Why 90% of Enterprise AI Will Run on Open Source

We're in very very early stages of this The amount of things, you know, that we're going to be doing with AI, it's minuscule compared to, you know, where it's going to be 5 years from now, 10 years from now. Open at some point had stated directly publicly that you were one of their main competitors, but at the same time, you're able to leverage and partner with them. So, I don't actually worry about competition. I've never worried about it.

dream started in early 2019 which makes us the world's first enterprise generative AI company. I particularly remember your vision was to build like Google for enterprises. That vision changed onethird of all employee time is spent just like trying to find things. So we wanted to solve for that.

When you work with you know thousand plus customers you kind of see like you know what usage patterns are today. A lot of application companies with their data advantage are also coming into the modern businesses. Cursor has composer etc. Is that also something?

Yeah, in fact today if all of us came together, if all the model companies, if all the application layer companies like us and became one company and we start to deliver products and services without competition to you know all the enterprises, we will still fall short. We still won't meet even that% of the demand. Hey Evan, thank you for joining us. Uh you're one of the technology luminaries of our time.

You started many massive companies. Rubric went public and now began clean. I know you and Kavan have a history together. He worked with you at Gle at Rudek and then this led to this partnership that we have with Glean and Composio.

Can you tell me like how you met Kavan, how that started? Yeah, I mean as you said first of all like thanks for having me. Excited to be here. Uh Karan and I have known each other like since our rubric days uh where he was one of our early star engineers.

So that's that's how we got to know each other and when we started this company and he was building you know technology which kind of made sense for us to like you know uh to partner on. So it was like it was a clear choice like you know cuz we believe in him. Yeah. Thanks.

I mean obviously I think I kind of rubric uh was the first company that I joined. Uh thanks a ton for having me there. I think that's where I essentially learned how to build systems, how to build the engineering team. Uh so kind of yeah and thanks a ton for believing in ush for the first time and kind of like that has shaped our product.

uh as well. Can you briefly explain like what clean does and its history as a company? I know you guys started as an enterprise search company sort of vector search vag systems. You moved to something much broader in a recent podcast.

You described it as something like a supererset of chat GPT and cloud code. So yeah like can you talk me through like what the product is today and like how you sort of got here what the history looks like. Yeah. So, Glean started in early 2019, which makes us the world's first enterprise generative AI company.

Well, I mean, just to be clear, the term generative AI did not exist at the time. Uh, but we were the first ones to to play with transformers and language models and bring these technologies to the enterprises. And we were doing it for uh for our first product that we had built uh which was search for employees within their own company. Um I don't know how many people know this but the origins of transformers and language models is initially these things were built at Google um with the goal with with the single purpose of making Google search smarter more intelligent.

uh so when we started um glean and we were also building a search product for businesses like I said it made all the sense for us to use this technology so that's how we became the first ones to bring transformers to the enterprises um the reason we wanted to solve that problem was um our own experience at rubric and in fact actually my experiences at Google before that you know where I used to work um it's always been incredibly hard to actually find the information that you need to get answers to questions that you have that will that allow you that allows you to then go make progress on the things that you're working on. Um it's deeply frustrating. It used to be uh at rubric it was our lowest scoring uh uh question in our pulse surveys. People complaining that you know it was so hard to find information that they needed to do their work.

So that's how we got started. Now we've evolved from that. So in addition to actually of course you know having this product that looks like Google or Chat GPD inside your company one place where people come in um and ask questions or give us some work uh to do and what Glean does is it uses um both like you know the best models that are out there um that can actually helps you know solve this task or answer this question and these models come from companies like OpenAI, Enthropic Group uh you know, Google and all the open source models. Uh but more importantly, what Glean does is it's connected to all of your company's internal data and systems.

It knows where every single piece of knowledge or data is inside your company. It knows who has permissions to use that information internally. Um it's also um understands deeply like you know what content, what data is fresh, up to date, relevant, high quality. So we sort of had that you know that context of your business your data your knowledge and how work happens inside the company and we're able to use that context and that knowledge along with all these great AI models and technology to then start to work for you and the way you work with glean is you ask questions you give us tasks you can think of us as a supererset of Jad GPD claude uh co-worki like we will actually solve that task using all of that company's context and in fact you know picking the right models the best models for the each of the subtasks you know that we will have to execute to complete that task.

