为何 AI 需要你公司的上下文

大家好。我叫 Christian。欢迎回到 TNW。今天我们的嘉宾是 Arvind，Glean 的创始人兼 CEO。首先，请多介绍一下你自己。

Hello everyone. My name is Christian. Welcome back to TNW. Today, our guest is Arvind, the founder and CEO at Glean. First of all, please tell us more about yourself.

嗯，感谢邀请。我叫 Arvind，是 Glean 的创始人兼 CEO。对于不了解 Glean 的人，我们是一家企业 AI 公司。它成立于 2019 年初，这使我们成为世界上第一家企业的生成式 AI 公司。当我们起步时，我们把自己看作是你工作生活中的 Google。所以，Glean 进入你的公司内部。它连接你所有的企业系统。它找出你公司内所有数据和知识的位置，然后它是一个每个员工都可以用来在公司内查找信息的工具。这就是我们的起步方式，然后多年来随着 AI 变得更好，这个工具演变成了今天，它是一个非常强大的 AI 同事，连接到你公司的所有上下文，并且拥有其他人正在构建的所有优秀 AI 模型，作为那个协作设置的一部分，这样你就可以进入 Glean。它会感觉像是 ChatGPT、Claude、Gemini 的超集，基本上用它来完成你的工作。它成为你非常亲密的伙伴。然后我们也是一个智能体平台和 AI 平台，最终就像我们在 Glean 的使命是理解你的业务，理解人们如何工作，在公司内完成工作，并将他们的智能，他们的集体智能带入一个上下文平台，并以一种方式打包，使得 AI 智能体也能从那种智能中受益，并达到你的人类专家的水平。

Well, thank you for having me. My name is Arvind and I'm the founder and CEO of Glean. For those who don't know Glean, we are an enterprise AI company. It got started in early 2019, which makes us the world's first enterprise generative AI company. When we got started, we were think of us as a Google for your work life. So, Glean comes in inside your company. It connects with all of your enterprise systems. It figures out where all the data, all the knowledge inside your company is and then it's a tool that every employee can use to find information inside the company. That's how we got started and then over the years as AI got better, the tool evolved into today, it's a very powerful AI co-worker that's connected to all of your company's context, as well as has all the great AI models that other people are building as part of that co-working setup so that you can go in Glean. It will feel like a super set of ChatGPT, Claude, Gemini and basically do your work with that. It becomes a very close companion to you. And then we're also an agent platform and an AI platform that ultimately like our mission at Glean is to understand your business, to understand how people work, get work done inside the company and bring their intelligence, their collective intelligence into a context platform and package it in a way so that AI agents can also benefit from that intelligence and work at the level of your human experts.

说实话，我在六月的伦敦科技周上做了 Glean 的演示，我很震惊地看到这样的东西存在，而我之前不知道。所以基本上，你把所有的知识都放在一个产品里。你进行的任何对话，任何同事在电子邮件、LinkedIn、内部、Slack 上与某人交谈，你都有平台内的每一条信息。所以这太棒了。恭喜。我想问一下你看到的最常见的用例是什么，以及为什么公司需要像 Glean 这样的东西。

To be honest, I did the demo of Glean at the London Tech Week in June and I was shocked to see that something like this exists and I didn't know about this. So basically, you have all of the knowledge inside one product. Any conversation you had, any colleague that talked to someone on the email, on LinkedIn, on internally, on Slack, you have every single piece of information inside the platform. So that's amazing. Congratulations. I wanted to ask about the use case that you see most with companies and why they need something like Glean.

是的。嗯，我们一直关注的是如何让人们在工作中更高效？如何帮助他们利用自己的才能做出最非凡的工作？哪些事情常常成为摩擦？而信息实际上是关键之一。比如拥有正确的数据、正确的知识信息，最终会消耗你作为员工的大量时间。所以这就是我们解决的问题。但就用例而言，如果你想想人们试图用 AI 做什么，AI 具有类似人类的能力，这意味着企业非常兴奋地想要实际承担一些人类今天花费大量时间的工作，并希望通过 AI 的帮助来完成它们。无论他们能在哪里用 AI 替代部分人类工作，那都会给他们带来杠杆效应，对吧？所以这就是为什么这项技术令人兴奋的原因。企业正在考虑所有不同的部门，看看他们能在哪里用 AI 带来这些效率。所以今天人们用 Glean 做什么，我们是一个水平平台。我们连接你所有的企业系统。这些是你公司内不同员工使用的系统。例如，我们连接像 Salesforce 和 HubSpot 这样的 CRM 系统，这是你跟踪客户信息的地方，你的销售人员经常使用这些系统。我们还连接工程系统，如 Confluence、Jira 和 GitHub，开发人员在那里做他们的工作。

Yeah. Well, so we've always been about how do you make people more productive at work? How do you help them do the most extraordinary work that they could with their talent? What are the things that often come as way of friction? And information is actually one of that key things. Like having the right sort of data, the right knowledge information is what ends up consuming a lot of your time as an employee. And so that's sort of what we solve for. But in terms of use cases, so if you think about what people are trying to do with AI, AI has human-like capabilities, which means businesses are very excited to actually take some of the work that humans are spending a lot of time on today and actually hopefully do them with the help of AI. Wherever they can replace parts of human work with AI, that gives them leverage, right? So that's the reason why there's a lot of excitement for this technology. And businesses are thinking across all of the different departments and seeing where they can bring those efficiencies with AI. So what people do with Glean today, we are a horizontal platform. We connect with all of your enterprise systems. And these are systems that different employees inside your companies use. For example, we connect with CRM systems like Salesforce and HubSpot, which is where you track information about your customers and your sales people use those systems a lot. We also connect with engineering systems like Confluence and Jira and GitHub, where developers do their work.

