Fireside Chat with Arvind Jain & Noa Topaz

No, I'm excited to be here. And hopefully you can understand what I'm saying because I have lost my voice. I've been talking too much. Not working enough, I think. Just talking.

Yeah.

Well yeah that's a good question. So Google did try to build Glean for many many years. In fact we had a product at Google. Well, I used to work at Google before. Before I started Rubrik and then Glean and actually before the ads business took off, the way Google used to make money was via enterprise search. We had this yellow box. We would sell it to enterprises. But it never really worked. It was really bad because once google.com started to make money, nobody in Google wanted to work on enterprise search. It was boring and the company didn't really care so much about it. So that was sort of one reason.

No. So well no Google was a search engine that came out of Stanford but the first business you know where it started to make some money was by selling that same search service to businesses. And so we did have that business. It was actually there for about 16 years. It never really scaled. I think at the peak it was maybe around $100 million which at Google obviously nobody cared about. And so but there were many reasons why it didn't work. One of them of course was focus. Google was not interested in building that product but also I think all that time like from 1999 till 2015 I would say it was also not the right time to build that product. It was just too hard to make search work inside the enterprise. It was easier to do it on the web.

Yeah. So the challenges with search when we started or I mean like why search never worked in the past was that enterprise systems are actually quite complex. So first like even if you want to build a product like a search product inside a business you have to go and do a lot of work just trying to connect with enterprise systems and in the pre-SaaS world they were very bespoke in nature. So for every business you had to actually go and spend two years just trying to even figure out like you know what their systems are what their where their knowledge was. So it had to be like a services business as opposed to a product company. SaaS changed that because with SaaS like systems became standard. All enterprises sort of like look similar in some way and you have standard APIs to actually connect with all these systems. And then the second unlock was that enterprises typically don't have a lot of signals like Google learns from you know the intelligence of humans. So today like you know if you go to Google and you type in a few words like two words you may misspell it but somehow it still reads your mind and the reason why it's so smart is because it has actually learned from similar misspellings that other people have made in the past and see them correct see all those people correct those spellings later on. So there's a lot of learning that you get to have when you are serving in a product that actually do to more than a few billion people. In enterprises you don't get that. So you needed some other signals and one of the core capabilities that came to us in 2019 when we were starting Glean was transformers and language models that actually really allowed us to fundamentally understand enterprise knowledge you know at a more conceptual level which would allow us to do a much smarter sort of matching of people's questions with enterprise knowledge. So that was so there were some of these you know nice technologies that were sort of coming our way and plus like you know I think well that was not the reason why we started the company like we just started it because I was so frustrated like not being able to find things ourselves and at Rubrik you know where we grew very fast that was the number one complaint that our employees always had. So initially I was just trying to actually bring a product inside Rubrik to help our people and that's when I was shocked to see that nobody was actually building a search product for businesses and I got excited about building it.

It was simple. It was just building a Google for you in your work life one place where you go ask questions and we'll surface the right information back to you. And you know that information could be in one of like you know hundred or thousand different systems that you have in the enterprise. So our job was to actually find the needle in the haystack like find the right information that you need to do your work from one of those systems. So basically Glean goes and it connects with all of enterprise systems understands what what data what knowledge is where and it's going to now start to sort of surface the right information based on whatever questions you have. That's how we got started.

Yeah that's an interesting journey for us. We started as a search company and when we did that like you know usually like actually like you're asking this question that hey why did nobody start this company or why did Google not have success part of it was because nobody cared about a search product in the business world. Nobody really thought it was important. You know, I would ask people that, hey, like, you know, what if we made, you know, gave this product to you? It's going to save so much time, you know, for your people. And most of the reaction that I would get in 2019 or 2020 was, well, yeah, seems useful, but like it's like a vitamin like not really necessary and I don't have the budget for it. So, like, so we had to go and actually really go and create that category. It took us a few years to do that. But the evolution of the product has been natural like I was as I was telling you you know we're the world's first enterprise generative AI company. We're the first ones to use transformers and language models and bring them to enterprises train models that understood you know their own corpus their own data. And over time you know so the models like in 2019 they're already pretty good in terms of understanding information conceptually. So we built a semantic search engine from the get-go in 2019 before the terms vector search those kind of things even got like you know these things were not coined at the time. So we used to call it embeddings based search like that's that's was a more technical term for it.

