Let's say I'm hiring for a finance role. You don't want to tell them not to use a calculator or Excel formulas or macros. The AI is the same way. Just think of it as a tool. It's a very powerful tool, but use it as much as you want. So one thing that has happened is that our practical exercises for testing have all become more complicated. Now what we ask you to build is 10 times more complex. You will not be able to succeed unless you use AI.
Welcome to the Main Branch podcast. This is your host, Cagla Kaymaz. Main Branch is a show where I talk to founders and early adopters about building, deploying, and scaling world-class AI startups. My guest today is Arvind Jain, founder and CEO of Glean.
So Arvind, you and I have known each other for a number of years now and I feel super lucky to have been an investor at Glean at my old firm, seeing you as an amazing leader and epic executor, taking the company to where it is today. I'm super excited to double-click and learn from you and learn from your journey. Thanks for joining me.
I'm really excited to be here. Thanks, Cagla.
So I suspect most people are already familiar with Glean, but it's always great to hear from you. What does Glean do and how do you help your customers today?
Glean is an enterprise AI company. First, think of us as the Google or ChatGPT, but inside your company. Imagine if ChatGPT knew everything about your company, your people, your customers, your business, your processes, and it can use all of that context and knowledge to help answer any questions that an employee has or do some work for them. We are the enterprise version of an AI assistant and a coworker. And Glean is also an AI platform that powers other AI applications within your company. All the AI applications that you have in your business that need that broader and deeper enterprise context, data that's besides the data that they have in their own applications, you use Glean as a platform to get that context to power AI within your own products.
Amazing. I definitely want to talk about the product journey and evolution. But when you first started back in 2019, I remember you telling me you were starting as an enterprise search company and given how successful you were at Google and Rubrik, VCs told you, 'Hey, I want to back you, do anything but enterprise search.' And I think you said, 'No, this is what I want to do,' and you went ahead with it. Can you talk a bit about that journey and how did you build conviction to work on something a lot of people told you not to work on?
That's a great question. The reason why I wanted to work on search inside our work life was because it's a pain that I felt personally every day. No matter what job I did, whether I was at Google or at Rubrik, it was always a big problem. Whenever you're looking for information, whenever you need a quick answer to a question that you have, you struggle. Inside enterprise, you have hundreds of thousands of different systems. A lot of knowledge has become outdated. It's obsolete. So getting answers to simple questions has always been really, really hard. And in fact, even today, in the world of AI, it still remains the same for most enterprises. So when we started Glean, we didn't have the thought that we're going to build a really amazing company that will succeed a lot. We'll build a huge business. We were starting from the mindset that here's this problem and I face it and everybody I know faces the same problem and it's got to be solved. It's going to improve people's experience. And I think there were studies that showed that one-third of all working time by knowledge workers is spent just looking for information. So we also knew that it was a big-impact problem to work on.
Now, the interesting thing is that nobody actually built a successful search product inside businesses before. There were many efforts over decades, but no successful company. And so that was sort of the non-exciting part of it for any investor that I went and talked to. When you see that nobody has succeeded for many years, you start to make assumptions based on that. One of the assumptions is that it's probably not a problem worth solving. And that was what I heard a lot as I was planning to raise funds. But my conviction came from my own experience. And that's actually important for any entrepreneur: if you have experienced a problem yourself, then you can stick with it.
I remember trying some of the earlier search products and they just didn't work right. So everyone is like, 'This experience sucks. Someone should do this, but this has been tried.' And I think there were a couple big acquisitions before Glean, even a decade ago. Then after they acquired those companies, the product basically died. It just didn't work.
Why do you think they failed to succeed and what did you think you were going to do differently that would make it work?
It's kind of interesting. Now we're in the modern AI era, so some of these things that were on our mind seven years back may sound foolish now. But the reason why all search companies in the past failed was because search is a really hard problem to solve. In the pre-AI era, one of the first things you need to build a good search product inside an enterprise is to be connected to all of your data systems. You need to know where the knowledge is, where the information is, so that you can create an index and help people find that information. Just connecting to systems was incredibly complicated in the past because in the pre-SaaS world, enterprises had custom applications they built themselves or installed in their data centers, and connecting to those applications was completely bespoke. You had to spend multiple years just trying to get hold of data. And then once you get hold of data, there's not enough context to understand what data is good, what is bad, what is high quality, what's up to date versus what has become obsolete. All of those signals were very hard to get hold of, and nobody bothered to work at such a deep level at those search companies. So the complexity of the problem was higher. But as we went through technology shifts like SaaS and cloud, they were big enablers. The fact that we can build super-scalable systems in the cloud without spending too much time on infrastructure is a big enabler because search is a very intensive technology. Similarly, the fact that businesses have standardized on common SaaS systems makes it easy for us to interconnect and get hold of data. But the biggest enabler was transformers themselves.
