one application or use case for AI today in the world is information seeking and question answering. Everybody's doing that. So you see the entire team, as well as our customers, they're all doing that. Everybody's asking questions, getting some basic summarization and information synthesis. But the advanced use cases are limited to about 5% of the employee base.

作为领导者，你会做些什么来尽可能积极地推广 AI 吗？我们在帕洛阿尔托的 Nikesh 每周都会开领导会议，就像展示会一样，每个人都需要站起来展示他们那周用 AI 做了什么，取代了他们的工作、改进了他们的工作等等。你有什么可以做的吗？

Is there anything you do as a leader to try and infuse AI as aggressively as possible? We had Nikesh in Palo Alto on every week he has a leadership meeting where he's like show and tell and everyone needs to stand up and show something that they've done with AI that week that replaces what they do, improves their job, whatever it is. Is there anything that you can do?

是的，这实际上是个好主意。我考虑过这样做。我们限制了 token 最大使用量的仪表盘，我一直认为仅仅奖励消耗更多 token 的人不是正确的想法。我觉得我们不需要那样做。我们本身就是一家原生 AI 公司，人们已经足够了解，他们会在需要时使用 AI。但高管分享成功故事——我们没有要求每个高管每周都做。但我们在全员大会上有一个展示环节。我们总是会请人们分享这些成果。每次全员大会都有一个专门的部分，介绍团队正在使用的新 AI 智能体，以改变工作方式。

Yeah, that's actually a really good idea. I've thought about doing that. We limited the token maxing dashboards and I always thought that was not the right idea to just reward people who are consuming more tokens. I felt like we didn't need to do that. We are a native AI company ourselves and people are already educated enough and they will use AI when they need to. But executives sharing success stories—we haven't demanded it from every single exec every single week. But we have a showcase in our town hall, for example. We'll always ask people to share those wins. Every town hall has a section dedicated to these are the new AI agents that teams are using to work differently.

我能问问关于高管和你手下的人吗？我认为招聘从未如此困难。

Can I ask in terms of the execs and the people that you have? I think recruiting has never been harder.

如今招聘有多难？正如我们所说，一些最大的模型提供商支付着前所未有的巨额薪水。

How hard is recruiting today with some of the largest model providers as we said paying just enormous salaries that we haven't seen before?

是的。我实际上想说，也许在过去 2 到 3 个月——我们先把这个放一边。我会说，与 SaaS 高峰期相比，招聘实际上变得更容易了。为什么？因为公司更加——他们没有增加员工人数。如果你看看最大的科技人才雇主，很多实际上并没有增长。很多公司一直在持续裁员。以 Meta 为例，对吧？每年都有大规模裁员，我不知道总人数——我猜可能比 2021 或 2022 年的峰值要低。所以实际上市场上的人才比以前更多。但如果你现在开始谈论 AI 人才、机器学习人才，顶尖人才比以往任何时候都更受追捧。而且薪资水平也完全改变了。不仅来自模型公司，甚至来自初创公司，因为初创公司也在给太多钱来竞争人才。所以即使是现在的初创公司也付很多钱。我们必须这样做。

Yeah. I would actually say maybe in the last 2 or 3 months—let's put that aside for a minute. I would say that recruiting was actually getting easier compared to the SaaS peak. Why? Because companies have been more—they haven't been growing their headcount. If you look at the largest employers of tech talent, many of them actually haven't been growing. Many of them have been laying off continuously. I think about Meta for example, right? Every year there's significant layoffs, and I don't know if the overall headcount—my guess is that it's probably down from the peak of 2021 or 2022. So actually there was more talent available in the market than before. But if you now start to talk about AI talent, ML talent, top people are sought after way more than ever before. And also the pay scales have completely changed. Not just from the model companies but even from startups, because startups are also giving them too much money to compete for talent. So even startups these days pay a lot. We have to.

是的，我们必须这样做，因为他们的替代选择也很大。

Yeah, we have to because their alternatives are so large, too.

而且实际上，如果一名优秀开发者的成本是 30 万、40 万、50 万美元。

And actually, if it costs three, four, 500 grand for a great dev.

是的。

Yeah.