只针对一个部门吗？

Only for one department?

它是针对整个公司的。

It's for the entire company.

没错。

Exactly.

所以，我们连接所有这些。我们连接 HRIS 系统、ITSM 系统。基本上，今天人们工作的所有不同部门，Glean 都连接所有这些系统。它实际上观察并找出每个系统内有哪些数据和知识。它还理解今天的人类流程。比如，每天执行哪些不同的业务流程？人们实际上如何聚集在一起完成那项工作？然后从他们的智能中学习，现在开始把人们花费大量时间的每个业务流程，看看你是否真的能把 AI 作为辅助或作为完整解决方案来运行那些业务流程。对吧？所以，我认为这就是为什么在不同部门构建智能体的方法对于基本完成 AI 转型之旅至关重要，或者尽可能多地在公司内带来自动化，比以前做得更好。所以，这就是为什么它如此重要。所以，我们看到所有这些团队的用例。我给你们举几个例子。

So, we connect with all of these. We connect with HRIS systems, with ITSM systems. Basically, wherever people work today across all of your different departments, Glean connects with all of those systems. And it actually observes and figures out what data and knowledge sits inside each one of those systems. It also understands human processes today. Like, what are the different business processes that get executed every day? How people actually come together to finish that work? And then learn from their intelligence and now start to take each one of those business processes where people spend a lot of time and see if you can actually bring AI as an aid or as a full solution to run those business processes. Right? So, that's why I think like this is this approach of building agents across your different departments is critical to basically completing that AI transformation journey and or bring as much automation as you can inside your company do work better than before. So, that's why it's so important. So, like we see use cases around all these teams. I'll give you like a few examples.

请。请。

Please. Please.

例如，销售团队使用 Glean 来寻找潜在客户和新客户。

Sales teams, for example, they use Glean to prospect and find new customers.

而且不要寻找相同的客户，对吧？

And don't prospect the same customers and don't right?

没错。是的。然后实际上要聪明地寻找他们。例如，找出是否有关于那家公司的相关新闻？找出我们是否已经对那个客户的一些人做了很多外联，所以不要给他们发垃圾邮件。所以，所有这些内部外部的上下文帮助他们更好地寻找潜在客户。同样，如果你要开客户会议，作为卖家你如何最好地准备？

Exactly. Yeah. And then actually like prospect them smartly. For example, figure out, is there some news about that company that is relevant? Figure out if we have already done a lot of outreach to some people in that customer so don't spam them. So, all of that context internally externally helps them prospect much better. Similarly, if you're going to have a customer meeting, how are you best prepared for that as a seller?

所有的知识、讨论、新闻、融资轮次，一切。

All the knowledge, the discussions, the news, the funding rounds, everything.

没错。所以基本上，有了 Glean，你的卖家会得到会议简报，比如我如何以最有效的方式实际运行这个会议并提高你的赢率。同样，你的客户支持团队，不必完全由他们自己解决每个客户问题和工单，你可以使用 AI 实际解决大多数工单，因为无论客户有什么问题，很可能其他人以前也遇到过，AI 可以从过去的历史中提取信息来实际帮助更快地解决工单。所以，用例实际上跨越所有不同的团队，法律、财务、人力资源、IT、工程。而我们作为一个水平平台实际上是企业的热爱，因为有了一个平台，现在他们可以把 AI 带到所有不同的部门。

Exactly. So, basically, with Glean, your sellers will get a meeting brief, like how do I actually run this meeting in the most effective manner and increase your win rates. Similarly, your customer support teams, instead of them having to resolve every customer problem and ticket all by themselves fully, you can use AI to actually resolve most of the tickets because whatever problem the customers have, most likely other people have faced that before and AI can pull from that past history to actually help resolve tickets faster. So, the use cases are actually across all the different teams, legal, finance, HR, IT, engineering. And us being a horizontal platform actually is what enterprises love a lot because with one platform, now they can bring AI across all of the different departments.

但为什么不使用像 Claude 这样的东西，例如，把你部门的所有数据都放在 Claude 上？为什么人们应该使用 Glean？有什么区别？

But why not use something like Claude, for example, to keep all your data in your own department on Claude? Why should people use Glean? What is the difference?

是的。

Yeah.

我想说的第一点是，你可以使用多个模型。

I would say the first one would be that you can use multiple models.

是的，这是很重要的一点，对吧？因为你知道，Claude 是非常昂贵的模型，而如今 85% 到 90% 的企业任务，我们宁愿用其他模型而不是 Claude，因为有些模型，比如 GPT 的模型，成本低一个数量级，但在完成大多数企业任务上同样出色。所以这是 Glean 的一大优势。把它作为你的核心 AI 系统和平台，意味着你基本上能享受到行业里所有的创新。你可以使用各种不同的模型，为任何给定任务挑选合适的那个，而不会让你破产。

Yeah, that's a big one, right? Because, you know, Claude is a very expensive model, and for 85 to 90% of enterprise tasks today, we would rather use some other model than Claude because there are models like from GPT even that cost an order of magnitude less but are equally capable of doing most of your enterprise tasks. So that's one big advantage of Glean. Using it as your core AI system and platform means you basically get to enjoy all the innovation that's happening in the industry. You get to work with all the different models and you can pick the right one for any given task, which is going to break the bank.

