Hi everybody, good afternoon, good evening, and good morning wherever you are. Thank you very much for joining another episode of the Boardroom Club. My name is Yosi Gmazi and I'm joined here also by a good friend of mine, Senat Mully.
Hi everyone. Thank you so much for joining us on the Boardroom Club today. So today we are fortunate enough to interview none other than Arvind Jain, founder and CEO of Glean, a company that's revolutionizing how organizations discover, access, and leverage knowledge across workplaces through AI. Most recently they raised $150 million Series F at a $7.2 billion valuation. Arvind is also the founder of Rubrik, the cyber unicorn. So Arvind, thank you so much for joining us today.
Thank you for having me. Really excited to be here.
The pleasure is all ours. So Arvind, I want to start with a question about all the companies that you founded, as I mentioned in the intro, because you built around two data problems, right? With Rubrik, it was about protecting the data. Now with Glean, it's about making it useful. But was there any critical mindset shift, something that you needed to unlearn when you transitioned from Rubrik to Glean?
I mean, these two companies couldn't be more different. Of course, both companies are enterprise software companies, we build products and bring them to enterprises who need them, but the nature of the business, the problem was so different between these two companies. Rubrik was a product that was in an established market. We had many, many different competitors at the time when we started the company, and we had to go with a better technology, a modern product, and tap into the established budgets that companies already had to buy a product like ours, which was a data protection product. And now when we started Glean, suddenly the environment was completely different. We were selling a product that nobody was used to buying. There was no budget for it. There was no precedent for actually buying search products in the enterprise. And so we had to create the market. We had no competition, nobody to actually really go and compare ourselves with, but we had to actually convince the market that a product like this was needed. So these were two very different journeys. Also, the Glean product is a product that every single person in your company uses. The Rubrik product was a product that was used by only one or two people who were also the buyers of the product. So it fundamentally changes how you build the company. For example, in Glean we couldn't actually go and start selling the product and make revenue while at the same time building it, which is what we did at Rubrik. At Glean we didn't have that chance. We had to actually go and first build a product, get it to a really, really good place, because all the users of our product, they were not actually people that we knew. We didn't have relationships with one person at a company, but then there are a thousand other people who are going to use your product. So these were very different journeys, a lot of things that we had to unlearn, a lot of things that we had to do differently. And I can go endlessly on just this one question, but I'll stop there and see if you have any specific questions there.
It's fascinating, moving from selling only to the C-suite of the company to selling to all of the company's employees at the end, because they are the users, right? But what was the aha moment when you knew that the Glean idea was not only a nuanced business but rather a massive market opportunity? I've heard that you started at the Kleiner Perkins basement, but what was the moment that you actually understood that there is actually a big market opportunity there?
Well, I knew it. I felt it even before we started the company, and I had a very simple reason. I don't do market analysis. I don't do the total addressable market. In fact, it's funny that I didn't even know what TAM is when we were starting the company. I mean, I'm joking, but I kind of never really thought about that. What I knew was that every person who I knew found it hard to find the information that they needed to do their work. And you would have people complain all the time. In fact, we saw this problem at Rubrik. Everybody in our company was complaining that they cannot find the information that they need to do their work. In fact, they also could not find people who could help them figure out and help them in their day-to-day jobs. So the problem has kind of always been there, but nobody was actually able to build a successful business out of it. And when we were starting the company, I talked to a bunch of people and they all mostly identified that yes, this is a good problem, their employees have it, but they all also felt that maybe this was not the most important problem to be solved and it would be hard to build a company. So the advice that I was getting was the opposite, do not do this kind of a thing. But I always had the belief, and it was a journey, it took time before people started to understand it. But the moment when I felt that yes, we are going to be a huge big company, was when we got the first kudos from our customers who told us that hey, this is the very first time search has ever worked in the enterprise for us. You're the first ones to build that product. In fact, I hear it almost on a daily basis. Today I was in the morning in a meeting with one of the largest database companies in the world and I got exactly that same comment from them, that you are the first ones to make it work. So I think when we started to get that recognition that we built a good product was when we knew that yes, this is going to be a huge thing.