So that's sort of one product like very simple um use us just like how you used uh but underlying that is this platform uh this platform that's connected to all of your enterprise systems where we're able to read information where able to take actions and where we actually partner with you um the we are able to understand the security and governance of all of that data and we're able to build this deep context and knowledge graph of our enterprise and all of that platform is also just like how it's available to our own product that we deliver to our customers that platform is also available to our customers to directly integrate into their other AI systems. So many of our customers for example connect their cloud code or cloud co-work or cursor with the clean MCP uh gateway that brings that intelligence and context of your company to those tools and make them better. So so that's that's how you should think about glean an enterprise AI platform uh and a coworker. So um I remember I think even in rubric days I particularly remember your vision was to build like Google for enterprises.

Yeah. And I I as from what I understand you started glean with also that like vision. How has that like given like during that time language models became like came along the way etc. How has that vision changed or has it remained?

It has expanded so much. So like we were all I mean I think the most exciting thing for me personally was number one to build a product for myself like you know and I was that that was actually the driving factor and for everybody who I know and I knew that everybody struggled with finding information. So we were actually quite happy first just actually solving that use case which is that every employee whenever you need some information we're going to make it super fast for you. Onethird of all employee time is spent just like finding trying to find things.

So we wanted to solve for that. You know the AI of the day was kind of like you know capable of doing that. Uh it was capable of helping us understand data knowledge content. It was it was capable of us helping understand what questions people are asking and do a good matching between them.

But AI AI in that time was not capable of writing on its own or doing deep reasoning or thinking. Um but it was still a step change like you know when we started glean we used transformers and our search was already u this you know what later on like two or three years later the industry started to call it semantic search uh or vector search uh we built them like well before these terms were coined by the industry we used to call it embedding search internally you know that was the name that we chose for this concept that we had built um the and and remember that sometimes you know people think of Google as one type of experience and chat GPD as other um where Google is you know that I ask a question and you surface you know 10 links to me chat GBD is I ask a question and you give me an answer but that dichotomy is actually not true Google for decades have actually also tried to when you come and ask questions try to give an answer and it didn't need generative AI models for a long time to do that so we used to do the same when we started clean often times when people came and asked questions we would surface an answer back to that user but that answer was extracted ed from your corpus not generated by AI and then as AI got better we you know obviously were able to now start to use that generative capabilities to in fact answer questions for people. So over time like you know like our our sort of goal and mission remains the same which is we want to make um people uh do extraordinary work like we want to untap you know help them untap you know their potential to do extraordinary work that's our mission statement and and for that like you know graduating from helping them find things to answering their questions to actually now doing some work for them to actually becoming that holistic platform at an enterprise level you know that allows the company to actually operate with a new level of intelligence. We feel like you know it's just a evolution and expansion of our uh mission which you know this great technology has allowed us to actually uh enable.

I guess it's a question about like a a directory choice, right? you guys started going deep on what you called embedding search within the industry later called semantic search or vector search and then like later we sort of see this like both in terms of what you're describing as the the ability for agents to do work but also you're seeing agentic search sort of calling the tools and as an agent in a sandbox or a loop generally seems to be much more successful and you're able to do much more with it or it sort of democratizes this like vector search product in some way. How did you guys think about that inside glean and how do you think about that transition because you guys had a very successful business in this you transitioned to this other thing also extremely successfully. Well, I mean I think internally for us like sometimes we get a lot of credit for pivoting the company from A to B to C and internally we think about that like I don't think we're getting more credit than we deserve.

