Latest / TeqTalk with Jas / Why Enterprise AI Struggles to Reach Production | Inside CoreWeave’s Playbook
Transcript
- speaker-1: Thank you so much Prashant and ⁓ welcome to a tech talk. I have been really looking forward to this conversation today. speaker-0: Thank you, Jess. Thanks for inviting me. This is a privilege. speaker-1: We start with something which I am seeing in the ecosystem right now. So we have reached a very, very interesting point with the enterprise AIs where adoption right now is not a question, right? Everybody has implemented AI agents into the application. And also from one of the reports from Mackenzie, it says that eighty eight percent of the organizations now ⁓ report that they are using AI, at least in one of the business functions. But the issue is most of them are not able to scale it to production. So that is a problem. And same kind of you know pattern I can see in my customer conversations as well. The decision may start with the model or you know agent, but very quickly the question comes to do I have the right context? Will the agent be able to really get the output? And how we can trust that agent with systems. Agent is going to touch and how do I actually govern, you know, the action? And most importantly, did any of these actually imp going to improve my business outcomes? And it's very interesting to talk to you because Core Weav just is scaling at an extraordinary scale. I have the number as well for that. So CoreVeeve reported 2.575 billion in Q2 2026 revenue with roughly $104 billion in revenue backlog at the End of the quarter and that is amazing, right? Which makes me a point here is that the discussion which we are going to do is not going to be the surface level discussion. We will love to go ahead and deep dive into how you are implementing. Before we do that, I would love to understand and know a little bit about your journey. ⁓ we know that you have spent more than a decade at Equinex Tech. So what that organization has taught you that you are still carrying it forward here at Core Viv as your primary skill. speaker-0: Sure, hey, thanks, first of all, inviting me and this is very interesting topic, close to my heart. And definitely we will have a lot more deeper conversation. So very excited about, you know, next 30, 40 minutes. ⁓ Equinix, my responsibility was enterprise systems and data and AI enablement. And that was 10 plus years, as you said. at Equinix and that was a journey when the AI started, machine learning started and then ⁓ at some point in time, LLM started and the ecosystem of AI was changing rapidly. So ⁓ we had a lot of successful ⁓ AI use cases powered by machine learning and data. But a couple of things I quickly realized at my previous company in order to get the scale and the impact from AI has to be democratized. We talked about data democratization, data quality with the guardrails, but even the AI tool itself needs to be democratized to help people to get their problems solved and take it to the production level. Right. And since like it's one and a half years at Covive, that was the one learning that helped shape how we are driving AI inside Corvive. And Corvive as a company, mean, all the top AI models, builders are our customers. So definitely we need to embrace AI. We need to adopt AI. And what you said is the revenue growth and the backlog that we have. when you have so much of ⁓ demand from your customers as an IT organization, where my role is again, enterprise systems and AI enablement. an IT organization, my job is to unblock my business as quickly as possible. So I really cannot wait and ⁓ try to find third party systems and take long implementation time. And that's where the AI acceleration happened and we can go more deeper, but that was the driver. Like how do you ⁓ cope up with this type of growth and speaker-1: scale. Very interesting. You have mentioned that the AI should be democratized, right? And ⁓ could you give couple of best practices which you have been doing here in a way that is really helping in democratizing AI? speaker-0: So why AI need to be democratized? if there is a core team with a small number of people just enabling AI app, that team doesn't scale. That's the reason it needs to be democratized. And there are infinite possibilities available with the AI. And we need to make sure that every employee has access to AI and explore those possibilities and implement their ideas. I can tell you the process that we took to democratize AI as fast as possible. The way we democratized AI at Core Weave is first of all, we opened all the standard AI tools to all employees. is like Cursor, Codex, Cloud, ⁓ know, Chat, GPT, Gemini, ⁓ Glean. And the reason for that is depending on where you sit in the organization, you have an affinity to certain tools. There are certain tools popular within the engineering community, development community, research community, or deep research community, right? And content generation community, automation community and all. So definitely we did not make it too restrictive, like just go and use one tool. But at the same time, we are not saying like bring your own tool. We also have a good process to evaluate and bring the tools. But then these tools are effective only when there is a quality data available. So