It's interesting because there were a few companies that tried but struggled very much, right? Like HP and even Google Search, but you succeeded. I was curious if you can tell us, maybe share with the audience, what was one of the contrarian technical decisions, technical bets that you took at the beginning that turned out to matter the most?
So first, I think the problem itself in some ways breaks the rule of startups. We started by saying that we're going to build a general product that's going to help every single person inside the company. And so it was kind of the opposite of focus in some ways. We were not actually saying that hey, can we go and help this particular user population, can we go and build a product for the customer support teams or for engineers. We said no, we want to actually solve the problem in a more horizontal manner for everybody, for every person in every company. And so that was sort of a little bit at odds with typical startup advice that you would get. But we did it because that was what we felt was a big opportunity for us and we wanted to build a broad, very large company, not a point solution for a specific user base. But in terms of technical decisions, one of the decisions that we made, so remember like 2019 is when we were starting and this is very much the peak of SaaS and every new tech company, they would think about building a product in a SaaS model. But for us we had a very interesting decision to be made, which was when we go to a company, imagine this, let's say we go to a large company and we say that why don't you actually give us all of your data and we will actually take it from your environment, put it into our SaaS cloud, and then we're going to build a search experience on top of that. And we are a startup and we are two months old and we have five engineers and we don't have a security team, and nobody was going to feel comfortable doing that. So instead we made a decision that we will actually build the technology in a single-tenant model, and you could deploy Glean inside the customer's own environment. So none of their data leaves their VPC, their sort of security boundary. And that was actually a very critical decision that was not the norm, but it actually helped us gain that trust and comfort that large enterprises needed to actually solve that problem that they had. And so that turned out to be a very, very good decision for us till today.
It turned out to work out very well because you reached $100 million of ARR at the beginning of '25 and then doubled it by the end of '25. But it's interesting because you mentioned all the SaaS companies, that usually when companies get to this $100 million ARR milestone they go to sort of like a growth wall. So I was curious, was it driven by a change in sales motion, like moving from engineering teams to the C-suite, or maybe a deeper shift in enterprise budgets for AI agents that are more inclined to do so and pay?
Our business is accelerating and actually it's accelerating faster every year. And so we are well past that $200 million number also that you mentioned. And a lot of that growth is driven by the fact that we are also in this amazing technology moment. Every enterprise today wants to embrace AI. They want to bring automation through agents to different business processes that they have. And Glean is the pioneer and the most robust context platform to make AI work inside your company. Imagine, take anything that you want to do with AI inside your company. You have to work with data and information inside your company. You have to take actions in your enterprise systems. That's what we do as humans and that's what AI agents have to do. And the fact that we did all that investment over the last seven years and built this truly horizontal, comprehensive RAG and search platform over all of your enterprise data and information, that became one of the most important building blocks in your AI architecture. So today, every enterprise, when they're looking at their AI strategy, you can think of AI strategy or AI tech stack at three different levels. At the top level, you have agents and applications that users experience, where they're seeing AI capabilities. At the bottom layer, you have the intelligence that comes from models. And in between, you have this context graph and enterprise connectivity layer which is bringing the specific data and information and the know-how of how your company works to these AI models. And so that middle layer is something that we are known to be the strongest player in. And that's why every company, as they think about the AI strategy, we have become a core part of that technical architecture that they are designing.
There's been a lot of noise about token volume as sort of like a KPI, right? Like $100,000 or $1 million per engineer or something of that sort. So how do you think about the real measure of value, and how does this specifically, let's say, impact Glean's business model, if at all?
