AI is Making Enterprise Search Relevant, with Arvind Jain of Glean

0:05Hi, listeners. Welcome to NoPriors. This week we're speaking to Arvin Jayne, CEO and co-founder of Glean. Glean is an AI-powered enterprise search and knowledge management platform, which allows you to not only access all the different internal documents and slacks and other things that your company may have, but also allows you to enhance workplace productivity by using different applications on top of that. Prior to Glean, Arvind had a really storied career. He co-founded Rubric. He was early at Google, worked on search there, amongst other things. And so we're very excited to have him here today.

0:32Arvind, welcome to NoPriors. Thank you for having me. So I'm really excited about this. I've known you for years and Alad's known you for maybe 15 more years than that. You're an amazing, repeat successful founder with Rubrik and Glean. I want to start by just asking you about search. You've been a search guy, you know, since before it was cool for a long time when it felt like not solved, but not as dynamic. How broadly has search changed because of LLMs? I've been working on search for almost 30 years now, long, long time. The paradigm has completely shifted. I think I would say that search had been static for a long time.

1:10It was this keyword-based paradigm, like, you know, people ask questions, you find words and try to find them in documents and bring them up, you know, to the users. But LLMs have completely changed it. Like, you know, it has actually, the main thing it has done for search is that it has allowed us to really deeply understand a question that a user is asking. And similarly, it allows us to very deeply understand what a document is about. And you can actually, you know, match people's questions with, you know, the right information conceptually. And that gives us so much more power. It's not brittle anymore.

1:42And I think it's been a foundational technology to really evolve search into these new experiences that you're seeing these days, you know, where you can go far beyond just surfacing a few links, you know, to an end user to actually deeply understand their questions. and answering them, put them directly using the knowledge that you have. If I remember correctly, Glean got started in the more traditional search world. And that as these foundation models and these LLMs have come to the fore, you've really kind of shifted how you think about both the capability set that you provide and how you approach things.

2:14Could you tell us a bit more about how you started off building the systems and how that's shifted and then how you've kind of mapped new use cases against it? Because you're now effectively like this really interesting platform that can be used in all sorts of ways inside of an organization, around the corpus of information they have. but I'd just even love to hear the technology transition. Like, how did you think about that? When did it happen? I think you really lived through it in a really meaningful way. We had good timing, I would say. So, you know, we started thinking about building Glean in late 2018.

2:42I started the company early 2019. And so interesting thing is that transformers as a technology had emerged by then. Now the whole world was not talking about it, but in search teams like at Google, you know, we saw the power of embeddings and how it could fundamentally change search. And so we had that luxury to actually see this in action. So the version, one of our product actually already used transformers for semantic, you know, matching. Like, you know, we didn't have these terms. Like, nobody used to call it vector search. You know, we didn't have that. Like, these terms had not been invented yet, or generative AI for that matter.

3:17And so, like, internally, we used to call it embedding search. And it was a core technology that we started out with. So you were super early to it, actually. Yeah. And, you know, the models at the time were not as powerful as today. We started with this BERT model that Google had put in Open Domain, which was trained on all of the Internet's data and knowledge. And we would then take those models and then for every customer of ours, we'd actually build custom embeddings on their business content. And then that would sort of power the semantic part of the search. But remember, like, you know, search as a technique, there's a lot of focus on embeddings and vector search over the last few years.

3:54but that's actually only one part of building a good search system. Because if you think about an enterprise, imagine a company that has been around for a few decades. You know, there are tons and tons of information spread across many, many different systems. A lot of that information has become obsolete, you know, now because it was written like many years back. And so when you build a search product, it's not just enough to say that, hey, I want to understand somebody's question and I'm going to match it with the right information sort of semantically or conceptually matches what the user is asking.

4:29Well, you've got to solve for other problems too. You've got to actually pick information that's correct today, that is up to date, that has some authority, like somebody who's an expert on this topic has actually written that document. So you have to do all of those other things too to actually truly sort of pick the right knowledge and bring it back to people. So we started with building the product in that shape and form. It was a very different product, actually. Nobody had to actually search enterprise search as a problem before. In fact, the interesting thing that I remember is that even though I was coming off of a successful company, like with Rubrik, we had good success, I don't think people really wanted to invest in enterprise search or me, for that matter, because this problem was not exciting.