we democratize and I call it like MCP, MCP is now a verb now. We MCP all our structured data and enterprise content or unstructured data. So what that means is any data coming from our third party systems like ERPCRM ⁓ plus any data that exists in Jira or Gmail. ⁓ Google Drive, Slack, or Confluence pages, all the content, or some document management system. Everything is classified for security guardrails and then available through MCP servers. all our AI tools are getting the context very consistently. Everybody has access to the data to build their applications. But one thing we did uniquely to ⁓ you know, accelerate adoption is we created our own application layer app layer where what you do is like you use the AI tool, build your app or whatever chat bot or workflow, anything that you want to solve for. And then you select certain templates like here, depending on what type of app it is. You check that template and then deploy it to our internal app. Why that is important. because you said like, people can develop the app and, they just run it on their local host or a laptop, but it doesn't go to the production or even if it goes to the production grade, production, people don't know whether it's a production grade or not. But that whole process of get taking your app from your idea to the production grade is also AI. And that's where we saw a lot of enthusiasm from people because it is easy to build. And then I am also like, taking it to the production. when you go to the production, you are also advertising your app through some internal catalog. And then we also ⁓ ship the Grafana observability along with that. I know exactly whatever apps I ship, who is using it, who requested for it, and then number of users and adoption. It feels very good when you see live from day zero who is interested in your app. So that's the approach we took. speaker-1: I think it's very interesting and couple of points ⁓ which I want to again highlight what you mentioned that democratizing AI is so important because in that way you are not letting any specific department to control the permission to launch it, right? It becomes so hard. And in a way you are educating the other people as well that how to use it. And you have built internal end to end process where you have a use case, you create it and put it in with the app and deployment and all. I mean you are covering all the aspects of the development process. You're doing observability, you're doing traceability. I think this is giving end to end information and education to the other people as well. And that is real democratization, I believe. speaker-0: Yep, agree. speaker-1: Yeah. But what is the fundamental difference you have seen about transformation inside a rapidly scaling AI infrastructure company, which is Coreviv, right? You have been talking to different other organizations. Where do you think there is a big difference? speaker-0: Yeah, so I've been talking to many data leaders and AI leaders across the globe. So one difference I can see is the AI strategy itself. So a lot of folks get stuck into AI strategy for a long time to enable AI. They recruit like different types of roles that takes time and goes through multiple rounds of approvals that takes time. Yeah. And that's where I see Core Weave has a different approach. So we move very, very fast. But at the same time, we are making sure that the data is protected, the guardrails are there, and certain roles are doing the job within their swimplanes, right? Mm-hmm. With a compliance policy and all. Even then, our approach is like, OK, make AI enabled with the publicly available data, right? Without any friction. Just just available, let people play with it. And a lot of companies worry too much about the tokens and token costs, which is a valid concern. But what we are doing is like we are observing all the behavior. And then if somebody is experimenting, you know, for a small duration, building some app, learning something. And if that app is not going to, you know, become a popular app, then obviously there will be won't be any token usage. We don't worry too much about those usages, but there are certain apps or AI automation that really picks up basically because it really solving some problem and it is becoming popular within the community. Then we definitely approach that team or whoever has built it and we help. And this approach is more for optimizing the app and branding more ⁓ so that it becomes like adaptable. So that is the approach we are taking. ⁓ speaker-1: Quick question on this app as you are highlighting it. This is something which has been given to the company employees, right? Do you have any use case or there is a specific business use case you guys provide and then the apps are created? How does that go? speaker-0: So right now our focus is to solve our own problems. Okay. So there's at least within the IT strategy, instead of building, buying third party app and implementing, can we solve this problem using, you know, AI powered app? So that's the approach. So our employees, we call them business engineers. So these are like business folks with engineering mindset. They know their pain points better than anyone else. They need not to write the requirements for someone to explain their pain points, but