我明白，我明白。那是一个很大的优势，但实际上只是部分优势。事实上，我们并不认为那是 Glean 的主要优势。Glean 的主要优势其实是我们能够以非常深入的方式连接所有企业系统。所以不仅仅是像 MCP 服务器那样能看到那些系统中的一些信息，而是真正更深入地理解你的业务，构建这些复杂的知识图谱，理解公司内部的核心概念、核心实体，围绕这些构建完整的上下文，因为即使是 AI，要让它发挥最佳水平，你也需要给它正确的原材料、正确的输入、正确的上下文、正确的信息，才能帮助它做得非常好。否则，就是垃圾进，垃圾出。所以，因为我们的根基是搜索，在任何给定主题上，Glean 都非常擅长收集最新、新鲜、高质量且最适合使用的信息。这让我们实际上比 Claude 或 ChatGPT 这类 AI 原生工具工作得更好。事实上，不要把 Glean 看作 Claude 或 ChatGPT 的竞争对手。虽然我们为能成为一个更强大的同事而自豪，拥有更多模型、更多公司上下文，但你可以把 Glean 当作一个无头平台来使用。所以甚至不要把我们当作独立产品。只把我们看作一个上下文和智能层。你可以把这一层连接到 Claude 内部。所以作为一家公司，你可以继续使用 Claude。

I see. I see. That's one big advantage, but that's actually only the partial one. In fact, we don't think of that as a primary advantage of Glean. The primary advantage of Glean is actually our ability to connect with all of the enterprise systems in a very deep manner. So not just like an MCP server that can sort of see some information that's in those systems, but actually go more deeply and understand your business, build these complex knowledge graphs, understand the core concepts inside a company, the core entities, build full comprehensive context around those because even AI, for it to do its best work, you need to give it the right raw materials, the right input, the right context, the right information that's going to help it do a very good job. Otherwise, it becomes garbage in, garbage out. And so, because our roots are search, on any given topic, Glean is very good at collecting the information that is up-to-date, fresh, high-quality, and the best one to use. That allows us to actually work much better than these AI-native tools like Claude or ChatGPT. In fact, think of Glean not even as a competition to Claude or ChatGPT. While we are proud of being a more powerful co-worker with more models, with more company context, you can use Glean as a headless platform. So don't even use us as an independent product. Only think of us as a context and intelligence layer. And you can actually connect that layer inside Claude. So as a company, you can keep using Claude.

我明白了。

I see.

而个人用户可以进入 Claude，运行一个智能体，或者提问，或者给它一个任务。Claude 会完成工作。

And individual users can go into Claude and run an agent or ask a question or give it a task. Claude will do the work.

提供知识来自

Giving the knowledge from

是的，但 Claude 不必实时连接 100 个不同的系统来构建上下文，它只需向 Glean 发送请求，获取所有正确的上下文，这样——我们已经发布了一些研究——我们让 Claude 快两倍。我们让它消耗一半的 token。我们让它的工作被偏好五倍于现在。所以我们在这里所做的只是解决 AI 如何为你工作的拼图中的一部分，即把正确的上下文和正确的智能，以及工作在你公司内部如何发生的经验，带给像 Claude 这样的优秀 AI 工具，让它们工作得更好。所以我们非常自豪地告诉客户：“嘿，看，你不必把我们当作终端用户产品。你只需引入 Glean，它就会让你现有的所有 AI 平台变得更好。”

Yeah, but Claude, instead of having to build that context in real time by connecting to 100 different systems, it'll just send a request to Glean, get all the right context, and that way—and we've released some studies—we make Claude twice as fast. We make it consume half as many tokens. And we make its work preferred five times more than what it is right now. So all we're doing here is we're doing one part of that puzzle of how AI can work for you, which is bring the right context and the right intelligence, the experience of how work happens inside your company, to these great AI tools like Claude and just make them work better. So we're very proud to actually tell our customers, "Hey, look, you don't have to use us as an end-user product. You just bring Glean in, it's going to make all of your existing current AI platforms better."

你不觉得这有一天会被取代吗？

Don't you think that can be replaced at some point?

是的，我是说，但那没关系。就像，看，作为一家科技公司，你只有持续创新才能留在市场并取得成功。我认为如果你不能——你不能通过制造锁定来获胜——或者至少，那不是我们的心态。我们不是试图做一些让企业难以取代我们的事情。事实上，恰恰相反：我们的目标是让我们非常容易被取代。因为我们想让企业掌控。我们希望他们觉得，有了 Glean，他们完全拥有 AI 随时间建立起来的所有那些惊人的学习成果。这非常重要

Yeah, I mean, but that's fine. Like, look, as a technology company, you only get to stay in the market and succeed if you're going to continuously innovate. And I think if you cannot—you cannot win by creating lock-in in our—or at least, that's not our mindset. We're not trying to do something where it becomes hard for the enterprises to replace us. In fact, the very opposite of that: our goal is to actually make us very easily replaceable. Because we want to put the enterprises in control. We want them to feel that with Glean they are keeping full ownership of all of those amazing learnings that AI is building over time. And it's very important

数据。

data.