5:09It was traditionally a very bad problem, right? So there's all these search engines fast. I remember when the early Google days was sort of an enterprise search engine, I think, based in Norway. There's lots of attempts at this. A lot of attempts and no successes. Why do you think it didn't work? Because it felt like an awful market. It was like a graveyard, like, you know, of all these companies that tried to solve the problem and it didn't. Part of it was just that I think search is a hard problem. In an enterprise, like even getting access to all the data that you want to search, it was such a big problem.

5:37In the pre-SaaS world, there was no way to sort of go into those data centers, figure out where the servers were, where the storage systems were, try to connect with information in them. It was a big challenge. So SaaS actually solved that issue. So like search products, like most of them, most of the companies started in the pre-SaaS world, they failed because you could just couldn't build a turnkey product. But SaaS actually allowed you to actually build something, you know, which is my insight was that like, look, you know, the enterprise world has changed. We have these SaaS systems now and SaaS systems don't have versions.

6:09Like everybody, all customers have the same version. You know, they are open, they're interoperable. You can actually hit them with APIs and get all the content. I felt that the biggest problem was actually solved, which was that I could actually easily go and bring all the enterprise information and data in one place and build this unified search system on top. So that was actually a big unlock. So it was the rides of these connectors and APIs internally. So you're using Google Docs instead of older school systems or using Slack or using these new tools that now provide you access to the data or underlying content.

6:41You guys must remember Google Search Appliance. Yeah. The idea of like, I need to slurp your data continuously into a hardware appliance in order to actually do search is ludicrous. It was a challenge. The, you know, search as a, and by the way, the origins of Glean is, so at Rubrik, you know, we had this problem. Like, you know, we grew fast. We had a lot of information across 300 different SaaS systems and nobody could find anything in the company. And people were complaining about it in our pulse surveys. And I was, you know, I always run ID in my startups. And so there's a complaint that, you know, it came to me, like I had to solve it.

7:15So I tried to buy a search product and I realized there's nothing to buy. I mean, that's really the origins of how Gleam got started as a company. And so that was like, you know, one big issue, like, you know, the search SaaS made it easy to actually connect, you know, your enterprise data and knowledge to a search system. So that actually made it possible for us to, for the very first time, build a turnkey product. But there are a lot of other advances as well. You know, one is, you know, like, look, you know, businesses have so much information and data. One interesting, you know, fact, one of our largest customers, they have more than 1 billion documents inside their company.

7:47Now, hear this, you know, when Alar and I, you know, when we were working on search at Google, you know, in 2004, the entire internet was actually 1 billion documents. You know, there's a massive explosion of content like inside businesses. So you have to build scalable systems and you couldn't build like a system like that before in the pre-cloud era. I would spend all my time just trying to build that scalable distributed system, which, you know, we don't have to anymore because of thanks to, you know, all the great cloud technology. And then, of course, Transformers. Like, you know, that's really the big unlock, you know, that we had was that we could actually understand enterprise information more deeply and was very necessary in the enterprise compared to on the web.

8:23On the web, even if you don't have good semantic understanding, there is so much that you can learn from people's behavior because, you know, you have a billion people, you know, coming and using your product. In the enterprise, you don't have that luxury. So you have to sort of like, you know, make up for that, you know, lack of signal from users, you know, with other techniques and transform is one of them. It sounds like you feel a combination of, I'd call it like more traditional IR and search techniques and embeddings is relevant. Do you think that persists? Like where would you want bespoke infra or, you know, signals like freshness and authority or like how much do models just do in the end?

8:58Yeah, I mean, I think there's always this thought of that, like, you know, the models will have near infinite context windows and you can just give them everything and they can figure things out automatically. But I don't think, you know, like they're anywhere close to, you know, that happening. I'll give you an example. Let's say that models are mimicking human intelligence. So they're actually getting more and more capable of like, you know, how we work like humans. But as a human, like, you know, imagine like, you know, if I were to actually give you, like, let's say I give you a question and then I say that here's, you know, here's everything.