they can play with the app ecosystem and You try to solve their own problem. The advantage there is when you Solve your problems with this approach. You are very very surgical about it Yeah, so you're not going to solve for the entire world yourself solve for yourself speaker-1: Pretty interesting point before you go ahead that you're giving it to your business engineers, which means the business users, right? Mm-hmm. So unlike other organizations where business is the one to take care of the business requirements, right? And ⁓ engineering team is different, IT department is different and they actually build the product. So now When business users or business engineers are building, they understand that whether this capability is really possible or I'm just in some other world dreaming of something which cannot be achieved right now. Because there is a lot of discrepancies I have seen with the within the organizations where business users and the IT department thought process doesn't sink in. So make me understand if I'm right that you are referring to business users as business engineers or it's a part of your IT department. speaker-0: calling business users as a business engineers. Okay. Great. example, we have ⁓ like in our data center world, there's a community of people we call data center technicians, which is a business user. Yeah. They have created app to solve their own problem. We have people from inventory management, supply chain, shipping, know, ⁓ that shipping is their shipping optimization is their day job. ⁓ Planning is their day job. or inventory management is their day job, but they have created app for receiving and processing ⁓ transactions within warehouse. have people from revenue management team. have people from legal team, HR team. They do their day to day job, plus they are creating app to solve their own problem. So these are like end users. speaker-1: Mm. Pretty interesting if every organization should do that, at least in not this scale, but in little scale where, you know, everybody gets hands on and try to have their own problems solved. speaker-0: Yeah, so I have a comment there. If you make it very easy for anyone to adopt AI and solve their problem, then there will be lot of curiosity and enthusiasm to go forward. But in a lot of organizations, like if there is too much friction to even get the access to the tool, speaker-1: Yeah, then you wouldn't have i that kind of motivation, right, to start and do and go ahead. And what is what is the motivation which you are providing to the business users to start using the app? speaker-0: So we have hackathons that we conduct regularly. And then we support people in the hackathon. Like, you have an idea. Come here. I will give you like a CTO. We will give you architect sitting next to you. And then, yeah, as a team, you build the app. that's another one motivation. OK, I can do something using the I have support of building. speaker-1: Support of building, yes. speaker-0: And then, definitely successful app gets awards. But besides that, ⁓ there is not much friction to access the data which can be publicly available. And moment, like, of course, like if you are accessing sensitive data of the company or classified data, there's a huge friction. You have to go through that process. So we have that card rail. But I would say ⁓ a reward is ⁓ I'm not spending any annoying time just to get the approvals. can have very less friction to get started. that probably ⁓ helped adoption. speaker-1: Yeah, and a quick turnaround basically. You you can start it and you have the support and you can have a quick turnaround and solve the problem. You also mentioned ⁓ Prashant before that you guys have enabled lot of tools instead of standardizing on one specific tool. Tell me more about it. Why do you think this kind of ⁓ mindset is really important? speaker-0: I believe it is important to stay competitive in this market. ⁓ in order to have a competitive advantage and distinguish yourself, the speed is important. The speed with the precision is important. And that having democratized the platform. So our AI strategy is like a platform strategy. You democratize data and tools through platform and let people build. ⁓ Then you achieve a speed and precision both at the same time. Speed because it is easy to build app and deploy. Precision because you're only working on your own pain points. You're not buying a software which is made for a thousand different customers and you're force fitting your processes, but you are like creating specifically for your needs. speaker-1: Very interesting, ⁓ Prashant. Would you also like to discuss the use cases which you are implementing in Core Weave? speaker-0: Yeah, sure. I will try to stay as generic as possible. But as I said, like every business function has adopted AI. I'm seeing a lot of adoption on the finance side, ⁓ especially whenever they need to read long documents and get value out of it. Like it can be a leasing contract, legal contract that you have to instrument into the system and process. So people are using. automation using AI, I'm seeing a lot of complex revenue calculations are basically got enabled through AI with a human in the loop, course, basically. And then tons of reconciliations. mean, any like a Neo-Cloud