AWS Marketplace Newsroom | Ash Kulkarni, Elastic

Clips
More from AWS Marketplace Newsroom

Speakers

Ash Kulkarni

CEO

Elastic

play_circle_outline Introduction to Elastic CEO Ash at AWS re:Invent 2025 and industry context.

play_circle_outline Emphasis on data abstraction and the role of agents in enterprise AI.

play_circle_outline Elastic and AWS Team Up for Enhanced Data Analytics and AI Development: Announcing a New Competency Award

play_circle_outline Elastic's Vision: Pioneering Foundations for Unstructured Data Applications in a Multi-Billion Dollar Marketplace

Info
Transcript

Ash Kulkarni, Elastic

Ash Kulkarni

CEO Elastic

In this interview from AWS re:Invent, Ash Kulkarni, chief executive officer of Elastic, joins theCUBE’s John Furrier to discuss the critical role of unstructured data in the emerging era of agentic AI. Kulkarni details Elastic's latest strategic announcements, including a new integration with AWS Agent Core, achieving AWS AI Competency status and a joint solution with Accenture available on the AWS Marketplace. The conversation centers on the concept of "context engineering," where Kulkarni explains how Elastic serves as the essential data layer that allows L... Read more

explore Keep Exploring

What is the focus of theCUBE's coverage at AWS re:Invent 2025 and who is being interviewed? add

What is an update on Elastic and re:Invent, and what is the significance of agents in relation to data? add

What were the key announcements made regarding the integration with AWS and the AI competency recognition? add

What is the vision and purpose of the company's approach to artificial intelligence and unstructured data? add

bolt Powered by CUBE AI

Ash Kulkarni, Elastic

search

John Furrier

>> Welcome back around to theCUBE's live coverage here in Las Vegas for AWS re:Invent 2025. I'm John Furrier, host of theCUBE. It's our 13th year covering AWS. And we've seen the growth, we've seen the abstraction away around the infrastructure. And now with agents, you're starting to see the abstraction around the data and the work. And now you're seeing a whole nother level of applications. Ash is here. CUBE alumni going back over 10 years. Now the CEO of Elastic. Ash, great to see you. Thanks for coming on.

Ash Kulkarni

>> Great to be here. Thanks for having me.

John Furrier

>> You guys have been in the center of DevOps, search, cloud native. For a long, long time, we've been covering you guys like a blanket. But the world has spun in the direction that gives you guys a real lift with what you guys have been building. Give a quick update on Elastic and re:Invent here on the news. But then let's talk about the agents and what that means because it's a data problem.

Ash Kulkarni

>> Indeed.

John Furrier

>> And a data opportunity.

Ash Kulkarni

>> Great question. Great tee up. So the way we think about the world is our value to the world is making it possible for you to exploit all the information that you have in your unstructured data systems. And in the world of AI, as you know, John, it's all about context engineering. It's all about providing context to these large language models so they can actually do their job correctly. And at AWS, the announcements that we made this week have all been around that same theme. So the first thing that we announced was our integration with AWS Agent Core. That just allows anybody who is building agents on data analystic search, integrating with the AWS agent core technology to have a seamless experience. The second thing that we announced was that we were awarded the AI competency. One of the first ISVs that AWS awarded this competency and just based on all the work that we've been doing in the AI space around retrieval augmented generation, around context engineering, around agent development. And then the third thing that we announced was a joint solution with Accenture that includes both our platform, their skillsets, to help customers jointly build technologies around agentic workflows. And that will also now be available through the AWS marketplace. So it's really a three-way kind of relationship, but those were the big announcements.

John Furrier

>> Yeah. And you guys are well positioned on obviously the data side. We've talked many times around the value of the data, the data layer. Matt Garman, when I interviewed him for his re:Invent keynote in Seattle before the event, he said, and I'd like to get your thoughts on this, the next 80 to 90% of enterprise AI value will come from agents. And then of course, Nova Forge and the frontier agents kind of set the table for the direction, which is bring your data to kind of a half-baked model, however you want to look at it with the checkpoints, and then make that a frontier model without paying for it. So that only works with data. Okay. Frontier agents is a system. It's not a chatbot toy. It's a system architecture because it's got deterministic and non-deterministic kind of capabilities. Share your thoughts on that because you got that value clearly with agents. Everyone sees that enterprise value.

Ash Kulkarni

>> Right.