I think like we were simply um taking advantage of the technology that was being built around us and it's been like you know in some sense very natural progression of starting with search as the first problem that we solve for people to answering questions for them to actually doing work for them because it's all like you know it's all correlated with the advancements in the AI models and technology I guess there's always sort of this like you know this trade-off of curve that you're making when you're building products about like focusing on what's working, focusing on the new thing. I think that's where the credit is sort of coming or the question about like was there like an innovator's dilemma or sort of a push and pull within the company like hey this thing is working super well and we have to go down this new bet this is the new I think that was obvious that was obvious for us like you know like the world and I think there's there's there's a good parallel in the consumer world like and we didn't have to actually wait for that was al obvious to us too like you know if I have a question I don't want to actually go read a long document to get an answer for that like I do prefer you know a chat GPD like experience so and it was clear to said we also have to deliver it like you know you know yes jpd was initially launched only for consumers and but like you know in an enterprise of course is the same people who use that product in their in their personal lives so so I don't think they had a choice we had a you know question or um you know in in that sense like you know we were basically we had to do these you know um uh like advance our product and it actually made all the sense and people were excited and I mean that's how our R&D team works like you know they are always looking at like you know what are the new things you know we could be doing uh within the company also we have a uh this concept you know that we learned from Google uh this concept of 7030 where 70% of our team is focused on short and medium-term deliverables like you know thinking about the current product road map how to evolve it and then we have another 30% who are you know working on moonshot bets uh trying to think about new things you know that we could be building so that's that's been the um you know the kind of journey you know for us um so I I I know for this sort of agentic search and like actions transition and maybe it's not a pivot natural road map I know we played a huge part as like composure and helping in that we'd love to talk about like where we were helpful where we sort of came into the picture how that partnership came about. Yeah. Yeah.

So look like the and part of it is also like you know users who you serve. So as as we um as we started to actually launch these agentic capabilities as people could do more and more advanced analysis um deep research with lean uh the national question from all of our customers was that well I mean I think yes it's great to bring information to people it's great to actually give them these tools to analyze data uh to get them to write some content for you but ultimately like you know if you know like everybody started to see the potential of that you know now AI was ready to actually do some work like you know do like complete the cycle like not just do the first part of that like you know of any given task which is research and data collection and analysis but now you can actually do complete the work like you know if you're going to save that work in some of your systems. So that required us to have like start to think about actions. We actually built um actions also like you know probably like roughly the same time that you started to build it.

Um but for us like you know our model at you know at Glean has always been we want to actually work on things that others are not and we we are very much a partner focused company and so if we see um great innovation that is happening elsewhere and if we have the ability to actually embrace that innovation then we'll always make that choice. And so when you think about you um it's very much like you know like the the partnership that we have with you is kind of similar to the partnership we have with OpenAI or Google or Enthropic you know there's great tech you know that they've built and they make it available as a platform so when they do we want to actually leverage it as opposed to sort of having to reinvent that. Uh and the same uh motivation we had you know was with you. You were deeply focused on that one you know one specific area of actions.

Um and we felt that you know you would go fast you would you would actually go deep um you know you remember this like you know actions as you it's been a long journey and while like you know it's easy to build a quick um uh sort of system you know that can actually make a rest API call and you know and make an action like take take you know execute a rest API call but like doing it the right way with the right parameters and and sort of you know uh taking a task which requires you to make like four different API calls like you know ultimately start to run into lots and lots of complications and for a long time like you know LMS were not good at it and and so we we we knew that like you know for a for a company that's focused on making that happen they'll do a good job and and we would like to partner with them so I I guess yeah that sort of leads us uh thank you for you trusting us betting on us so early leads us into a next question that I was uh been thinking a lot about you sort of me mentioned partnering with open AI anthropic these people, how do you think about competition uh in the space, right? Because like when I see Glen's product, clearly you guys are doing exceptionally well. Uh the product is growing really well. Like I've heard speculative revenue numbers that I don't necessarily want you to comment on, but like you guys are growing really fast, doing really well, but very much in the product road map of something like an anthropic or open the way they stated.

I think open at some point had stated directly publicly that you were one of their main competitors but at the same time you're able to leverage and partner with them and if you look at your like contemporaries in coding like a cursor cognition are doing exceptionally well the they have all maybe even more fierce competition. So how do you as an entrepreneur think about navigating this and how would you suggest like we navigate this in terms of thinking about building products? I think we we first you know need to have this awareness that we're in very very early stages um of this AI revolution that we are in the amount of things you know that we're going to be doing with AI right now is minimal I mean it's it's a lot I mean I guess in some way some some sense you know AI has made made a lot of progress over the last few years but it's actually it's minuscule compared to you know where it's going to be 5 years from now 10 years from So, so there's a very very big opportunity for all companies that are leveraging AI to build valuable products for customers. And and I actually find it kind of ridiculous to to sort of think in the in in a way that hey like one or you know a model company or another model company will basically make it unnecessary for all other companies to even exist.