business has lot of complexities in their transactions. And we land up in a lot of reconciliations, a lot of financial reconciliations that are happening there. I'm seeing huge adoption in the supply chain side of the warehouse automation ⁓ receiving. I'm seeing people are deploying their own AI app where you can take a picture of bill of lading and automate the receiving or the ERP. I'm seeing AI adoption for planning tools or even functions like MRP. We are building our in-house capabilities using AI powered tools. So yeah, across the board, wherever there's a confidence that clean data is available and the not exactly idea, but you know exactly what you are solving for. Problem is well defined. I'm seeing people coming up with the app pretty quickly and getting it updated. speaker-1: Very interesting. And how ⁓ you have mentioned that you're building a lot of things in house. What strategies are you seeing around in your organization ⁓ with respect to build versus buy? Are you open to buy or it's mostly build within the organization? speaker-0: Great questions. ⁓ With lot of adoption that we have seen in past few quarters, definitely our inclination is towards build. ⁓ But definitely there are some areas where buy totally makes sense. Specifically, the app has to deliver for certain compliance, right? And if it is inbuilt into the app, it totally makes sense. Sometimes there are areas where the data model itself is very complex. All the third party tools, have champion data model for all the years, then why not to leverage it? So there are compelling ⁓ use cases ⁓ where ⁓ third party products make sense. But even then our strategy would be make sure that the data is MCP or available. And then it is also available to build some applications out of that. So every app not only need to bring some features, but every app brings some valuable data to the company or creates valuable data. And that can be also leverage for towards building app. speaker-1: What exciting things are you planning to do in coming next six months or so? speaker-0: There are many right now because AI has created infinite possibilities. ⁓ But what we are focused on right now, like how we enable adoption faster. are investing in some kind of a, like enriching our current AI enablement program, making sure that it is advertised properly, making sure that some kind of a training material is constantly available. So we have some kind of webinars or ⁓ I call it bootcamps, different bootcamps available. And then all the builders need some level of constant help. So we are building some office hours concept where anybody can come and ask questions. So I'm seeing a lot of traction in that area. ⁓ But definitely besides this, know, enablement process. would be excited to take more complex problems where typically we would have thought like, hey, no, we should go and buy third party ⁓ software. ⁓ But definitely would love to explore in-house first, even if it is too complex for ⁓ traditional IT shop to handle. speaker-1: Do you have any example on what kind of use cases you really think could be built? speaker-0: When our data centers become operational, mean, we deal with tons of telemetry data and machine data. Right. We have to collect data from all the devices and compute devices, network devices, as well as the infrastructure devices like generators, chillers, right? And these data coming from multiple sources have some kind of a core relationship. What we are trying to solve is using this data and AI, can I predict problems before my customer finds it? that when we are running this massive infrastructure, we detect the problem based on any pattern recognition anomalies in data collected from all these devices. And we ⁓ identify a problem and dispatch right person at the right time. to solve the right equipment, right, like right machine problem. So I'm very excited about exploring our own platform. There are tons of platform trying to solve this problem, ⁓ but the scale at which we are growing, ⁓ we definitely have unique set of challenges that we will definitely look at ⁓ solving through our own platform. speaker-1: With that said, I remember you also mentioned that you guys are making app and you're tracking the tokens but you're not very concerned about it. So you treat AI consumption kind of as a behavioral signal rather than something to restrict with the users, right? So what do you find when you go back and ⁓ talk to people, those who are heavy users? speaker-0: So what we find is the app, which has heavy users, that the developer or app builder really knows the problem and it's very relevant to the company. That's why a lot of people got attracted and tried to, you know, play with the app or even started using it in their day to day function. So that's the uniqueness. Like if somebody is solving a right problem and it is getting right traction, it is either simplifying the process or it is eliminating the process. Right. or standardizing something. So those are the, I would say, unique traits in the app where there's a huge adoption. speaker-1: Have you faced something like the app is gaining a lot of traction from the other users, but there is a governing body. So where does this governance body sit into the picture in your current organization? speaker-0: ⁓ So our governance is embedded in the platform. ⁓ So we do not have like a councils or bodies. I mean, we do have some councils definitely, ⁓ but we don't have like too many committees or too many bodies to, you know, evaluate the governance because based on your role, you have only limited access to certain set of data to even start building the app. And it is all secure because I said, right, like you're not just creating in local host, you have to deploy it in ⁓ in production through our own app platform process. ⁓ So it has, when the app is live, right, and available in our internal, you can say marketplace, it has gone through a lot of guardrails in most the automated way. And finally, there will be approvals ⁓ before the code goes into the production. So I would say a lot of guardrails are already taken care of. Now, if the app is really getting traction and now it's not a guardrail, it's more about what help you need. And also we found opportunities where we can optimize the token usage because if it's not written right way, then we offer help and optimize it. And also sometimes we find like, okay, ⁓ the person ⁓ had a great idea but struggling with some features and then our core team, they can help that app. speaker-1: That's interesting that the whole guardrail and the governance is throughout the process and not at the end to stop it or to provide the governance, which becomes a kind of hindrance to you know, ⁓ in the overall process. So I think that's interesting. And ⁓ tell me about your leadership space that what is driving you to innovate to this level? speaker-0: Couple of factors driving me, the kind of environment we are all in, as I said, the speed and precision is important and the traditional product building processes are not good enough now ⁓ in this age of AI. And that's where we need to shift our mindset and see like, what is going to work and accelerate the impact basically. So that's one factor. Second thing is, ⁓ ⁓ I like to solve problem. So I love to find problem. I want to make sure that I work on a problems. Those are really relevant and those are really big. Those can be solved. So in a traditional way of software building or product building, solving a complex problem takes long time, even identify with a clear clarity of thought that takes long time. What I'm saying ⁓ is like if you outsource that function to person who really owns that problem, then the real problem surface very quickly. is helping us to even find problems, right? And then helping us to prioritize the right problem. So that excites me because now I don't need too many whiteboard workshops or a lot of consultant just to tell me what is going on. ⁓ the people on the ground are telling like, this is what I need and this is how I want to solve it. ⁓ now, the last thing that excites me is ⁓ if you enable that through some kind of a platform, then basically you are creating a flywheel of finding them, prioritizing problem and solving it, right? So yeah, those three things. speaker-1: The other thing which I think ⁓ is definitely a part of this ⁓ innovation is that the processes are limited. So like how you mentioned before the democratization of AI tools and all of that, which is making it so simple and people are curious to know, build themselves. So that would be I think one of the ⁓ factors of motivation as well, right? ⁓ tell me more about ⁓ What do you love doing other than your nine to five job? speaker-0: I like golf. at least one of the like Saturday or Sunday, I go with my son or friends and play nine holes. So that has become standard routine for the last couple of years. Besides that, I like to make some cartoons. speaker-1: Is it caricature? ⁓ my gosh. speaker-0: Yeah. Yeah. And then I like singing as well. speaker-1: Great. So I think I'll post some of your caricatures at the end of the podcast, ⁓ saying that speaker-0: ⁓ I don't know, I will make them public but yeah I love doing it myself. speaker-1: ⁓ why not? I mean if you have that skill, why not show it? If I remember you were writing a book as well, right? speaker-0: book is called Leaders Guide to Data and AI Culture and ⁓ it is written by two authors in my college, Emelon network. ⁓ And what they have done is like they came up with the framework to measure the AI culture in your company and what all needs to be done to cultivate AI culture to drive impact. And then they approach many data and AI leaders across the globe and ask to contribute a chapter. So I contributed a chapter there, which talks about how can you leverage AI ⁓ culture to drive digital transformation. Great. That's where I discussed certain case studies from my experience and thought leadership. And what I believe is ⁓ the impact and culture reinforces each other. So you have to start somewhere, show the impact, and then that reciprocates into the right light level of culture. So I have contributed one chapter in the book. speaker-1: Great, great. ⁓ would love to read the book and would love to see that what differentiator ⁓ is mentioned in the book by you. Thank you so much, Prashant, for being on the show. Really appreciate your time, your insights, your useful information. And I'm looking forward to meeting you in person again. speaker-0: Thank you so much. Thanks for inviting me.