嗯，200 万美元的种子轮根本不够用。但创始人组建团队时，即使自己不拿薪水，如果要雇四个人，我也需要 600 万美元。没错。你认为创始人应该筹集大额种子轮吗？

Well, the $2 million seed round just doesn't go anywhere. But the founder building the team, even if they don't take a salary, if I'm going to hire four people, I need 6 million bucks. That's right. Do you think founders should raise large seed rounds?

我认为更好。我一直倾向于从一开始就尽可能多地融资。

I think it's better. I always prefer to raise as much as you can from the get-go.

哪一轮融资感觉估值最高？

What round felt the most highly priced?

首先，除了第一轮，我们其实从未主动出去融资。总是有人找上门来，通过长期建立的关系成为实际投资人。我觉得我们的 C 轮可能感觉最贵，因为我们几乎没什么业务——肯定不到 200 万或 300 万美元，也许是 500 万，我记不清了——但估值却超过 10 亿。这太极端了。但我想我们只能接受现实。

So first, we never actually went out to raise except for our first round. We always had somebody come in, and it was a relationship built over time that became the de facto investor. I would say our Series C probably felt the most expensive because we barely had any business — definitely sub $2 or $3 million, maybe $5 million, I don't remember exactly — but the valuation was north of a billion. That was extreme. But I guess we take what we get.

这太不可思议了。你在做的时候会担心 Scaling（规模扩张）到那个估值吗，还是埋头觉得这很棒——高估值低稀释？

I mean that's incredible. Would you worry about scaling into that when you're doing it, or do you just head down and think this is great — low dilution for a high price?

我们更看重的是向潜在员工传递一个信号。我们希望市场明白我们在打造特别的东西，这给了我们认可。

The way we thought about it more was that there was a statement to be made to prospective employees more than anything else. We wanted to make the market understand that we're building something special, and that gives us validation.

员工在乎你的投资人是谁吗？

Do employees give a who your investors are?

当然。很多创始人认为最优秀的人不在乎——他们为使命而来。但我向你保证，如果你有 a16z、DST 或红杉，优秀的候选人突然就更愿意和你聊了。

Absolutely. A lot of founders think the best people don't care — they're there for the mission. But I promise you, if you have a16z or DST or Sequoia, great candidates suddenly want to talk to you a lot more.

是的。投资人的声誉直接影响你的声誉。

Yeah. Investor reputation directly impacts your reputation.

回顾今天，过去 12 个月里你最大的观念转变是什么？

When you look today, what have you changed your mind on most in the last 12 months?

我个人一直过于自律，但这可能不再是正确的策略了。我从团队那里得到反馈，说我们过于保守，试图让资本更持久，但这种心态可能让我们错失圈地机会。所以我感到压力，需要改变对支出和投资的看法。同时，我坚信企业建立在纪律之上——你必须为产品收费，它必须为客户创造价值，营销的每一分钱都要有良好回报。你不能指望一直融资来弥补缺失的东西。

I've always personally been too disciplined, but that might not be the right strategy anymore. I get feedback from my team that we are trying to be conservative, making sure our capital goes a long way, but in that mindset we may lose the land grab. So I'm feeling pressure to change how I think about spending and investing. At the same time, I have this fundamental belief that a business is built on discipline — you have to charge for the product, it has to generate value for customers, and every dollar in marketing must have a good return. You cannot assume you just keep raising money to make up for things that weren't there.

但像 Uber 这样的例子证明，糟糕的商业模式可以通过规模变好，你同意吗？

Do you agree with that when you have examples like Uber which prove that a bad business model can turn good with scale?

是的，这就是我说的——我感受到的压力，也许我的思维方式不对。

Yeah, that's what I'm saying — that's the pressure I feel, that perhaps my way of thinking is incorrect.

你认为现在是圈地运动吗？

Do you think it is a land grab?

我们绝对处于圈地运动中，毫无疑问。今天世界上每家公司都想要我们这样的产品，要么我们现在进入，要么未来难度会大 10 倍。

We are absolutely in a land grab, no question. Every single company in the world wants a product like ours today, and either we get in today or it's going to be 10 times harder to get in the future.