9:30Like, you know, in a completely non-organized fashion, I give you like a whole bunch of like 1 million documents and let's imagine you have, you know, the memory powers and speed, but it still just feels like, you know, a very complicated thing. Like it's very hard to make sense of information that is, for example, being given to you out of order. Like, you know, I give you one document that is something from today, something from four months back, something from three years, then something again from two days back. If I give you information in a manner where it's sort of not organized in any shape or form, then as a human, you're going to have a lot of difficulty reasoning over it.

10:05So we think about the model the same way. There is a good amount of work that you have to do and present the information to the model in some organized fashion. That's when they're going to actually do a much better job reading that information, reasoning over it and giving the answers. And sure, you can actually give them more and more over time, but still, it matters how you provide them with the right information. Now that you have this sort of corpus of information, you've basically aggregated all the internal documents of a company, which in itself is incredibly useful just for search. But you've also got down the route of enabling applications to be built on top of it in different ways.

10:40Could you talk a bit about that and what are some of the common use cases that you're seeing? So we started with this vision of building a Google in your work life, But then as models got better, developed like these reasoning and generation capabilities. So first, like, you know, it changed our product and like our new product, like Green Assistant, you know, it sort of looks and feels more like ChatGPT. So instead of like, you know, me going asking questions and seeing a bunch of links coming back to me, you know, now, of course, you converse with Green, you ask questions. And it works just like ChatGPT.

11:10You come and ask a question. It's going to actually take all of the world's knowledge. And also, additionally, it's going to take all of your internal company's data and knowledge and use that in a safe and secure manner, like knowing who you are and what information you can really use within the company to answer questions back for you. So that's sort of like the first progression in terms of our product. We evolved from being a Google to something that looks more like ChatGPT, a more powerful version of ChatGPT inside your company. As you build that, this Glean Assistant actually, you can think of it more like a personal assistant that you're actually giving to every employee in your company.

11:49It's a tool, you know, it's your sidekick, you know, it's always available to help you with whatever questions or tasks you have. It's going to use all of your company's context and data to help you with, you know, with your work. But, you know, businesses are actually a lot more interested in not in that, but in actually thinking about how they can transform their company with AI, how they can take specific business processes, you know, where they're spending a lot of money and how do they bring automation in that with AI. So we've been asked before agents became, you know, the talk of the day and everybody's a force building agents.

12:27but early last year when agents had not yet taken off, people were asking us for that, hey, we need to build more curated applications using this data platform that you have. So as an example, HR teams would come to us and say that, look, we love Green Assistant. People come in there, ask questions about benefits and PTO and vacation policy and whatnot. And it works great, but sometimes it uses content that's not authorized or blessed by us. And if somebody is coming and asking questions on people-related topics, we want Glean to only use the curated content that our people team has created. And we want it to behave in a particular way, particular tone, and all of that.

13:11So that was a request that we started to get last year that can we create more specific curated experiences, function by function for different use cases. So we started to build that. And we were not calling them agents. We were calling them apps. Now, of course, like, you know, the people think of, you know, them as more as agents because it's no longer just, you know, asking questions and getting answers. But you want, you know, these specific functional experiences to actually replace the business process, which also involves doing some work for, you know, not just answering questions, but actually doing some work in those systems.

13:44Arvind, when you talked about, you know, access to the right data with the right authority and also like it really begs the question of like access control. in a platform like Lean. When you have all this unstructured data, this seems much more complicated. What is your overall stance or how you think this is going to work in the future? Yeah. Well, so look, enterprise information, in some sense, it's governed and it's protected. Most of the knowledge, I should say 90 % of the knowledge inside the company is private in some shape or form within your company. You'll have a document that maybe is private to you or you share with a few other people.

14:23So that's the nature of enterprise knowledge. That's the fundamental sort of way it works. And you can't actually build, for example, a model inside your enterprise and dump all of your internal companies' data and knowledge into it and then make that model available to everybody in the company. Because if you do that, you're leaking information. Like inside a company, you're letting somebody in the engineering team see sensitive stuff, which probably only HR teams should be able to see in the example. So any AI experiences that you build inside the company, it has to think about security and governance and permissions at a fundamental level.