那是非常重要的数据，我要谈两点。一是协同工作。未来我们都会大量与 AI 一起工作。随着时间的推移，我们的同事，我们的个人同事，会非常了解我们喜欢如何工作。凭借所有这些学习，它对我们变得非常高效。它拥有关于我们、我们的需求、我们的风格的记忆。现在，谁拥有那个记忆？对于 Claude、OpenAI，他们拥有那个记忆。一段时间后，替换那些工具就变得困难，因为他们对你的员工了解得比你还多。这样一来，企业实际上失去了控制。那个智能现在不再归他们所有。所以我们在 Glean 的做法不同。我们建立的关于个人及其偏好和喜欢的工作方式的所有记忆都以标准格式存储，并且实际上归企业所有。所以这样，我们变得容易被替换，因为所有智能——对吧？是的。但我认为企业喜欢这样，因为他们希望如此。智能体也一样。当你构建智能体时，通常智能体第一天表现不好。但大多数智能体中通常有人类在环。随着时间的推移，人类给出足够的反馈，使智能体工作得非常好。但在大多数其他智能体平台中，那些复合学习成果归平台所有。而在 Glean，那些复合学习成果归企业所有。所以这是一个策略。这对我们来说是一个商业策略。我们想要与众不同，我们想要凭借我们技术的优点获胜，而不是制造锁定。

That's very important data and I'll talk about two things. One is co-working. So we're all going to be working a lot with AI in the future. And over time, our co-worker, our personal co-worker, gets very good at knowing how we like to work. It becomes very effective for us with all those learnings. It has the memory of us, our needs, our style. Now, who owns that memory? So with Claude, with OpenAI, they own that memory. And it becomes hard to replace those tools after a while because they know so much about your employee more than what you know. And so, but that way the enterprise actually loses control. That intelligence now is out of their own ownership. And so we do it differently at Glean. All the memory that we build about an individual and their preferences and how they like to work are stored in a standard format and actually is owned by the enterprise. So that way, we become easily replaceable because all the intelligence—Right? Yeah. But I think enterprises like it because they want it to be like that. And same for agents. As you build agents, often times agents don't perform well on day one. But there's a human in the loop typically in most agents. Over time, the humans give enough feedback which makes the agents work very well. But those compounding learnings in most other agent platforms are owned by the platform. And in Glean, those compounding learnings are owned by the enterprise. So it's a strategy. It's a business strategy for us. We want to be different and we want to win based on the merit of our technology as opposed to creating lock-ins.

我真的很有一个个人问题刚刚出现在我脑海中。你们如何处理员工隐私？例如，如果他们想保持电子邮件私密，或 LinkedIn 消息私密，或对话私密，这是发生过的事情吗，你们如何处理？

I really have a personal question that just came to my mind. How do you deal with employee privacy? So, for example, if they want to keep their emails private or their LinkedIn messages private or their conversation private, is this something that happened and how do you handle it?

是的，首先，在企业环境中，Glean 完全尊重企业内的隐私框架。例如，作为员工，你的公司邮箱收件箱只有你能查看，公司内其他任何人都不能。

Yeah, so first of all, within the context of an enterprise, Glean fully respects the privacy frameworks that you have in the enterprise. So, for example, as an employee, your corporate email inbox is only viewable by you and nobody else inside the company.

好的。

Okay.

对吧？这已经是现状了。

Correct? That's already the case.

除了法律保留或类似会产生访问权限的情况外，其他人看不到。Glean 会自动遵守这一点。即使它连接所有用户的收件箱，也不会让用户一访问用户二的收件箱。它尊重权限和治理。

Except for legal holds or similar situations that create access, nobody else can see it. Glean automatically respects that. Even though it connects with all the inboxes of all the users, it's not going to let user one access user two's inbox. It respects permissions and governance.

你只是搜索不同的东西，它连接到邮箱，但不一定连接到邮箱。告诉你这是获取信息的邮箱，对吧？

You just search different things, and it connects to the mailboxes, but doesn't necessarily connect to the mailbox. Tells you this was the mailbox that got the information, right?

是的，基本上，当你在 Glean 中工作时，我们知道你是谁，我们不会让你使用公司内部其他人的私有信息。

Yeah, so basically, when you work in Glean, we know who you are, and we will not let you use information that is private to somebody else inside your company.

这是由 AI 决定它是私有的吗？

And this is something that AI decides that it's private.

不是由 AI 决定的。它实际上更程序化、确定性。我们不能使用像 AI 这样的技术，因为它模糊且可能出错。所以信息保护必须更程序化和确定性。我们就是这样做的。这是我们解决方案的一部分。第二件事是你还提到了一些技术上不是企业数据的数据，比如你的 LinkedIn 消息。Glean 同样不会让其他人使用它。你仍然控制你的数据，事实上，这些数据永远不会进入企业范围。你可以用它做一些工作，但数据留在原处，不会提供给企业内的其他人。

Not decided by AI. It's actually more programmatic, deterministic. We cannot use a technology like AI because it's fuzzy and it can make mistakes. So information protection has to be more programmatic and deterministic. So we do that. That's one part of our solution. Second thing is that you also mentioned some data which is technically not enterprise data, like your LinkedIn messages. Glean again will not let anybody else use that. You remain in control of your data, and in fact, that data will never enter the enterprise purview in that sense. You can use it to do some work, but that data stays where it is and it's not going to be made available to anybody else within your enterprise.