我们谈到了岗位替代。你认为哪些现在不存在的工作，在 3 到 5 年内会变得非常普遍？

We spoke about job displacement. What job does not exist today that you think will be incredibly common in 3 to 5 years time?

复合型角色会非常普遍。比如，能构建产品的人——我不知道该怎么称呼他们——但他们像工程师、产品经理、设计师一样工作。同样在市场推广中，有人能销售产品，不仅能够进行商务谈判，还能演示产品、讨论用例。不再区分客户经理、解决方案工程师和销售方案架构师，我们会看到更多角色通用化，远离专业化。事实上，我在自己公司里也在极力推动这一点。

Composite roles will be very common. For example, somebody who can build a product — I don't know what to call them — but they act like engineers, product managers, designers. Similarly in go-to-market, somebody who can sell the product, capable of not only business negotiations but also demoing the product and talking about use cases. Instead of segregation between account executives, solution engineers, and sales solution architects, we'll see more generalization of roles, away from specialization. In fact, I was trying to drive that very hard in our own company.

但抱歉——我这是好意——复合型角色恰恰与维持团队规模的想法相悖，因为如果你把四种不同专长合并到一个角色里……

But I'm sorry — I mean this in a nice way — the composite roles go exactly against the idea of maintaining team size, because if you have composite roles where you bring in four different specialties into one...

那就是更小的团队。

That is smaller teams.

是的。

It is. Yes.

但正如我所说，未来你必须做 10 倍的工作才能从客户那里获得同样的收入。你用更小的团队来完成过去同样的工作量。我们只是被迫做得更多。

But as I said, you have to do 10 times the work to get the same amount of revenue from your customers in the future. You have a much smaller team to deliver the same amount of work that you used to deliver before. We are just forced to do more.

那么我们现在拥有的哪些角色将来会消失？我们会看着说，“天哪，真不敢相信我们以前还做这个。”

And then what role do we have today will we not have? What do we look at and go, 'Oh my gosh, I can't believe we used to do that.'

很多分析师角色——比如不是商业思考者的数据分析师。他们被分配任务查看数据、制作特定仪表盘、配置后端系统。这类工作肯定会消失。商业智能将变得非常不同。业务负责人将直接得到问题的答案。所以业务分析师、数据分析师——这是其中之一。许多 HR 角色，比如招聘中的寻源专员，会被整合到全周期招聘角色中。

A lot of analyst roles — like data analyst roles which are not business thinkers. They were given a task to see data and produce specific dashboards, configure backend systems. That kind of work definitely goes away. Business intelligence will be very different. Business owners will directly get answers to their questions. So business analysts, data analysts — that's one. Many HR roles, like sourcers in recruiting, will get consumed into a full-cycle recruiting role.

我最后必须问一个问题，我们现在在欧洲，这引出了主权问题。美国和欧洲坦率地说在开源方面没有达到标准。

I do have to ask one final one which is we're sitting here in Europe and it brings about a question of sovereignty. The US and Europe bluntly have not come up to muster so to speak on open source.

是的。

Yeah.

你认为我们会迎来一个主权模型的世界吗？考虑到过去一个月左右的情况，我们是否需要对我们的模型拥有主权？

Do you think we will have a world of sovereign models and do you think given what we've seen in the last month or so that we need to have sovereignty over our models?

对主权模型的需求很强烈，实际上，我觉得一年前可能比现在更强。至少我感觉听到的少了。有一段时间，每个国家都认为自己能造一个，那时 AI 还处于早期阶段。但后来很多国家意识到这条路行不通，所以他们愿意让本国企业使用 OpenAI、Anthropic 或其他模型。我不是专家，不知道这个趋势是在上升还是略有下降。

The desire for sovereign models is strong and was actually, I would say, probably stronger a year back compared to now. At least I feel like I'm hearing less of it. There was a period when every nation thought they could build one, when AI was still in its early stages. But then a lot of those nations actually figured out that this is not going to be the way, so they're okay with letting their enterprises within their own countries use OpenAI or Anthropic or all the other models. So I'm not an expert. I don't know whether this trend is on the rise or sort of on the fall a little bit.