15:01And that's what we do in Glean. So when we connect with all these different systems inside our enterprise, if we index a particular document from Google Drive or a conversation from Slack, we also keep track of who are the users who can actually access that information. And this is fundamental. Like any access to data that's going to happen through our platform is going to actually match. Like, you know, the users have to be signed in and we will actually only let them use information that they have permissions for. And this is important as a problem to solve. Like, you know, unless if you have infrastructure like that, you cannot hold out AI safely inside your device.

15:36I learn a lot from people who work on search, especially like search with any sort of scale because you get all sorts of weird user behavior. And so related to your idea of, you know, us with our personal assistant team, what are some behaviors you see from end users in terms of how they're using Glean or AI in general that you think we should just do more of? Right. Like I, you know, I'm always very surprised when I learn from Google people about just like the behaviors around navigational search and how many are one word queries or what the popular queries are and those sorts of patterns. And so I'm sure you see like Glean and AI super users.

16:10One of the biggest surprises for me, I always felt that we're building such an intuitive product. There's no UI, there's one box and you ask a question, you put in a search and what's the big deal? Why do you have to learn how to use this? And we realized that as we added more and more of these natural language capabilities and the ability for you to actually ask a really long question, like a paragraph-long set of instructions that are given to us. And we realized that people won't do it. I think everybody has been trained over the last 20 years to actually type in one or two keywords. Google has taught us on what search can do.

16:55So with search, we never had a problem. We launched a product, we made it high usage. Nobody was confused how to use the product. with assistant, people didn't know what to do with it. Some people with more curiosity and they will ask all kinds of questions that we couldn't actually answer. For example, somebody says, hey, what should I do in my life? So I think, but anyways, coming back to this, that was one of the key learnings is that AI is actually very unintuitive. For most people, you have to actually really expose to them these capabilities in a sort of a incremental fashion. You know, like some things, you know, which sort of are more meaningful to their day-to-day work.

17:35For example, if I'm an engineer, like, you know, prompt the user sometimes that like, look, you can actually learn about a new piece of technology. Like I can actually give, you know, create a two-page tutorial for you right now. And you sort of have to understand like, you know, what people's, you know, people, like, you know, what their core work is. And then you have to actually give them these, you know, sort of prompts, like prompts for them to sort of start experimenting and get excited about like, you know, trying something out with AI. One thing, in fact, which I would also add here is, like a lot of time, you know, with AI, businesses are excited.

18:07Like, you know, they have a lot of dollars to spend on AI. And, but they're all like also asking for ROI that, well, like, you know, I'm going to make all this, you know, investment. What are the returns? Like, what are the efficiency gains, you know, that I'm going to be getting? Or what are the top line, you know, improvements that I can make to my business? There's a lot of focus on that. And one thing that often gets overlooked is education. Because, you know, the world is changing. Imagine like, you know, three years from now, you wake up, you're the CEO of a large enterprise. What do you want to see in your workforce?

18:38You actually want to see people like who are trained and are AI first. Like, you know, they're experts. They know how to leverage the strengths of AI because this is a difficult technology. Like, you know, it's not perfect. It's not easy. It makes mistakes. It hallucinates. But yet it's powerful. and if you become an expert, you can get a lot done with it. That has to be like the objective today is like, you know, with like as leaders think about AI, how do you sort of build people tools that sort of motivate and motivate them to bring AI in their day-to-day work? You had an amazing career between being early at Google, starting Rubric, now starting Glean and running it.