John Furrier

>> But then you've got these models now are going to take data, which has been your business. And then you got the frontier agents that do the work. And so you almost have this kind of next level cloud concept of abstracting away the data and the work where the agents are now doing the frontier agents of doing the work, but it doesn't work without the data. So tie that together, the whole...

Ash Kulkarni

>> Yeah. So there are lots of areas where Matt Garman and I are incredibly aligned in terms of our view of the world. So the value, so when I talk about what I think is happening now and is going to happen over the next few years, I think every system process, business workflow, that can be automated will be automated with AI. I truly believe that. What that means is you're going to have more and more agents. Data will be created in such a way that it can be easily consumed by agents. If you think about documents today, right? So we create a lot of documents in process workflows in our business, but we don't think about it as, who's going to read that document? Our first thought is it's a person. It's the HR person, it's the engineer, whatever. We will be increasingly creating content with the view that it'll be agents that will be consuming those documents. They'll be consuming that information and then acting upon it, right? And you're going to find various forms of these agentic workflows, some that are human in the loop, some that are more autonomous, but that spectrum of autonomy will be quite interesting as it evolves. I also agree with Matt that it's all about the data. At the end of the day, it doesn't matter if it's a Nova model, whether it's a Gemini model, whether it's an open AI model or an open source model like Llama or Mistral, all of these models are trained on publicly available data sets. And so they really have no context about what's private to your business. And that's really the task of context engineering. That's the task of connecting the dots between all of these models and your private data. Now, where I would say that we feel that there is going to be an evolving world and maybe there are some differences in how we see the world than what AWS looks at it, I think there's going to be more than one model. I don't think that Nova or Gemini or OpenAI or Mistral, nobody's going to win the war and be the one single model to rule them all. You're going to have specializations. But what's going to happen is you're going to see the price of inference start to come down, because the more competition you have, it's better for consumers, it's better for users, that's going to bring down the cost. Everybody is going to try and one up each other. So customers will demand choice. Customers will look at things not just from a model perspective, but they'll say, "Okay, I have a choice of models. I have a choice of cloud computing platforms. I have a choice of GPUs or TPUs or Trainium chips or whatever. How do I now look at things from a data perspective and go, 'What's my new platform?'." So I think the new platform is going to be these data systems, these data platforms, right? So I do agree that there's going to be an emergence of a new type of cloud, but in my mind it's going to be a data cloud. And the role that we are very determined to lead in is the role of being this data cloud for unstructured, messy enterprise data that has the greatest value in it.

John Furrier

>> Yeah. I like that. I agree with you, by the way, on the one model not ruling the world, because there is a power law. And one of the things, if you look at the big models, it's clear from this event and all the people sharing their stories and innovation, yeah, they're limited, but they're biggest. So they're the biggest. They have everything, but they don't have everything. So the domain specific intelligence that's going to come from the integration of the data makes that. So these large languages don't have my company data in it. They're not trained on that. So again, that's where I think you nailed it. Now on the data cloud, it's interesting. After I met with Matt, my headline was something to the effect of the new cloud era is the agent cloud, because what they're doing is providing the inference and the silicon to try to make the cost per token per watt, a metric as low as possible, give some agents out there. They announced three agents. So they're already kind of telegraphing that they're kind of doing the Amazon playbook for agentic infrastructure.

Ash Kulkarni

>> That's right.

John Furrier

>> So it's IS and pass, but it's a whole nother thing.

Ash Kulkarni

>> And they want to bring down the cost, which is great, right? Which is great for us. It's great for our joint customers. It's great for the market. I think there's another thing that's really interesting that's going to happen here is when you look at where data gets created, data is often found in silos. So you look at a typical enterprise, they'll have an ERP system, they'll have a CRM system, and you know that Salesforce wants to be able to control all the data that's within Salesforce and have customers use Agentforce. And ServiceNow has its own Aagentic framework and they want to control the data that's in ServiceNow and so on and so forth. So the hard part is going to be figuring out how to stitch all of these things together in a standards based way. And really, that's the role that we want to be able to play and provide to our customers that we will work with all of these data stove pipes and we will make sure that you are able to access all of that information in a seamless way, fed rate where needed, have relevance on that data in the most accurate way. And I think this is what makes it very exciting.