Um in fact today if if all of us came together, if all the model companies, if all the application layer companies like us, um platform companies like yours, if all of us came together and became one company and we start to deliver products and services without competition to all you know all the enterprises, we will still fall short like we still won't meet even 10% of the demand that enterprises have like what they want to do with AI. That's the first realization that every startup needs to actually have like you know that don't don't focus too much on competition and what other players are going to do. Um the your biggest competition or your biggest challenge is going to be just you yourself. Um and can you execute?

Can you build a high quality product? Can you actually you know talk about it? Can you take it to the market? Uh can you earn the trust of your customers?

Can you do well for them? You know those are the real challenges not whether somebody else is also doing what you're doing. um because the it's not as if you know there are only 10 companies you know that we're all selling the products to like you know the the the customer set is actually almost infinite so so that's sort of that's my take so I don't actually worry about uh competition I' I've never worried about it a lot of people talk about like us being directly like you mentioned like an open has mentioned it or um and and mark the market does you know recognize us as the pioneers you know in enterprise AI and search And whenever another large company and this is quite often like you know uh when when they come and you know build a product like glean now we get the recognition that you know they're building um a product like ours. So, so that is there like you know there there a lot of companies you know we you know that do they're doing things that we are um and the only like real reasonable response as an entrepreneur that you know that I could have is well I mean we will be focused on our customers we'll be focused on our product and we're just going to do better and we're going to do it you know with our eyes wide open and if there are other players in the market and they want to compete um well I mean like I think first let's see that you know um whether uh we can uh try to partner with them and that's what we do with you know and others that they they build great technology which they as a platform company make available to companies like us.

So, so that's sort of like you know our strategy like you know by and by and by kind of like you know always like you you can't actually of course partner with everybody you you cannot like if you also you also cannot say that if somebody else is building something that I won't build it but in general like you know it makes sense to have that strategy where you try to do things you know which are unique you know uni uniquely suited to your strengths. So for us you know that is you know search that is u building these deep knowledge crafts context crafts you know building an understanding of how your business works and we still like you know that particular area is something that we are the best in uh we focus on that other people are not focused on that and everything you know surrounding that is where we will have a partner for strategy if we can get something to the market we'll use it and that allows us to compete so there is kind of uh a thing where a lot application companies with their data advantage and other things are also coming into uh the model business as well to certain extent like cursor has composer etc. Is that also uh something that yeah like I think for glean too that's an important one I think before even talking about companies like us for enterprises um we talk to a lot of them um every day one of the biggest um um you know sort of concerns that they have today is over overdependence on model providers and you know this independence that you need um as an enterprise like you want to make sure that model is a technology that you can use uh but the context of how your business works your proprietary business processes your sort of unique differentiators that all of those all of those things you know you remain in full control of yourself there's this concept of that you know over time AI is going to do more and more work in the enterprise um but how's AI getting better you know those learnings are compounding uh and who owns those learnings and enterprises need to be in full control of that um so that so even at a c at a individual customer level you know there's a lot of focus on how to actually not fully depend on uh model providers uh for for applied AI companies like ours Um the it's the same thing like you know like we cannot have a huge dependency on any one provider for our technology stack. Uh open source models are actually fantastic.

So you know glean being multimodel like you know we actually uh work with all the different models that are out there including open source. I guess this sort of transitions very nicely into like what like if you had to say like what percentage of tokens on glean are on like a frontier uh American model versus some of these like Chinese open source models. Yeah. Well, today's dominantly like you know on on the on the American models um the but it is like the inflection point actually just recently happened.