No we never no I think we actually did I mean if that's the question you want to ask me we'll talk about all the mistakes we made on marketing. But I guess the first mistake was calling Glean the workplace assistant when we launched and nobody in the world actually really understood what that meant. So it was a big flop. This was like you know I think we called ourselves the assistant like you know well ahead of when you know before this sort of ChatGPT revolution. So.

Yeah. So at the time like you know we actually remarket like you know we actually go back and call start calling out search even though our vision always was to be the assistant for people to help them with their work life but the evolution of the company so we started with search models kept getting better at some point they developed the ability to actually generate and write answers and synthesize information so we could go one step further in our product you know our product used to look like Google but now you know when somebody comes and asks a question we could go one step beyond like you know just surfacing the right information. We could actually have an AI model go and read that information and give an answer back to you. So there you go like you know we launched a ChatGPT like experience a conversational AI bot roughly in like early 2023 right about the time when ChatGPT did the same for consumer and you know like the model capabilities have sort of gotten better and better over time. So today like you can think of Glean as you know say think of it as a superset of what ChatGPT or Claude or Gemini any one of these products are like you know we don't build these frontier models ourselves so you can come to Glean you can ask questions you can give us tasks and we'll use all of your world's information all the world information but also all the information inside your company all the context of how your business works and combine that with the power of the best models from you know like Claude and Gemini and GPT and all that and then sort of answer your questions or complete your tasks. So that evolution has happened as models got better. So today like you know we can go well beyond answering questions to actually doing real work for you and be your agentic co-worker.

Yeah. Well, I kind of like you know really dislike the term agents because you know now whenever I go and ask you know anybody in our company like you know like can we do this faster and somebody will say yeah I'm going to build an agent for it and they never actually build one. And so like everything like everything like you know whether you're building an application whether you're building a prompt like you know agent has become a very generic term for like basically like any body of work that you're going to do in the enterprise. But people are seeing a lot of success now like we see I mean of course like codegen and customer support like they've been the largest application of AI where a lot of automation is actually now already in place at many businesses. There's so many companies that don't even actually allow developers to actually write code anymore like in like I mean I know of many of our customers where the code can only be written by AI now and developers can only review that code. So there's like you know that full huge impact you know that AI is making in certain functions. Last year was mostly software code generation and customer support. This year actually it has been or in the last 12 months it has actually transformed that to like you know basically every job every function in any business in the world. You can actually now drive significant automations. I'll give you some examples like legal teams they don't need to review these 100 page contracts manually anymore. They can use AI to actually do the first version of redlining and save like weeks and weeks of time. Your sales team they don't have to respond to RFPs that you get from large enterprises which are like 500 questions long and it used to take 3 weeks of your solution architect's time to fill them out. You can fill them out you know now easily with AI. And so you're seeing like you know that automation actually now taking hold like you know across all different business functions you know. Yeah. So agents are for real for sure.

Well I mean so Glean also evolved into an agent platform. So as we were sort of we started with search then we had the AI assistant. And to build like search and AI assistant we had to do a lot of hard work underneath. We had to actually go and connect Glean with all of your enterprise systems. We have to understand security and governance of all of that data and knowledge. We have to build a retrieval engine, a knowledge graph like where we go and understand like how your business really works, who are your people, what work they do. So all of that infrastructure that we built was exactly what like any agent needs to do its work. I mean if you think about an agent, what does it do? It does some piece of work for that. It needs to actually read some information inside your business. It's going to use AI to actually do some work use some judgment of you know the models and then ultimately like you know when the work is done it's going to actually save that back into your enterprise systems so sort of I can think of an agent as a three-step process read information and context from your enterprise use AI to sort of now do the work and then finally save that work in your enterprise systems. If you think about that journey the first and the third step are the ones that are the most clunky the most complicated because reading information from enterprise systems is complex. Writing it back into those systems is also like fairly involved process. You have to do it with the right security and governance. But we done all of that work already since we connected with all the enterprise systems. So it was very natural for us to make our platform be the one that actually allows you to build agents not just in our own agent builder but any other agent builder that you have in the enterprise.