我认为这绝对是在上升，考虑到特朗普政府禁止 Anthropic 最新模型，以及很多欧洲人意识到我们不能依赖一个可能切断我们情报获取渠道的美国个体。

I think it's unequivocally on the rise given what we saw with the Trump administration banning Anthropic's latest models and this understanding from a lot of especially Europeans that we cannot rely on a US individual who could ban our access to intelligence.

是的。

Yeah.

但成果在哪里呢？

But where are the results from it?

那才一个月前的事。所以期望三周内就拿出一个模型是很难的。

I mean that was a month ago. So I think to expect a standup model within 3 weeks would be tough.

是的。但即使在那之前，也没有发生。世界上唯一在美国之外产出模型的国家是中国。

Yeah. But even before that, it just hasn't happened. The only country in the world that has produced models outside of the US is China.

是的。

Yeah.

然后可能法国有 Mistral 一点点。

And then maybe a little bit in France with Mistral.

这仅仅是一个激励问题吗？美国缺乏开源社区，为什么我们没有真正有分量的美国开源模型？

Is that simply an incentive problem, the lack of open source community in the US and why we don't have any US open source to a real degree and substantiveness?

不，我认为美国在许多其他领域有很好的开源社区。

No, I think there is a good open source community in the US in many other areas.

但美国有什么开放模型呢？

Well, I mean what open model from the US?

不，你说得对，我们没有开放模型，但这并不是因为开源运动或概念在美国薄弱。实际上它很强。但模型需要大量前期投资，这在很多方面对开源不友好。很多开源软件是秘密项目，开发者没有资金支持却仍能做出东西。但模型不能那样构建，所以自然行不通。你需要不需要超高投资的技术。

No, you're right that we don't have open models, but it's not because open source as a movement or concept is weak in the US. It's actually quite strong. If you think about models, they require a lot of upfront investment, which is not open source friendly in many ways. A lot of open source software has been skunkworks, with developers getting no funding and still getting something built. They couldn't build models that way, so naturally this thing didn't work out. And you need techniques where super high investment is not needed.

那你看到现状会担心吗？我花了很多时间在 OpenRouter 上，看模型使用量和流量。今天 Anthropic 是第一个美国模型，排第七；前六名都是中国的。我们就这样出局了吗？谁在乎中共在资助前六名模型？

Do you worry then when you look at the state? I spend a lot of time on OpenRouter and I see the model usage and traffic. Anthropic today was first US model, it was seventh; the first six were Chinese. Do we just get out of bucket? Who cares that the CCP are funding the top six models?

你可以在那个受控环境中实际运行推理，这让人们感到放心。但我不认为美国会对这个趋势完全放心。现在有一些好的工作正在推动美国的开源和模型开发。有些模型正在出现。

The fact that you can actually run inference on those in that contained environment makes people feel comfortable. But I don't think as a US, the US won't feel absolutely okay with that trend. There's good work happening now to actually promote open source and model development in the US. There are some models coming out.

另一种可能是 Sam 和 OpenAI 给 Trump 5%，然后他对 Anthropic 和 OpenAI 进行监管俘获，并攻击开源。

The alternative is that Sam and OpenAI give 5% to Trump, and then he puts regulatory capture on Anthropic and OpenAI and puts attacks on open source.

我希望不会。我怀疑那会发生。

Well, I hope not. I doubt that's going to happen.

不然 Sam 为什么要给他们 5%？这是交换。我需要你，你需要我。

Why else would Sam give them 5%? It's a quid pro quo. I need you, you need me.

嗯，我只是更相信美国体制，最终会遏制……顺便说一句，我不认为现在需要遏制开源；它在美国太落后了。

Well, I guess I just believe more in the US system and ultimately curbing... I don't think right now, by the way, you need to curb open source; it's too far behind in the US.

你不觉得 Sam 和 Dario 会想，‘哦，哇。我们低估了这一点，这对我们的业务是核心威胁。’

You don't think Sam and Dario are sitting going, 'Oh, wow. We underestimated this and this is a core threat to our business.'

他们可能确实这么想，但我不认为他们能通过监管解决。

They probably are thinking that, but I don't think they can fix that through regulation.