That's actually the top question that every enterprise today has on their mind. Or actually two questions. One is just the rising cost of AI. While you can have these vanity metrics like how many tokens every individual is consuming, and that's a measure of how AI-powered you are, but actually the other problem with that metric is that it's actually real cost. The more tokens your employees use, you have to figure out where to bring that money from. And if there is no value being generated, that becomes a really hard thing for companies. So most enterprises are actually in that mode where they feel like value realization has not happened but the costs are increasing. So from the point of view of how we tackle that, number one, the Glean platform actually is designed in a way which minimizes your AI spend. If you take the basic AI tools from the cloud or ChatGPT, and if you start to connect those tools with your enterprise systems using MCP, one thing that happens is that even the simplest tasks that you need to do with AI, these systems are very brute force and they spend a lot of tokens just trying to assemble the right context to complete that task, and that actually is where all of the cost comes from. So when you bring Glean inside the company, you can actually connect Glean as a backend, as a context provider into ChatGPT and Claude and Gemini, and we can actually for any given task immediately give the right context, the right data to these AI tools so that they can then complete their work, and as a result, halve your AI token consumption.
So that's sort of one part of how people think of us as just a hard cost savings for their AI stack. But then the second thing is, we are also of course building an AI product, and the way we actually drive value is that we focus on actual use cases. We're not actually selling AI technology. We allow you as a customer to actually automate and build agents for the most expensive business processes that you have inside your company. So imagine, for example, legal teams redlining contracts, which is very expensive, sales teams filling out RFPs, customer support resolving tickets on a daily basis. As customers are filing all of these use cases, you can actually build agents in Glean to bring automation, almost full automation or 90% automation, to these different tasks. And the good thing is that each one of those tasks have an associated cost. For example, companies already measure that their cost to resolve a customer ticket is about $120, let's say. And now you bring the Glean agent to actually do the first round of triaging and resolve maybe 80% of the tickets. You can actually bring down that cost from $120 to $30, let's say. And so now you have a hard measure of ROI. You're not just talking about personal productivity or time saved. You're actually solving a specific business problem and you have a clear understanding of how much it cost you before and now how much it costs you after the Glean agent. And so that's how we help businesses compute the ROI of AI and then justify their purchase of AI.
I think the case study that you are alluding to was about Claude and about the MCP specifically. I will mention it for the audience: Glean preferred two and a half times as often as off-the-shelf MCP tools and specifically consumed 30% less tokens than Claude, which is obviously a big deal for all the enterprises. But I was wondering, when you talk with all these big enterprises like say Dell or T-Mobile, is the conversation more about accuracy, maybe the token economics, or even about stability, knowing exactly which document informed the AI reasoning at the end, or is it something else?
All three of them. So first, are agents actually doing things correctly? Are they using the right information and data? Second, are agents reliable? Do they always perform in an expected manner or do they sometimes go off the rails? And third, the cost. So all of those are important considerations for businesses, but there's a change in overall sentiment. Last year most of the discussion used to be focused on the accuracy of agents, and in the last two months most of the discussions that we hear is no longer about accuracy, because first the models have gotten better, so AI agents are now performing better than ever before. But the real pain point now immediately is the cost of AI, the token consumption. And every day, we talk to enterprise customers and they share that they had a certain budget for AI this year and they've already consumed it in the first quarter of the year itself. And some of those companies are saying their token consumption is doubling every week. So today the top-of-the-mind consideration for most enterprises is how do I keep these systems from bankrupting us. That's sort of the right now the bigger concern for businesses.
It seems like an endless loop because you use AI more, so maybe you need less employees, but then you pay more for all the tokens that you use. And I wanted to ask you about another big product of yours, of course the rollout of the agentic engine too. You introduced fast but also thinking modes, where thinking modes achieve 94% answer completeness. But why was it important to give user control over the reasoning instead of simply making the AI as powerful as it gets? Was it also about the unit economics of it?