You're talking about like more BERT-like models?
Yeah, like BERT-like models. Those were a step-function change in terms of how much you could understand enterprise knowledge and data and do search at a more conceptual level as opposed to keyword-based search. So these were all big trends which made us feel like this was the right time to go and solve the problem.
What's really interesting about that is you were using some of those models before the mainstream caught up. And then as you were building the company and the product evolved, a lot started to happen on the more... Like, ChatGPT came out and gained a lot more traction in the mainstream. And then as you mentioned earlier, now you have this super successful assistant, you have a whole product suite. Can you talk about the product evolution and what was happening in parallel from a technological shift perspective?
Yeah. So when we started, our product looked and felt like Google, but inside your work life. You come, you ask questions, we'll surface the right information back to you. You could ask in natural language because from day one it was a transformer-based search-matching system. So we could understand people's questions in natural language and understand content semantically. That was one small difference, but otherwise the product looked and felt like Google. But as the models got better, so did our capabilities. One of the first things that happened as generative AI started was that AI could actually write and synthesize answers. So we could go one step further and instead of surfacing links, we could build a conversational AI interface where we would take all of that data. You come and ask a question, and we built the industry's first RAG, where we would use our retrieval engine to pick the right information and then make an AI model work on it to summarize and synthesize a response. That was the first evolution of the product, moving from Google to a ChatGPT-like experience inside your company. We were the first ones to build that. RAG as a term was not invented when we did it, but it became clear to enterprises after ChatGPT that you couldn't just take all the information inside your enterprise, train a model, and get a ChatGPT-like experience. People tried and nobody got it. They realized they had to combine a search and retrieval system with the power of the model like we were doing. That's when the term 'RAG' got invented. From there, the model capability has continued to evolve, and the product has changed from a place where you ask questions and get answers to where you also do your work. Glean today is very much a personal coworker for you. You can come and do a lot of things inside it: write documents, create artifacts, do data analysis. And of course, you can give Glean a lot more agency to complete long-running tasks for you. These are all incremental developments and product capabilities.
One of the main shifts is from information retrieval to where Glean works on behalf of you. It can send an email, update your CRM with latest information after a sales meeting. It starts to feel more like a coworker who can take half of your tasks and do them for you behind the scenes. That's the evolution on the product side. But an interesting thing for Glean as a company is that we had to build a lot of technology underneath to power these experiences: connecting to enterprise systems, understanding security and governance, building a deep knowledge graph of how your business works. And our customers started to say, 'Hey, you've done all this work. I've connected all my systems to Glean, and I want to build my own AI applications. I need an API from you to retrieve the right content or take actions inside enterprise systems.' So our customers pushed us to become a platform. Today, think of it as a horizontal AI platform that brings the right context that any AI application needs and the right actions capability to do work inside enterprise systems in a safe and secure way. That's what our customers are doing with Glean. They connect Glean with most of their AI applications, like Claude or Cursor, or enterprise applications like Zoom, Miro, or Sigma. All of these applications now use Glean as that context layer. For example, Zoom has a feature where if you ask me a question and I don't know the answer, Zoom AI Companion can show that answer by pulling context from Glean behind the scenes. So we've evolved into being a platform that powers almost all AI experiences inside a company.
Yeah. I remember it was a huge unlock when everyone was trying to use ChatGPT at work, and at the large enterprise, you obviously can't do that. With Glean, having the security and governance and RBAC and all these controls really help people adopt it and do what they want to do in a more controlled setting. What is next for Glean? And also, there's this ongoing concern about the labs moving up to the application layer, like Claude Cowork and the likes, and the hyperscalers who also have access to a bunch of your data. How do you think about Glean's next stage and how do you ensure you stay market leader?
Well, first I will start with the state of the AI industry. Today, we are in very, very early innings of this AI revolution. You hear this from people and sometimes you feel like, okay, somebody has drunk the Kool-Aid and is talking big. But I think the AI revolution is bigger than some of the fundamental shifts in our human existence, like the Industrial Revolution or the internet. AI is much bigger because it's fundamentally changing the composition of the world. For example, when you work, you don't have to work alone anymore. If you're a new grad joining the workforce, you still have a team of 50 people helping you from day one. You don't have to be the apprentice or the intern. You start with a lot of knowledge and skills from AI agents at your disposal. So how work happens is going to fundamentally change. We are really at the beginning. There's hardly any impact that has been made so far in the industry with AI. Right now we're in this learning phase, people are learning how to use AI. So when I think about competition or how do we survive here, with so many people coming into this space, my first reaction is that even if we all came together and became one company, we don't overlap on any products, we all build unique technology, we are still 100 times short compared to the demand and what businesses want from us. There's so much work that needs to get done, and technology companies simply don't have the capacity to deliver. I'm not talking about hardware capacity, but intellectual capacity and the ability to deliver the right products and experiences. So we're in this super early journey.