好的。我知道你的 ARR 达到了 3 亿美元，过去 15 个月增长巨大，我想多谈谈这个。请告诉我们公司的故事，你是如何获得第一百万收入的，如何增长，以及公司的下一步计划。

Okay. I know you reached 300 million in ARR and you have a massive growth in the past 15 months and I want to talk more about that. So, please tell us the story of the company, how you started getting the first million in revenue, growing beyond and what are the next plans for the company.

是的，Glean 始于 2019 年初。我们最初是一家搜索公司。我们想为你的新工作生活打造一个 Google。非常有趣的是，我们不得不使用 Transformer 和语言模型来构建那个产品。

Yeah, so Glean started in early 2019. We started as a search company. We wanted to build a Google for your new work life. And very interestingly we had to use transformer and language models to build that product.

好的。

Okay.

当时 Transformer 语言模型并不像今天这样常见。现在每个人都知道 Transformer。2019 年，没人真正关注它们。它更像是一个小众概念。最初，它是为了让 Google 搜索更好而构建的。所以，当我们说我们要为企业构建一个非常好的搜索产品时，使用它是有道理的。所以，我们就这样开始了。我们构建了一个非常好的搜索产品。我们很快成为最好的，事实上是唯一在企业内部工作的搜索产品。这就是让我们建立业务和收入的原因。旅程很漫长。在公司成立的头 18 个月里，我们甚至没有费心去赚钱。我们在构建这个产品，我们知道这是一个复杂的东西。复杂的产品，但一旦我们构建出来，它将对 businesses 非常有价值。

Transformer language models at the time were not as common as they are today. Now everybody knows about transformers. In 2019, nobody really bothered with them. It was more a niche concept. Originally, it was built to make Google search better. So, when we said we're going to build a really good search product for businesses, it made sense for us to use it. So, that's how we got started. We built a really good search product. We quickly became known as the best and in fact the only search product that works inside a business. And that's what allowed us to build our business and revenue. The journey was long. For the first 18 months of our company, we didn't even bother to make any money. We were building this product and we knew it was a complex thing to build. Complex product to build, but once we build it, it's going to be a very, very valuable product for businesses.

口碑是最大的增长渠道吗？

Was the word of mouth the biggest growth channel?

口碑事实上甚至今天也是我们最大的增长渠道，因为我们不是一家消费者公司。我们不对全世界每个人做营销。我们有超过 1,000 家大型企业使用我们。所以，我们有数百万用户使用该产品。通常，当这些人从一家公司转到另一家公司时，他们实际上会把 Glean 带到那些新公司。所以，我们是一个非常口碑驱动的业务。

Word of mouth was in fact even today, that's the biggest growth channel for us because we're not a consumer company. We don't do marketing to everybody in the world. We have over 1,000 large enterprises who use us. So, we have millions of users using the product. And typically, those are the people when they move from one company to other, they actually bring Glean in those new companies. So, we are very word of mouth driven business.

增长方面，过去几个月你增长了很多，对吧？

Of growth, you grow a lot these past months, right?

我们正以非常快的速度增长。确切地说是指数级增长。这是因为 AI 实际上企业有很大的胃口。每个人都想转型他们的业务，AI 被视为对大多数企业来说成败攸关的技术。如果你最好地拥抱它，你的竞争优势就会更强。所以这为我们的产品创造了很多需求。但此外，我们非常独特。从我们在市场中的定位来看，我们不与 OpenAI 或 Anthropic 或 Google 或 Space X 竞争。

We are growing at a very rapid pace. Exponential growth to be precise. And that's because AI is actually there's a lot of appetite that businesses have. Everybody wants to transform their business and AI is seen as a technology that's going to be make or break for most businesses. If you embrace it the best, then your competitive advantages are going to be stronger. And so that creates a lot of demand for our product. But then also, we are very unique. In the sense that if you think about how we are placed in the market, we're not competing with OpenAI or Anthropic or Google or Space X.

Soft Copilot？

Soft Copilot?

也许有一点。但我们的故事真正是关于上下文的。是关于理解你的业务如何运作。

Maybe a little bit with that. But our story really is about context. It is about understanding how your business works.

我认为对于智能体来说，拥有那样的上下文真的非常重要，对吧？

And I think for agents, it's really really important to have that context, right?

没错。上下文对于让 AI 在企业中发挥作用至关重要。而上下文是一个被严重误解的概念。今天很多人谈论上下文。我们开创了这个概念。我们大约三四年前开始谈论它。我们谈论时是在教育市场，你不能仅仅构建和训练能在你的业务中做所有事情的 AI 模型。你必须实际创建这个上下文引擎的概念，并将其与 AI 模型结合。它们一起才能真正驱动业务价值。所以，我们已经谈论了很长时间，但让我实际分享一下上下文到底是什么。它从根本上理解你的业务、你的员工、你的客户、你的各个业务流程，以及人们今天如何完成这些工作项。你的很多企业智能，工作如何发生，并没有记录在你的结构化数据系统或文档中。很多知识是隐性的。它在人们的头脑中。

Exactly. Context is super important to make AI work in the enterprise. And context is a very misunderstood concept. A lot of people talk about context today. We pioneered the concept. We started talking about it like three or four years back. We talked as we were educating the market that you just cannot build and train AI models that can do everything in your business. You have to actually create this concept of a context engine and combine it with AI models. Together, they can actually drive business value. And so, we've been talking about it for a long time, but let me actually share what context really is. It's about fundamentally understanding your business, your people, your customers, your individual business processes, and how people complete those work items today. A lot of your enterprise intelligence, how work happens, is not documented in your structured data systems or your documents. A lot of knowledge is tacit. It's in people's heads.