你不觉得 Sam 会直接打电话给 Trump，说‘嘿，中共在资助你最大的、我们最大的竞争对手，我们不能保证没有后门通向习近平。你需要阻止这件事，我给你 5% 作为辛苦费。’

You don't think that Sam will be calling up Trump, who he has a direct line to, saying 'Hey, the CCP are funding your biggest, our biggest competitors, and we cannot promise that there isn't a back door to Xi Jinping. You need to stop this and I'll give you 5% for your troubles.'

但论点不是反过来的吗？现在有很多非常好的开源模型，都是中国造的，美国需要自己造。美国不能被视为一个不进行技术创新的国家，所以美国造模型实际上是至关重要的。

Well, isn't the argument the other way around? Right now there are all these open source models which are very good and they're all built in China, and the US needs to build its own. The US can't be seen as a country that doesn't innovate on technology, so it's actually paramount for the US to build.

我认为 Sam 会说这需要数十亿美元和数年时间。Trump，保卫美国，支持 OpenAI 和 Anthropic，设置障碍阻止中国开源模型获得采用、征税、禁令。

I think Sam will be saying it takes billions of dollars and years of time. Trump, defend America and support OpenAI and Anthropic, and put barriers up to prevent Chinese open models from getting adoption, taxes, bans.

那些也许是的，但美国开源模型将获得很多顺风。这是每个湾区技术人员都在谈论的公认问题。有很多有动力的方想要推动，比如 Nvidia。他们投入大量投资推动美国优秀开源模型的开发，我希望他们成功。

Those maybe yes, but US open source models are going to have a lot of tailwinds. This is a known accepted issue that every technologist in the Bay Area talks about. There are a lot of motivated parties that actually want to promote, including Nvidia for example. They're putting a lot of investment in promoting development of great open source models in the US, and I hope they succeed.

绝对。多模型世界对我们所有人都很重要。听着，我要和你做一个快速问答。我说一个简短陈述，你给我即时想法。可以吗？

Absolutely. A multimodel world is important for all of us. Listen, I'm going to do a quick fire round with you. So I say a short statement, you give me your immediate thoughts. Does that sound okay?

好的。

Okay. Yeah.

你对今天学习计算机科学的人最大的建议是什么？

What's your biggest advice to someone studying computer science today?

学它没问题。不要因为别人说的而过于担心。

It's fine to study it. Don't get too worried because of what other people are telling you.

你认为哪家传统公司采用 AI 最好？

Which legacy company has adopted AI the best do you think?

你愿意称 Google 为传统公司吗？

Well, are you willing to call Google a legacy company?

是的。

Yeah.

那么 Google 可能比任何其他公司都做得更好，不仅在内部拥抱 AI，还推出产品。但我想他们是 AI 公司，所以把他们归入那个类别有点不公平。

So Google probably rates higher than anybody else in terms of not only embracing AI internally but also launching products. But I guess they are AI companies, so it's kind of unfair to put them in that category.

你创办一家新公司，只能带一位大使。你带谁？

You start a new company and you can only take one ambassador. Who do you take with you?

我想我会带我们现有的一个。我们和他们都有很好的关系。

Well, I think I'll take one of our existing ones. We have great relationships with all of them.

你会带哪一个？

Which one would you take?

我不知道。我不回答这个问题。我真的没有答案。

I don't know. I won't answer that question. I just don't have the answer really.

你最想改变当今创业生态系统的哪一点？

What would you most like to change about the startup ecosystem that we see today?

我确实认为如今初创公司可获得的资本太多了，这有时反而会为他们制造失败路径。我觉得他们没有真正理解打造一家伟大公司需要什么。举个例子：一家刚融完种子轮的初创公司决定花 50 万美元雇一个工程师，就像你之前说的。这种事现在正在发生，创始人觉得没问题，投资者也觉得没问题，但这绝不是一条可持续的制胜之路。他们付这么多钱，而谷歌却不这么干，谷歌知道他们不需要这样买人才。所以我认为，资本过剩正导致初创公司建立起一些不可持续的结构。

I actually do think that there is too much capital available today for startups, and it's actually sometimes creating failure paths for people. I think they're not getting what it takes to build a great company. I'll give you an example: a startup that has raised a seed round decides to pay half a million dollars to an engineer, like you were saying before. It's happening today, and the startup founder is okay with it, the investors are okay with it, but it's just not a sustainable path to actually win. And they're paying it while Google is not, and Google knows that they don't need to buy talent like that. So I think this overabundance of capital is getting startups to create structures which are not going to be sustainable for them.