19:19What was unexpected about doing Glean? Because you'd gotten to so much scale. We've done such amazing things in the context of Rubrik. What was hard or unexpected or just very different about Glean that you didn't anticipate? From a product side, one of the most interesting things for me was how hard was it to actually roll the product out to our customers? We had a very different journey in Rubrik compared to Glean. like in rubric were an established market like there were budgets several dollars and you had to actually replace an old technology with a new technology um here we were in a market where uh we had no uh budgets there was no concept of buying a search product in the enterprise and and everybody thought that yeah like this is an important problem but i'd like you know it's not a line item in my business priorities you know it's it's a it's a vitamin it's a painkiller people are living without it well yeah that's just true i mean you live without something you don't have like you know that's by definition you know you know you know true so we so we had a lot of challenge like we had to do a lot of evangelism to actually get the right like you know um folks like you know who wanted to be um the innovators like for them to actually make that bold call and and actually buy a product that's you know they're not used to buying so that's that's sort of first part of it like you know you have to create the market for this which actually was difficult and second which is actually very interesting one is you know we we thought part of actually working well like you know it was doing good search or they're you know letting people find things but then we started to hear from businesses that oh i'm scared of good search and what i don't want a good search product in my company because i have all these governance gaps i have like you know sensitive information all over the place and you know now people are discovering these things and we launched like you know for example you know people found like salaries of other people you know those like in one of our customers somebody found a sensitive m &a dog that was or something that was not yet happened.

21:15And you start, so people like you actually were very, very scared of actually having good search. So we had to actually, like that was an interesting challenge. We did some good work. We're doing it safely and securely, but you don't have good governance. And now we can't sell because the product is so good. It seems like LLM should be able to help with that, right? In terms of classifying documents and surfacing, hey, this one may be sensitive. Do you want to secure it, et cetera? Yeah, so in fact, that's exactly right. Like, no, so we actually were forced to build that. You're forced to actually go above and beyond respecting permissions in individual systems to knowing who you are, what you're asking.

21:50Like, you should have the right to even ask the question or, like, you know, when the information comes back, like, does it feel safe enough for us to show it to you? So we actually, in fact, you know, in that sense, you know, we actually ended up becoming a security product. Like a lot of companies actually buy us to fix governance in their sort of, you know, data and systems and become AI ready. like AI ready for the clean search product, the clean assistant, but also for all the other AI products that you can buy inside the enterprise. So that was actually a very interesting journey. But then for personally, you know, for me, you know, at Rubrik, you know, I didn't actually, I wasn't the CEO at R &D as one of the founders of the company.

22:28And here I had to actually learn how to become a CEO. And I don't think I've learned it yet. And like, you know, that's a constant, you know, challenge and like, you know, learning center that I go through because fundamentally, like, you know, I'm still an engineer. Everything I do, like, you know, like, you know, that's the mindset that I have. So, so growing, growing, you know, out of that into like, you know, being able to run a large business, you know, that's a, that's a personal transformation that I'm going through. One thing that I think is striking is that from a go-to-market perspective, you all have really focused on big enterprises, right?

22:57And you mentioned some of these enterprise data needs. A lot of people always just want to do PLG and you've really done sort of the top-down sale. It's been incredibly successful. You've done it twice now, right? Because Rubrik was actually that as well. Could you talk a little bit more about when it makes sense to do big direct enterprise deals versus a PLG mission and how you think about that as you build businesses? Because I think it's very differentiated and most people just can't pull that off. So I'm curious about how you think about when to do it and then how to do it. Just to be candid, clean.

23:24When we started, I mean, my dream was to do PLG. I'm an engineer and I wanted the community to have engineers and then product should sell itself, you know, on the web. Who doesn't want that? It was something that was a desire for us. But the problem is with our product, it is by definition a company-wide product. It's not like we cannot offer the product to one individual inside a company. Even one person, their search needs require us to actually search over all the entire company's information for them. So it's expensive. You have to actually index all of your company's data and knowledge. And so we never had that concept that we could make it available to one or two or 10 people.

24:03inside the companies. We're sort of forced just structurally to actually build in that fashion where it is, you know, like enterprise. It is like, you know, we roll the product our company-wide, you know, every employee. That's what makes it cost-effective. But like, you know, coming back to your question, the standard approach I think that companies prefer now is that like they think of PLG as basically lead gen as a funnel. You sort of nurture and expand using, you know, enterprise change motion. So the right recipe for me, like, you know, if I had a choice, I would actually start both the motions simultaneously.