John Furrier

>> Yeah. If you look at Amazon from 2006 to 2025, they basically abstracted away the data center or the servers. Now they're extracting work. If you believe that, you say, "Okay, what's the value process to a customer?" And I want to get your thoughts on this, because you guys have been playing in this data world for the cloud native side now with this new AI native side. The AI native startups that we talk to on theCUBE and our team, they're not thinking about coding to a model. They take your worldview of, "Hey, whatever the best model is, I'll just use the best model."

Ash Kulkarni

>> That's right. That's right.

John Furrier

>> Or use the best model for what I need. I think they use Reddit as an example in one of their use cases where they have their own data, moderation, all kinds of things that they're doing. They don't need to have the whole model. They just take a little bit of Nova and then bring their data in, with very little fine-tuning, they have solutions.

Ash Kulkarni

>> Yep.

John Furrier

>> That's an answer.

Ash Kulkarni

>> That's right.

John Furrier

>> So how does a basic IT department... Because everyone fears change, right? So like IT and people who are doing cloud native, they don't want to start a new thing. They want to have a continuation. They have data lakes, they have these things.

Ash Kulkarni

>> Right.

John Furrier

>> What's the customer orientation on your side where you see the customers? Where are they leaning into? What are they doing? Are they doing a data cloud with Elastic? And how are they thinking about agents from their perspective? Because they have the crown jewels, data.

Ash Kulkarni

>> I think that's the right question to ask, which is, what's the natural starting point for users, for customers? Because fundamentally, every day there is a new announcement. This week it's all about Nova. Last week it was all about Gemini. And next week it'll be something about Garlic, right? The new model that OpenAI seems to be talking about. So there's so much noise out there that customers genuinely don't have the time or the ability to like truly, truly research everything. So what they want is just a way to be able to work with all of these systems in an interchangeable way and then figure out as they go what they need and what's best for what use case. So that choice is incredibly important. The second thing is, like you said, it's all about the data. What are my crown jewels? What is the core competency that I have as a business? It's all the data assets that I have. It's my understanding of my customer buying propensity. It's my understanding of my supply chain and how I can use it to my business advantage. Whatever be your specific. In the case of Reddit, it's my understanding of all of the subreddits and all of the forums and how do I understand the sentiment in all of those forums and how do I extract most value out of it. I think that's the reason why what we are seeing is customers are starting the process of building an agent with their data. And so when we came out with our agent builder product, that was the fundamental thought, which was, "Hey, what's the most natural thing for a developer? The developer's going to start by looking at what data they have. So why don't we make it possible for them to almost have a conversation with their data?" Literally the experience looks like a ChatGPT bar and they start to chat with the data that's in Elasticsearch. And through that, then you go, "Okay, now do you want to take this and turn it into an agent?" And we just allow them to turn it into an agent that can be accessed over MCP or A2A protocol so they can publish it anywhere. It can talk to something built in AgentCore. It can talk to something built in AgentSpace, which is the Google technology. But fundamentally, the idea is that you want agents to be able to talk to each other and be able to work off of your data. And I think that's going to be the most natural approach.

John Furrier

>> Ash, you have a history, we've talked in the past. We've been on the data side for over decades, two decades. You mentioned MCP and A2A, those are the protocols. So if I look at the slides that they present, this is pretty much our industry, but here at Amazon, presented the slide with the Nova Sonic and all that stuff. At the bottom, it had protocols, only two, MCP and A2A. And then on the framework side, you had everything in there, including Strands, which is their thing, which I kind of like. But you have OpenAI SDK. So you have all the frameworks, you got the protocols. How are customers looking at that? Because now certainly choice is there. Then you've got coding assistance and transform. So now you have a whole nother dimension. You guys at Elastic own the developer community, you guys have great loyalty. Developers are, again, front and center here again, and they will be the drivers and agents on cloud native. What's your view of how this gets rolled out from a developer standpoint? What are they doing? What are they tackling first? Will they leverage these other agents? What's the developer pattern?