That's a good question. Well I was saying you know we made a lot of mistakes you know along the way. And for like you know for entrepreneurs like I would say that especially the ones who have technical background I think often times you know we like as engineers we can be very conservative and we will often like underinvest in all functions outside of engineering like whether it's marketing or sales and other places you know we did the same like even though we're the world's first enterprise generative AI company we actually didn't really invest so much in marketing you know originally we conservative. We didn't actually build like you know large sales teams you know from the get-go even though we had the product that was great that was ready like people were already loving it but we still sort of like focused a little bit more on like how to sort of grow responsibly as opposed to doing land grab. So today I would say that you know in the tech industry the right strategy for building a company is that you know if you have product market fit then you have to go really really fast because capital is abundant it's available and if you have the lead then you got to capitalize on that and establish yourself and you know at your customers as fast as you can.

Well, for us it was when people were actually sending us thank you notes and love letters back from.

We sold to well I mean we started selling to mid-market enterprise companies like you know with maybe like 500 to 1,000 employees first and then you know as time went by like you know we actually scaled up to you know now we of course sell a lot to companies with over 100,000 employees. But it's a good question like you know when we started the company we were actually not even thinking about business you know we were actually really thinking of you know us as a fun like you know we're building a product and was just you know an enjoyable fun journey to build that product and the first 20 companies that we signed up with we actually didn't bother to even ask you know charge them money like so we followed like you know all the wrong practices like if you go and talk to you know expert entrepreneurs they would say that like you know make sure that people pay you don't bring it like because that's the real test of whether they care about your product or not and we didn't do that you know we said that look you know as long as the company is a lot of people like you know the person who we're talking to doesn't really matter like whether they want to pay or not is not relevant we want to see if you know end users are going to use the product or not so we're very willing to actually go and give the product out for free so for us like you know the sign of whether we have a product market fit was not that somebody was willing to pay us. But when we saw that people were actually really really finding the product useful, they were using it, you know, 10 times a day and they were actually sending very positive comments back to us.

I mean like I guess like you know in a tech company you always have to like initially like you have to get builders like you know folks who have the mindset to go and like first you have to find people who are kind of like you know crazy because like you know this startup is a risky journey and you know they're sort of jumping into it knowing very well that you know they may not get paid or they're not going to be as financially successful as they could be you know at a large you know AI company or whatever and so we have to sort of find people who are really truly motivated by the mission you know that your company is about and so that's what we did like you know we had when we started the company actually was actually quite easy for us because when I told people that you know I'm going to build a Google for you in your work life I had no trouble attracting engineers because they actually had all faced that problem every single day you know in their current jobs. So the initial few people who we got were you know my ex-colleagues. Many of them were from Google and in fact you know some of them were folks who had actually said no to me at my previous startup. The company I built before Glean was Rubrik and but nobody wanted to work on that product which was like enterprise data security and protection like you know something boring. So nobody was excited about that but this you know people really loved you know because they could relate to it themselves. So but now who would I hire now? I guess the same thing like you know if you know people who are great you know like really good engineers you know you have working relationship with them that's always the best people to hire.

Well the number one mistake is actually not believing in yourself. Like you see like.

Yeah. Is that why is that?

Yeah. Well, I mean I think the reason I'm saying that is like you all you often hear from people that nine out of 10 startups fail, right? And I always correct them and I said no, it's more like 99 out of 100 because the other 90 are killed by the creator. You know, creator herself or himself like you know there's always doubts like you know you come with an idea and there are like thousand reasons why that idea will actually not work but that's not the point. The point is to actually like you know just keep at the mission and like you know you'll always it's like you know navigating like you know you're driving a car and there's not a paved road but ultimately you know you're going to figure out the journey and reach the destination and that's what a startup is about. So that's the number one mistake you know that people make. So don't do that like you know if you have you're on an idea there's a reason why you're on it and try to sort of you know keep going back to that original reason you know why you started the company whenever you actually had doubts.

Um yeah.

Oh yeah I mean like you know yeah like actually good point when we started the company this was like the worst thing that you know like you know I talked to investors remember I was already a founder of a company you know that like you know worth you know multiple billions of dollars. So in some sense like I was a prototypical you know specimen you know that you know an investor likes to you know invest in 100%.