Was that like a Kim K3 or a GLM52? Yeah, I would say GM52 was I mean Kim K3 came after that, right? But so GM5 52 was really when it became super clear that now you can actually fully rely on open source models for over 90% of all enterprise uh enterprise AI tasks. I guess I have two sort of um tangential questions.

there seems to be this like sort of cyclic every uh you know 8 to 12 months where open source sort of catches up to the frontier and then frontier pulls ahead and there a spread right like if you take back to deepsee i1 or even further back to mistrol 8x7b is like oh we're coming up to the frontier the frontier of delta is small and frontier pushes much further ahead and I guess a sort of tangential question to that is like how do you like as both a buyer of these language models and talking to enterprises buying and like buying these models How do you think about like you know like as an enterprise or as a purchaser of of tokens? Yeah. Can I sign a one-year contract with Anthropic if the thing is changing so quickly? So there sort of two questions which is like right now we're in this moment where like the open open source has caught up to the frontier.

Will the frontier again just move ahead? Yeah. Yeah. Yeah.

Got it. So so see first on the open source it's not like we are not in the same moment as we were 6 months back or 12 months back. I think this is the first time and I think like I say from for example our own engineering team like you know we've been always uh evaluating open source models and our team's decision was always that they were not ready like you know and they were of course comparing with the frontier at the time and the the the conclusion was that they were not ready and we never bothered to use them. uh that only changed for the first time now.

Um and so there's and and that's you know like a lot of other companies are going through that same thing that they're now feeling that yes like you know they can do a lot of work with open source. So this is so there there is a change um and and you will see in fact like you know my my own belief is that in the next 12 to 18 months we'll see majority of inferencing shift to open source uh in enterprises. Um and and of course you'll have uh US-based models in open source. I mean I think some of it is you cannot even I think it's not an option like you know we have to have that uh and companies are working hard at that.

Uh now on the second question of in this sort of really dynamic um landscape like like who do you like you know which model do you use you know who do you sign contracts with that's that's you're absolutely right that that's incredibly hard problem um and first think about individual enterprises and not I'm not talking about a um a a midsize company I'm talking about large even a large enterprise you know you know that do have budgets like you know they could potentially spend a billion dollars on AI today even for those companies it's extremely hard for them to actually figure out how much to commit to each each of these models and and put aside open source even within the frontier models from the from these top three or four companies um you know the race is so dynamic you don't you never know who's who's winning and so it's incredibly hard to actually like make that commitment And you you do need you know players uh who can sort of absorb that risk and make it easier for you. So and I think clean comes in you know um clean is a good solution for enterprises that way where we have our AI gateway and model card and we tell our customers that you can just work with us. um you sign one commit with glean and and and with that commit you can actually use as much of you know open AI or claude or gemini and you know and you can sort of flexibly allocate your spend there uh and you can also use open source so so we make the job easy for them now of course the problem shifts to us like you know as a startup now we have to figure out like you know how much you know uh how much commitments you know we make but I think for us you know it works it's easier because you know when you work with you know thousand plus customers you kind of see like, you know, what usage patterns are today. And and in a in a growing market when where you where you know that you're going to be spending double of what you're spending or like five times of what you're spending today, like it's easy to sort of bake in like you know some commits with each one of them.

Uh where the total commit for us also is probably like you know only half or a third of what we actually going to spend and that allows us to have like you know enough flexibility in our system as well. So many interesting points there but the first thing I wanted to sort of come back to is like we're talking a lot about like what like how enterprises are thinking to allocate budgets and like use models what's actually working and not working in the usage of enterprise like AI inside enterprises right like you had this tweet where you talked about you sort of cited the all-in thing where like Chamath is talking about how 45% token spend is going up every month or every two months and there's like 5% extra productivity there's all these questions about is the ROI there the not there. Where do you think like these models are quite diffused especially like I'm I'm curious in general but especially outside coding where do you think the the diffusion is weak and why that diffusion is weak? The number one issue right now with every enterprise is cost.

Um everybody rolled AI in a pretty aggressive manner to all of their workforce. People are doing things with AI and AI is such a broad technology you can do do it like use it to do all kinds of things. um it's becoming very hard for businesses to measure like actual return on investment on AI. most of them have no idea uh what's happening and you know they they didn't care for the last two years because it was it is obvious of course that you know every company should be investing in AI um first like you know just for the purposes of even modernizing your workforce making sure that they are ready for that future you know where AI is going to be more and more dominant so it kind of makes all the sense but it has reached like the expense has reached the level where like now they don't have the money to actually support this unless if they can actually prove do it with bottom line or topline improvements.