而且是最重要的部分。

And the most important part.

是的。可能吧。

Yeah. Probably.

是的，就像你知道的，想想当你想要做出非常复杂、高影响力的商业决策时，你最依赖谁？你知道，公司里的每个人都可以访问所有信息。

Yeah, like you know, and it's like you know, think about when you have when you want to make very complex, high-impact business decisions, who do you rely most on? You know, everybody in the company has access to all the information.

是的。

Yeah.

但仅仅访问信息是不够的。你依赖某个长期做那项工作的人，因为

But just having access to information is not enough. You rely on somebody who's been doing that work for a long time because

有上下文，有过去的经验，然后他们可以

has the context, has the past experience, and then they can

而很多经验只在他们头脑中，对吧？没有记录在文档中。

and a lot of that experience is only in their head, right? It's not documented in documents.

对。

Right.

所以当我们谈论上下文时，这就是它的含义。你必须以某种方式捕捉那种人类智能。其中一些表达在你的数据库、结构化数据系统中。

And so when we talk about context, this is what it means. You have to somehow capture that human intelligence. Some of it is expressed in your databases, your structured data systems.

有些信息体现在你的商业智能仪表盘里，有些在你的政策文件、用户指南和操作手册里，针对不同的业务流程。但其中 80% 到 90% 其实存在于人们的头脑中，不过他们确实会表达出来。他们在日常工作中表达出来，比如他们互相交谈的时候。所以这就是真正的上下文所捕捉的——人们在日常工作中所表达的那最后一点人类智慧。这就是 Glean 所做的。我们深入挖掘，观察每一次人际互动，看人们做什么，从而了解他们今天是如何工作的。然后我们把这些作为上下文，带入你的 AI 系统。

Some of it is expressed in your business intelligence dashboards. Some of it is in your policy documents and user guides and playbooks, you know, for different business processes. But 80 to 90% of it is, you know, in people's heads, but they do express it. They express it in their day-to-day work. You know, when they talk to each other. And so this is what real context is is capturing like, you know, that last ounce of human intelligence that people express in their day-to-day work. And so that's what Glean does. We go very deep. We look at every single human interaction. We see what people do. And that is how we learn how they would do do their work today. And now we take that as context and bring it to your AI systems.

不幸的是，我们剩下的时间不多了，但我会快速问几个问题。好吗？所以，请你直截了当地回答。自公司成立以来最大的挑战是什么？

Unfortunately, we don't have that much time left, but I will shoot a couple of rapid-fire questions. Okay? So, I will ask you to be straight to the point. Biggest challenges since starting the company?

是规模扩张的问题，因为市场需求巨大，要确保我们能招到合适的人，并让他们协同工作，与使命对齐。实际上，我一直觉得对我们来说，从来不是市场对我们产品有大量需求的问题，而是我们能否执行、能否作为一个团队做好工作的问题。

It's about scaling it because we've such huge demand in the market and like, you know, making sure that we can hire the right people and we can all make them work together, you know, aligned on the mission. That's actually I always find like for us it's never been the market where there's a lot of demand for our product. It's about can we execute and can we do a good job as a team?

我在伦敦科技周见过你的团队，我可以确认他们是非常好的人，价值观很好。嗯，你们现在实际雇用了多少人？

I met your team at London Tech Week and I can confirm they are really nice people, good values. Um and how many people do you actually employ now?

我们大约有 1500 人。

We are about 1,500 people.

好的，那相当多。嗯，你提到你们不需要再融资，资金充足，这方面没问题。你们计划未来融资吗？Glean 的下一步计划是什么？你们梦想上市吗？

Okay, that's uh that's quite a lot. Um you mentioned that you don't need to raise any more, you have enough money, you are good on that side. Do you Do you plan to do a future fundraising or what is the plan next for Glimp? Do you Do you dream to go public?

不，嗯，没有融资计划，我也要说没有真正考虑上市或其他事情。我觉得 AI 行业变化太快，现在很难有那种稳定感。

No Well, no plans for fundraising and I would also say like no real plans to think about, you know, going public or others. Like I feel like AI industry is so fast-moving and it is hard to like really have that feeling of stability right now.

甚至 3 个月的计划都不再可行，对吧？

Even a plan in like 3 months is no longer available, right?

是的，所以我认为这需要时间。我觉得很多 AI 公司必须花时间建立更稳健的业务。嗯，所以没有近期计划。当然，我们的业务规模不错，如果想上市也可以，我们也在确保内部流程到位，准备好能够做到这一点，但没有具体计划。

Yeah, so so I think like it will take time. Like I do feel like a lot of AI companies have to take their time and establish businesses, you know, that are more robust. Um so so there's no like near-term horizon plan. Like, you know, we Of course like, you know, our business is at a good scale. We could go public if you wanted to, but and we are making sure that our internal processes are you know, are getting there, you know, to be ready to do to be able to do that, but there no there no there's no concrete plans.