你是否担心退出选项越来越少，而且越来越现实？我的意思是，老实说，如果今天你没有 10 亿美元营收，就很难上市。科技收购方，那些大公司，对想买什么非常挑剔。私募股权公司正因投资组合里全是“奖牌”项目而舔舐伤口。

Do you worry about the lack of exit options that are now becoming more and more real? What I mean by that is, honestly, if you don't have a billion in revenue today, it's hard to go public. Tech acquirers, your big companies, are very specific about what they want to buy. PE is licking its wounds from having a portfolio that's full of medallions.

环境确实艰难。创业从来都不容易。事实上，在我看到的过去 25 年里，我觉得如今创办一家初创公司并成功退出的难度比过去要低。这是一场残酷的游戏。

It's a tough landscape. Startups have never been easy. In fact, in the last 25 years that I've seen, I would say it's been easier to build a startup and get a good exit from it these days than it used to be in the past. It's a brutal game.

从外部看，人们对创始人兼 CEO 有什么不了解但应该知道的事？

What does no one know about being a founder and CEO from the outside that they should know?

这不是一份光鲜的工作。它实际上是压力最大的事情之一，你真的得有点疯狂。

That it's not a sexy job. It's actually one of the most stressful things, and you really have to be crazy.

我觉得他们现在知道了。对我来说有一点是，你必须持续保持不满足。你永远不应该感到满足。作为 CEO，因为总有事情需要做，可以做得更好。

I think they know that now. I think one for me is that you have to consistently be unhappy. You should never be happy. As a CEO, because there's always something that needs doing, could be done better.

告诉别人你永远不会满足，他们会目瞪口呆。

Telling someone you will never be happy is something they're like jawed by.

对，说得好。这份工作方方面面都很艰难。我觉得很多时候，没做过的人会觉得它充满魅力，觉得这能赚大钱，生活会很精彩，会赢得很多尊重。但我觉得这些几乎都无关紧要。作为创始人，你必须真正以使命为导向才能生存。

Yeah, that's a good one. This is a tough job all around, and I think often times people who have not done it feel that there's a lot of glamour, they feel that this is going to make a lot of money and their life will be fantastic, they're going to have a lot of respect. And I think almost all of those things are irrelevant. You have to be truly mission-oriented to survive as a founder.

你的风格会随着金钱改变吗？你以前就成功过。

Did your style change with money? You've been successful before.

坦率地说，我认为创始人和投资者在已经富有时会做得更好。你会做出更理性、更稳健的决策，而不是出于经济上的急躁。对我来说，也许不是那样，但与此同时，我是一个需求极简的人，我的需求很久以前就满足了。所以我猜我创建这些初创公司时，确实没有那种“能否养家糊口”的担忧。所以也许这对我有帮助。但随着我取得更多成功，它并没有从根本上改变我。你仍然必须有那种动力，必须持续工作。你必须比公司里其他任何人都更努力，以身作则，不断推进。而且你必须有一种非理性的需求，去成就一番大事。

I think founders are better and investors are better when they are already rich, if I'm being blunt. I think you make more rational sound decisions that are not made with economic impatience. For me, maybe not like that, but at the same time, I'm a man with minimal needs, and my needs were already met a long time back. So I guess I've definitely built these startups without that worry of can I feed my family. So maybe that has helped me. But as I've seen more success, it hasn't changed me fundamentally. You still have to have that drive, you have to work continuously. You have to work more than every other person in your company, lead by example, and keep pushing. And you have to have this irrational need to make something big happen.

非常感谢你的时间。我为刚才讨论中态度强硬道歉。

I so appreciate your time. I apologize for being robust in my discussion back.

我觉得这和你做过的很多采访不同，它更像一场漫谈，但我非常感谢你的时间，你太棒了，伙计。

I think it was a different interview to a lot of interviews that you do, where it was more discursive, but I so appreciate the time and you've been fantastic, dude.

谢谢。

Thank you.