So that's the state of the industry. Um 45% 5% like all of that I don't know like you know they're all made up numbers. The the reality is that yes like it's like people enterprise don't have visibility into it and and there two ways to solve for that. One is that you to bring down the cost of this uh tech and I think open source models are going to actually help us a lot with that.

In fact like lean's mission like right now when we have conversations with customers like that's the first thing that we tell them that like look you know we understand your costs are so super high with AI we are the solution you know for you as an enterprise AI platform that's going to minimize your costs. Yeah, I think when you're saying like cost is the number one sort of concern that's implicitly saying like the like the technology is not useful enough that this cost is like it's you see it's useful everybody thinks it's useful but not enough but but hard to measure whether it's enough or not is not like not the complaint from people is it's hard to measure because a lot of it is very subjective like people are and I think what and part of it is just like you know connecting the dots like For example, think about your customer support team. They're using AI and they are actually likely resolving more cases every day than before or they're actually resolving those cases much faster than before and and actually you do have metrics for those like you know like customer service service teams actually track these as to their top two metrics and a lot of these companies have been able to see um a clear improvement. For example, one of one of the largest telos in the US um they use Glean to uh to run all of their customer care function.

Uh every time new tickets come, cases come like clean is you know actually helps those agents resolve those cases faster and and I'm talking about like a start from last year well before like now now we can do so much more even last year we had already improved their uh case um the time to resolve a case by 48%. So they knew it and that's why they're willing to pay for it and it's you know it's you know generates you know tens of hundreds of millions of dollars of savings uh for them. So when you when you use AI um to solve problems at a departmental level that's when you can start to measure u but a lot of companies haven't reached that state you know there's still the first thing of course what you do is you know just give AI to everybody in the company and let them use it and the the other one you know where coding of course is where in some ways you can claim you know a lot of productivity gains because of course you're writing many more lines of code um your number of commits are are going up you know on a daily basis number of GRS is all probably is also going up. Uh but even there like sometimes people feel like okay well like did we just shift the bottleneck from one place to somewhere else like are you actually shipping the product faster or not?

That is a question that is you know sometimes hard for companies to resolve. I guess like one way you can think about this is like uh like percentage of tokens relative to like headcount and then like how headcount moves there, right? Because like like at the end of the day it's like if I I don't believe in that by the way. That's I actually just don't like those two things to be in the same sentence.

Okay, count and token spend or budgets. I think for for AI vendors or technology providers to say that it's a trade-off between labor cost and how much AI you can have. I think it's the most ridiculous thing you know that like I've never heard of that in my entire like you know over three decades career of in technology like tech spend you know is is a very very small fraction of you know how your business runs. I guess the question is like you know like Salesforce reported they spent $300 million on tokens they ran out and for engineering and there was like 3% of like engineering salaries and like if you are believing the sort of thing that like anthropic or openi is selling about the future like the expectation is this 3% maybe becomes 20% maybe becomes 150%.

So it's it's not that this is a trade-off. This is like it comes from a opex budget and before software had almost no marginal net cost. It's like relatively cheap to keep running versus like now it has this extreme marginal net cost. I I think well that's just the that's just what we are where it is today.

And so your expectation is it'll again become something like a no marginal net cost. I I actually think that the cost of doing work is going to significantly reduce. there it was never zero like you know there's always any type of work that you do you always burning CPU cycles to do that work but it is basically so low compared to the utility you know that you know software would provide to you and that's why you would like you actually would say that it was like there's no marginal cost right and I think like AI should be in my opinion the same way like why is it like it is software it is like it is like you know uh running in friends you know on a machine learning model. So what makes it you know dramatically expensive is just because how it works today.