作为这家组织的 CEO，你如何度过你的一天？

How do you spend your day as the as the CEO of this organization?

嗯，主要是，我想说我希望做的事，我认为我现在做的事，我并不自豪。

Well, mostly I like I I would say what I would like to do I think what I do right now, I'm not proud of it.

我想说我需要做的首先是真正专注于建设我们的团队。嗯，我们有巨大的潜力。我们将成为最重要的公司之一，我们是先驱，是世界上第一家的企业级 AI 公司。我们今天将是一家非常重要的公司，我们希望成为未来最重要的企业级 AI 公司之一，这要求我们真正用优秀的人才来建设这个团队，所以作为 CEO，我应该把大部分时间花在建设优秀团队上，因为他们才是公司的支柱。

I would say what I need to do is first and you really need to focus on building our team. Um we are we have a huge potential. We're going to be one of the most important, you know, we are the pioneers. We're the world's first enterprise AI company. We're going to be a very important company today and we want to be one of the most important enterprise AI companies of the future and that requires us to truly um build this team with great talent and so I should be spending as a CEO most of my time building a great team because they are the ones, you know, who are going to be the pillars of the company.

你实际上是如何开始你的职业生涯的？你的计划是什么？你的热情是什么？你是如何——过程是怎样的？

How did you actually start your career? What what was your plan? What were your passions? How did you What was the process?

是的，我毕业后一直在科技行业，所以那是我的计划。我学了计算机科学，我想成为一名开发者，我想构建系统。嗯，但我从来没有真正计划过，我不是一个长期规划的人，所以主要是做有趣的项目。

Yeah, so my I've been in the tech industry throughout um after graduating and so that was my plan. Like, you know, I I I I studied computer science. I wanted to be a developer. I wanted to build systems. Um but, you know, I never really planned for I'm not a long-term planning person and so most it is about you do interesting, you know, projects.

直到成功。

Until it works.

嗯，是的，我觉得我一直喜欢做的一件事——我很幸运有这种奢侈——是我喜欢解决非常常见的问题，每个人都会面临的问题。我相信，当你解决那些问题时，你就有巨大的市场，可以建立巨大的成功。

Uh and yeah, and and I think like one thing you know that I've always uh liked to do um I'm fortunate to have that luxury is I like to work on problems which are very um common place problems. Problems that everybody faces. And I believe you know, when you solve for uh when you solve for those problems, you know, then you have a huge market and you can build, you know, big success.

太棒了。对于刚完成学业、想进入 AI 领域、崇拜像 Arvind 这样的人并想有所建树的年轻人，你最好的建议是什么？

Amazing. And what would be your best piece of advice for a youngster, someone who just finished their studies, um is looking into getting into AI, admires someone like Arvind and wants to build?

是的，所以我对刚毕业的人的建议，首先是减少焦虑。不要担心市场上那些无稽之谈，比如软件工程的工作还会不会有？其他类型的工作还会不会有？工作会有的，我们正处在繁荣的积极一面，希望未来也是如此。所以不要过度紧张，作为进入职场的人，我的主要建议就是努力工作、保持专注、不要想太多，并且每天学习。不管你学的是什么，第一份工作、第二份工作是什么，只要努力做好，你的职业生涯就会有着落。

Yeah, uh so my advice to like like somebody who's graduating right now, first of all is to reduce anxiety. Do not worry about all the nonsense that you hear in the market about yeah, are there going to be software engineering jobs? Are there going to be other types of jobs? Like there's going to be jobs like, you know, we are you know, on the positive side of, you know, you know, you know, prosperity, hopefully, like, you know, in the future. So, so do not overstress like, you know, you like as a as a as somebody entering the workforce, the my main advice to them is to basically just work hard, stay focused, do not overthink, and and try to learn every day. And and it doesn't matter like, you know, what you studied, what are your first job, second jobs, as long as you try to do a good job in it, you know, that is you will be set for your career.

我明白了。你实际上在谷歌搜索工作过，对吧？

I see. And you actually worked at Google Search, right?

嗯。

Mhm.

这对 Glean 的创建有何影响？

How did this impact the creation of uh Glean?

我是说，它直接促成了这件事。我们创办了 Glean。

I mean, it's directly responsible for it. Like we we started Glean

大约多少年了？

How many years around

现在大约 8 年了。我们开始的时候，我实际上是 Rubrik 的创始人之一，那是我在建设的另一家公司，

it's about 8 years now. We start and when we started, I was actually one of the founders of Rubrik and it was another company that I was building,

好的。

Okay.

它现在是一家上市的大型企业。嗯，当时我们发展迅速，这是我们内部面临的问题：公司里没人能找到任何东西。所以 Glean 的想法就直接来源于此。

which is now um a publicly traded large enterprise. Uh and at the time we were growing fast and this was a problem that we were facing internally. Nobody could find anything inside our company. And so, the idea for Glean actually just directly came from that.

太棒了。我明白了。

Amazing. I see.

是的。

Yeah.

显然，你对公司运作有很多上下文。嗯，那么，今天公司运作的方式中，什么让你感到惊讶？一些如果你在业务之外就不会知道的事情，因为你实际上通过 Glean 看到内部他们如何工作、如何沟通、如何完成任务。

And obviously, you have a lot of um context about how company work. Um so, what actually surprised you about how companies work today? Something that you wouldn't know if you're outside the business because you actually see inside Glean uh how they work, how they communicate, how they uh yeah, do tasks.