It's a very good question. As a user, generally you would not have an understanding of what thinking level to use in a model, or in fact what model to pick for a given task. In Glean, we support all the dominant models that are out there, including Gemini, GPT, Claude, Nvidia's models, and even open-source models. So we support them all. But as an end user, imagine you are a paralegal, or you're in the finance team, or in customer support, you have no idea which model is better for what task, and you also don't know how much you want the model to think. These are not things that you can do. Some advanced users may have a good understanding based on their repetitive use of AI. They've sort of gotten this feeling now that sometimes they want the AI to think more, sometimes less. So we give all of these options to users to manually go and select what models to choose for any given task, or what thinking level to use. But about 2% of users actually choose that, and 98% of them are using the auto mode. The system chooses for them what's the right model or what's the right reasoning level. So why do we even do it? Why do we actually give this option? Because ultimately, AI by itself doesn't know anything. All the things that AI today knows are basically things it learned from humans. And even though only 2% of the people actually choose models or choose reasoning levels, we need to learn from their brain, their insights. That's the reason why we give these choices to end users, so that we can then use that data to ultimately build a better auto-selector for models or for thinking level in the future. That's the source of learning that is going to help us improve our ability to pick the right models.
It's all about the user experience at the end, indeed. And okay, so Arvind, I do have one last question before I pass the mic to Senat. So obviously Glean search engine inside organizations, but what do you think search inside organizations will look like five years from now? Specifically considering that people say, okay, dashboards are gone, but do you think there will be another iteration of search inside organizations?
Search is changing very, very rapidly. Our own product, it started with you would ask questions, and you could ask questions in natural language. Even in 2019, we were actually the world's first enterprise generative AI company, the first ones to train transformer models inside the enterprise. But our first version of the product looked and felt like traditional Google. And now we see more than 50% of the users not wanting to use that method. In fact, they want to actually do Q&A, they want to converse, they want to ask questions, and they want a precise answer to come back to them. But it's not just questions and answers. You of course do a lot more things with AI. AI is actually becoming more like co-working with you. These things are blending. So search, knowledge access, knowledge management, assistance, and co-working, it's all getting blended into this one uniform product experience where you feel like now I have somebody on the other side, a very smart individual who knows everything about my work life, who knows everything about how our company works, and can actually help me continuously with whatever questions I have. I don't think anybody knows what's going to happen in the next five years because this technology changes so fast. So I think even in the next one year a lot of things are going to happen, and I'll share one of those, which is that search will actually also change from being a reactive technology to a proactive technology. And by search I mean search and AI assistance in general. So imagine this, you have a system now which truly understands you and your work life and OKRs, things you need to get done this quarter. It knows about all the weekly tasks that are on your plate, and it'll actually pick some of them and do it ahead of time for you. Or before you actually ask for some question, it kind of already knows that you need certain pieces of information based on the projects that you're working on, and it proactively surfaces that information back to you. That is going to be one big change in search and AI systems, is they're going to shift from being reactive to being very proactive, powered by the deep knowledge of that individual, which needs you of course to build that context graph for the enterprise, which is what we specialize in.
The future looks bright, and I can't wait, not to say five years, even in the next 12 months as you mentioned. So Arvind, thank you so much. It's really been a joy, and without further ado, the floor is yours. Take us on.
Yeah, of course. You know, thank you so much for joining us today and really interesting learning about some of the themes around AI cost savings and then maximizing work per token that you were talking about. You know, it's 2026. Every resume claims some level of AI expertise. What specific signals or questions would you kind of use to identify this AI builder mindset, and how important is it for people at Glean to have this?
We have this new process where we test for AI fluency of a candidate, but in reality, we're not looking for, hey, is somebody already an expert with AI? Have they used AI a lot in their personal or work life? That's not what we're really trying to look for. Ultimately, our hiring process is the same. We always look for people who are curious, who are always trying to learn new technologies, new ways of doing things, who are creative, who like to work hard, who are good team players. The criteria is actually the same, but AI is giving us an opportunity since it is the most fundamental new technology that we've seen in the last couple of years. It gives us a place to test for that curiosity and test for who's trying to reinvent themselves. So somebody has done a lot of creative things with AI, it shows us that they are those curious types who we want in our company.