So I think the long-term long-term like you know my I guess like sometimes you know it's a like nobody knows answers there's kind of more hope um than prediction um that AI technology is also going to become much cheaper and you can see like you know there's one we know one lever of how how that's going to happen is by you know AI models becoming more task specific once you specialize a model to do a certain You already know today like you know and we are achieving it you know through distillation through you know post- training we're already seeing that the same task like you know that that you like today opus is going to consume a lot of like you know cycles GPU cycles to actually compute you know if you can't you know if you create you know a very small model that does only that one thing very well it can achieve that same level of performance but like you know at a two or three orders of magnitude less cost. So, so that's sort of we know that that is one way like this tech is going to become cheaper. Uh there hopefully going to be like you know a lot of more other uh things like that. So I think when enterprises are deciding about their AI budget, I think to certain extent they are still kind of thinking about how much per person they are willing to spend right is that like given background agents are also becoming a thing where it's not exactly onetoone mapping to a person in the company is that thought process changing or does it need to change while thinking about how much you need to spend in well I actually like the this this um this process of establishing budgets um on a per person basis and you can have different budgets by different departments.

I I like that concept because ultimately businesses you know also like it's very hard to change them overnight. Think about Walmart. Walmart cannot basically just immediately completely change you know their working model, their budget planning process you know like how you know these things are super complex and so you do have to like you do have to think in that mode um and these new kind of things like for example background agents like the right way to probably think about that is that okay well you know I'm going to keep a discretionary budget pool and and those are the ones that I can assign to them but but I do like you know they already have very established model of how to actually think about uh their overall expenditure and a lot of that actually gets associated you know to you know like you know it's in proportion to the payroll and so it you know so so that model does make sense you know for majority of AI spend and then for some you can actually have this discretionary budget I guess like understanding like it's very interesting because yes there is some marginal cost of running a CPU cycle it's just so small you think about it as as rounding to zero so I guess like If I perceive the world model correctly, it's something like you are not expecting that uh token spend relative to headcount becomes like 20%, 50%, it is like very very small but we use way more tokens. Is that sort of the correct model?

Like it it's actually like we will spend way less than we're spending today but we will spend way more like order maybe multiple order. I think we will all do like lot more work with AI. Um I think we hopefully will also stop talking about tokens like we were never talking about like you know for a long time we've never talked about the number of CPU cycles or like you know that a task takes like who who like nobody nobody cares about that like you know and and so I think hopefully you know this over fixation on tokens also is going to go away uh in the next few years but but your point is correct that we are all going to be doing more not less you know with AI and still going to actually um cost us less than what it costs us today. I guess then there was a question about like is the like public market capex allocation like wrong because they're expecting that it like there's the revenue curves like you know anthropic might hit 100 billion in by the end of the year it's expecting to go up like like this this model is like amazing it sounds like amazing right you know it's like oh I will do more and it's like yeah like why should I think about a token or care about a token well well I think investors are smarter than me so like I don't I don't want to actually really comment you know whether they got it wrong or right and you know that's is a that's a very complex question and I think but I think let's let's talk about something that that I that I can comment on and that is that the today as we were talking before the value of AI the business value that it generates you talk to most enterprises they will say that it's lagging the investment by by a very significant margin and and so I I do think that you know there are two narratives in the industry.

one of them which was just all in on AI and bullish and that you know like AI is going to be half of you know your um of your sort of you know opex and and people are going to be other half of your opex you know that's and I and then you know and I think that narrative to me is problematic like I think you know it's it's kind of like the you don't deserve to talk about all of that until you first get businesses to realize that value and agree and I cannot talk about it. So I guess that's actually a very interesting thread, right? We've been talking about the cost side of this return on investment curve a lot. I guess like where do you find like that the value is not being as successful cuz like of course in code it's like it's able to write a lot maybe the bottleneck has moved to review or something else but in sort of like other tasks in like knowledge work tasks.

Do you find like where are the bottlenecks today in the model just being more useful because because like most of the time when you talk to one of these models it's like on one very specific domain they're likely smarter than you already but it's like in some way they have jagged and they have some bottlenecks. Where do you find those bottlenecks for like your own usage, Glean's usage, but also like your customer's usage? Yeah. Well, first first let's talk about like business value.