是的，我是说，我实际上坦率地总是惊讶于完成简单的事情有多么复杂。而且随着越来越多的人加入公司，以前更容易完成的任务变得更难了。

Yeah, I mean, I I think like I'm actually frankly always surprised at how complex it is to get simple things done. And as as more and more people join the company, tasks that were easier to do before becomes harder

真的吗？

Really?

因为这一点，我认为有更多——人更多，协调更多，我觉得变得非常非常困难。

because of because of this I I think there's there are more, you know there are more people there's more coordination and and I think it becomes very very hard.

所以大型企业会随着时间的推移变得缓慢，因为需要协调，因为很难让所有人保持同步。很难把流程都记录下来。所以我对事情变得如此复杂感到震惊。我确实觉得 AI 是一项非常好的技术，可以真正解决一些这样的问题：组织工作、组织知识、让人们更好地相互联系。所有这些东西——我认为 AI 将成为一个很好的工具来帮助我们。

So large enterprises, they become slow over time because of the need for coordination, because it's so hard to actually keep everybody on the same page. It's so hard to actually keep your processes documented. So I'm shocked at how things become so complex. And I do feel like AI is a really good technology to actually solve some of these problems of organizing work, organizing knowledge, making people connect better with each other. All of those things are going to be—I think AI is going to be a great tool to help us with.

最后你提到 Glean 平均每天节省 11 小时——抱歉，是每周。

And last thing you mentioned that Glean saves on average 11 hours a day—a week, sorry.

是的。

Yeah.

嗯，为什么是这样，这是怎么发生的？

Um why is that and how does that happen?

嗯，首先，作为知识工作者，我们工作时间的三分之一都花在寻找信息上，而 Glean 立即减少了所有这些时间，因为我们会为你带来任何主题、任何问题的即时信息。所以 Glean 节省的大量时间仅仅是因为我们可以把正确的信息带给人们，当你有问题时，你不必去找谁是合适的人，然后打扰他们，等待他们回复你。你可以自助服务。你可以通过 Glean 立即获得基于最新、准确、高质量知识的所有答案。所以这就是我们为人们节省这么多时间的核心。但除此之外，因为我们现在是一个如此出色的协作平台，我们也在帮助你节省时间，做那些过去消耗你大量时间的事情。例如，如果你是一名产品经理，你试图写 PRD 或生成 PowerPoint 幻灯片，你向组织其他成员展示，就像所有这些日常知识工作者的任务。Glean 非常擅长为你做这些。所以影响实际上将远大于为某人每周节省 10 小时或 11 小时。影响实际上将是根本性地改变人们的工作方式，让他们做自己喜欢的工作。

Well see first like a third of all of our time working time we spend looking for information as knowledge workers and Glean immediately reduces all of that time because we will bring instant information to you on any topic any questions that you have. So a lot of time savings with Glean happen just simply because we can bring the right information to people and when you have questions like you don't have to go and find who are the right people and sort of bother them wait for them to respond back to you. You can self-serve yourself. You can instantly get all the answers based on latest accurate high-quality knowledge through Glean. So that's the core of how we save so much time for people. But then on top of that because now we're such a great co-working platform we're also helping you save time doing things which used to consume a lot of time for you. For example if you are a product manager you're trying to write PRDs or generate PowerPoint decks you present to the rest of the organization like all of those day-to-day knowledge worker tasks. Glean is very good at actually doing it for you. So the impact is actually going to be much greater than saving somebody 10 hours a week or 11 hours a week. The impact is actually going to be fundamentally transforming how people work and make them work they enjoy.

是的。

Yeah.

那是最重要的部分。

That's the most important part.

让他们——是的，就像帮助他们，当你享受工作时，你会完成更多。

Make them—yeah, like help them and when you enjoy your work, you get a lot more of it done.

为了改进，你想要成长，你想要——是的，所以我们想成为 10 倍提升者。就是这样。这是我们想要设定的最低标准。我们如何真正帮助你实现，完成 10 倍的工作量——以及比今天能产生的多 10 倍或更高质量的产出？

To improve, you want to grow, you want to—yeah, so we want to be a 10Xer. That's it. That's the minimum bar that we want to set. How can we actually help you achieve, do 10 times more work than—and 10 times more greater output or higher quality output than what you can produce today?

如果我说 Glean 有点像 Asana 和 Claude 的孩子，这样说错了吗？

Is it wrong if I say that Glean is something like Asana and Claude had a baby?

嗯，我们得稍微想想。这是相关的。我们确实使用了 Claude 的技能。我们确实依赖存在于 Asana 或 Jira 等系统中的信息和数据。那是你跟踪工作的地方。我们做的更多。所以我们还依赖很多其他东西。但这是个好类比。是的。

Well, let's have to think about that a little bit. It's relevant. We do use the skills of Claude. We do rely on information and data that lives in systems like Asana, or Jira. That's where you track work. We do a lot more. So we rely on a lot more other things too. But that's a good analogy. Yeah.

非常感谢你，Arvind。谢谢。

Thank you so much, Arvind. Thank you.

谢谢你邀请我。

Thank you for having me.

谢谢。

Thank you.