At the same time, I think companies have to pick their domain. You don't want to just repeat what somebody else is doing. At Glean, we are very particular about where we innovate and we want to innovate in areas where others are not. When it comes to building large foundational models, we don't have the resources or capability to build that, but we don't need to because we partner with these awesome model companies. We get to use all the technology they are building and embed it in our product experiences. Our own swim lane in this AI architecture is enterprise context. It's building the deepest, most fundamental, and broadest understanding of how your business works today, aggregating that human intelligence inside your enterprise, the know-how, how work happens today, and bringing it all in one place. That's what we focus on, and we see others are not focused on that. It's a combination of that plus the amazing model technology that gives enterprises a solution to modernize their business processes. Because when you have that enterprise understanding from Glean plus the models, AI is not starting from scratch. It starts at the base layer of the current human intelligence of the company.
It's interesting to hear you talk about this like a rising tide lifting all boats. There's so much work to be done and you're just getting started. I think a lot of founders would die to be in your position where you're the leader, you have a head start, you've been working on this for some time, you have this enterprise context. If you were just getting started today in 2026, how would you think about finding an area to get started where there seems to be a lot of competition from startups, hyperscalers, growth-stage companies? If you put yourself in the shoes of a founder getting started today, how would you think about it?
There is a lot of opportunity in every area. It's hard for me to pick even a single thing where I would say this problem is being addressed by a lot of other people so I don't need to work on it. I just don't think that. It doesn't matter whether you think about AI for finance, legal, HR, or engineering. We are still super, super early. As a founder, you need to remember that we are in early innings. If you say this market is crowded and I don't want to be in, it's like saying there are too many people in the world so there's no point in participating in a race. Well, you're going to win the race if you work hard for it and train for it, and not everybody else is going to do that. That's my mindset. It's not about being smart about picking the right areas. The best ideas come from your own personal experiences. You work every day, see what challenges you face, and if those are generic enough, then just go and build a company on that. Nothing is solved at the moment. What drives success is long-standing belief and conviction. As an entrepreneur, you give up because you lose confidence in your idea. You forget why you started and worry too much about competition. There are a lot of large numbers. Imagine there are four startups already in this space. Now there are five of you. But if it's a broad enough problem with a million potential customers, the probability of having to compete with one of them at a client is minuscule. It doesn't matter. What wins is whoever builds a better product and is more persistent. So my take on the AI industry is that everything is super early. You pick an area, work hard, keep at it, and you will win.
I 100% agree. As a seed investor, I am betting on these people getting started today because you're just getting started. The opportunity is huge. I'm curious, one of the things you mentioned to me a while ago was the conventional wisdom of even if you have conviction, you don't want to wait too long to put your product out there. You want to do early, iterate, get feedback. Versus your approach: you already know this is a problem, so you're going to build in stealth. You don't even need a single design partner initially because you know what the problem is and what to build. Was that the right approach for you, and how did you build conviction that this is the idea, even if you're not working with any customers, you can go and work on this for a couple of years?
It is true that we didn't care about making revenue in the first two years of our company's existence because we had a few things that were true for our business. Number one, the problem itself is understood. People struggle to find information, making it easy is going to help them. So it was an understood problem and we knew what technology needed to be built to solve it. We also knew that we were going through a phase where businesses didn't have discretionary spend and this was not a budget line item. Nobody was used to buying this. So we knew we would have to evangelize, create a market, and have patience. But you always need design partners. Even if you know what you're building, you cannot build in a vacuum. So I spent a lot of time getting feedback. I used to send about 100 messages on LinkedIn every day to people, telling them I'm working on this idea, not selling anything, but wanting to know if it makes sense. That was a key part of making sure we were on track. Those conversations help build the right product, and you're also building it for people who feel obligated to try it out once you've built it. So you absolutely have to be aggressive about connecting with people. You don't have to be super aggressive about charging them. It's more important to get people to succeed. So we didn't generate revenue, but we were definitely working with people.
Okay, that makes sense. So you mentioned just now you reached out to a ton of people, just cold outreach on LinkedIn. How did you get your first one, three, five customers, and when was the first sales hire you brought on?
The first few customers we got were the design partners. We never signed them up initially as customers. We didn't propose a deal. They were working with us for six months or a year already. We had this product deployed as a beta for them, they were giving us feedback, we were improving it. And we let them use the product for free for a long time. At some point we said, okay, now...