Um, and I mean like we are able to with Glean, we focus a lot on that. Uh, we focus on departmental agents, uh, business processes that we can automate with AI. And as we go and work with our customers, we always talk to them about that look, you know, you're going to do this engagement. um you will be using glean, you'll be building all these agents, but don't do all of that work before you even know the metrics to measure um um that success like how would you know that you know that you know value was added.

So first actually agree on how you measure uh efficiency and productivity for that particular task. So we'll go for example take legal team like you know if you work with them um like one of the metrics for them would be like how many contracts you know can one one contract lawyer handle uh on on a daily basis for support we already talked about how many cases you can do uh for um for a sales rep um trying to actually do outbound prospecting you you basically measure how many meetings you know they can actually generate um on a on on a daily basis. So, so when you start to like go deep, you know, for many of these functions, you do have uh productivity metrics and then you you make that establish the baseline, then you actually build agents um and then you actually see improvements to that baseline and and that way now you have a very very strict way of measuring value. So that's the way to actually ultimately uh for enterprises to justify that investment in AI.

Unfortunately, it just didn't happen like that like you know AI was just there was there was this panic like you know every CEO every boardroom said that you got to invest in AI uh show me what what's happening and in fact like you know we were we were not talking about value like you know just four months back we were talking about token maxing you we're talking about that the success criteria for enterprises you know as as uh as pushed by the tech industry was that well like you know as long as your people are burning tokens you are in good shape right so so that's sort of like you know we are. So it's it's a matter of uh just getting that you know going uh having a more established like more formal AI programs at enterprises you know that will basically uh close the gap between measuring value but I guess the question I'm sort of asking is is more like where were they very good at generating value quickly and where are they like still very weak like what are like where are they hobbled where are they not as useful as they could be as their like intelligence suggests they are which functions of the with functions well I mean no that's what I was saying that the people people are seeing value across engineering support um sales um quite a bit uh also in marketing so they seem like my point to you was that the with the right approach you are seeing success everywhere like in all functions like we're seeing value being generated in all of them of course you're like you know for many companies the uh R&D or sales tend to be the the biggest department so the most value realization potential actually share happens in those two uh or in customer care. So like I would I would like if you ask me this question one year back I would say that like you know the real impact is only being felt by engineering and customer support but today like that's not the case. today you can actually generate value across all the different departments and the stumbles are actually not so much um that hey for this particular function AI is not there yet actually AI is there for all knowledge work now um it's more that did you like follow the right approach did you actually get the right tools um um that actually is a more more of a determinant of success versus not and like okay like I'd love to speak to that like like we talked a little bit about like okay if you want to capture value you have measure it and think about it methodically but like what are those like tools or things that you need to do that where you are able to sort of make the models more valuable for specific functions.

Yeah. Like what are you guys doing that's sort of enabling that? Um well I think you need first of all for AI to work in the enterprise to add value for your different business processes you have to you have to actually be able to go deep into understanding you know how your business works. You have to be you have to be able to go and connect into all of your enterprise systems and you guys come into play there a lot because without your platform like it's there's a lot of work that a business has to do to actually go build that agent that can you know complete that work take actions in your enterprise systems.

Of course you need the basic technology in place which includes uh models which includes uh enterprise um uh like search and retrieval technologies. it requires you know the actions platform um like these are the core piece of infrastructure that you that you have to first put in place um that and this is now going to actually allow you to uh generate generate value um the one thing that we know which I feel today often times you know people build these agents um and they use MCP as basically the primary method to to gather context to build those agents and and then they fail because you know these agents are not performing at the um that their humans were um and the reason for that is simple because these agents kind of are like you know your employee on day one yes you pointed them to some documentation but they don't really have that context and you know that immersion that has happened like you know to a person who's been there doing that work for 2 years so we're increasingly seeing this demand for investing in that deep context graph understanding your subject matter experts understanding how those experts today get things done you know which is not the way of doing those things is not documented properly. So, so that has sort of become a big bottleneck and that's where we come in, you know, and and that, you know, that kind of, you know, closes the gap between having agents that work like your day one employee versus your tenure employee. I think that makes sense and I think I'll wrap it so we don't go too long.

Thank you for spending time with us. Yeah. Yeah. Thank you.

Thanks for having me.