Ash Kulkarni

>> What we are seeing from most developers that we engage with is they are starting with very specific problem sets that they want to build an agent for. So it isn't this grand, "Build an agent to solve everything in my enterprise." It's like, "I have a particular need." I mean, literally, I was talking to somebody earlier today. They are trying to create a finance agent to help their rev rec teams be able to improve the speed with which they're able to do the close for the quarter, like all the books. And it's a very specialized need, right? It's going to involve some of their financial data. They're perfectly fine using any model because it's not a super complicated problem, but they want to start with that data and quickly build that thing. Now their vision is, "Okay, once I build that, then wouldn't it be great if that agent were able to talk to another agent that my FP&A team was building that could use that information about the revenue close in their forecasting tool, in their forecasting agent, so to speak?" So I think it's going to be built in terms of these modules, these agentic engines that will be specialized for a certain problem, and you will compose them using the protocols at the simplest layer. I almost find MCP to be as simple as HTTP. And the reason why the internet became so successful is because the protocols were dead simple. They weren't overly complicated. We didn't try to put too much sophistication in them. I think that's why MCP is so wonderful.

John Furrier

>> And it has organically grew.

Ash Kulkarni

>> That's right. And MCP, once it was published, everybody just loved it and glommed onto it. I think A2A is taking a little more time because each of the model vendors are trying to inject their own little nuance in it.

John Furrier

>> Yeah, classic. Yeah, classic open source kind of dynamics. You mentioned some of the frameworks. I want to get your thoughts on this because I want to go back in history. Back in the big data days, and up until, say, a couple years ago, data hygiene was kind of like the talk.

Ash Kulkarni

>> Sure.

John Furrier

>> In order to have a good data platform, you got to do your data hygiene. Is there an equivalent version for agents? Because with marketplace, I could literally get Elastic, plug it into my stack, vertically integrated stack, get my data in, do whatever I want with the models. So it's very easy to deploy and get going. And talking to some of the engineers here in the hallways, the comment was you can automate anything, but if your product doesn't get tuned up properly. And they were kind of speaking to the metaphor of hygiene in my comparison. So is there an equivalent, get your data hygiene right to get the data right for agents? Because there seems to be, you can have a workflow, but it's not tuned up properly.

Ash Kulkarni

>> Sure.

John Furrier

>> What's your thoughts on that? It's kind of a weird question, but I want to kind of get-

Ash Kulkarni

>> No, it's actually a really insightful question. Now here's the thing. I mean, I've been in the data space for the last almost 30 years. And you talked about data dictionaries and data correctness and all of that. We've been building data dictionaries and metadata frameworks and ontology libraries and all of this stuff for, I don't know, decades, right? And how successful has it been? It's I mean-

John Furrier

>> A lot of hard coding.

Ash Kulkarni

>> And it ends up being incredibly brittle because these pipelines keep breaking. Your dictionaries become obsolete by the time you put them out into production. And that's the reason why when you talk about master data management or dictionaries and so on, there's no company that you can point to and go, "oh, that's a multi-billion dollar business." Right? There just isn't one. And why is that? It's because of this problem that data is really hard to organize. And by the time you think you've organized it, somebody has messed up all of your logic. And so the way we look at this is LLMs are actually giving you the opportunity to stop worrying about the exact categorization of data. Let the LLM help you in real time, figure out what is that data for, what is it most relevant for. You can ask the data questions to figure out, "Is this data the right data I should be using in this particular use case?" So use the power of these language models to do the work that humans have in the past attempted to do in a very static manner. Do it dynamically by, instead of having a human do it, have the machine do it, right? And so that's what it's all about when we talk about the search lake. So when we talk about our approach and why it's different, we don't believe in creating ontologies and creating these lake house gatekeeper thingies that some of the more structured cloud providers are talking about. Our view on this is data is going to be messy. Good luck to you if you think you're going to be able to perfectly categorize it. The analogy I use is like, when was the last time you were successful in foldering all of your emails and like, are you happy with that system? I haven't met anybody who is.

John Furrier

>> Yeah, exactly. And also we heard a lot of that MIT study, which we debunk, but there are experimentations going on. People will fail with agents.

Ash Kulkarni

>> 100%.

John Furrier

>> And so that will point to where you optimize for.

Ash Kulkarni

>> Yeah. And you want to optimize for things in a modular way so you can actually deliver success and you want to use the agent to help you refine the data. And so that's the reason why when we built our agent builder product, the whole idea, John, was start with the data, use the large language model to help sift through the data, identify what's the best way to use that data, solve a particular problem, and then modularize it and keep modularizing it and build that kind of structure that then allows you to do some very sophisticated things like entire financial planning, automation, so on and so forth.

John Furrier

>> Yeah. Well, you guys are in a good position, Ash. I really appreciate you coming on. I have to ask you about the business.

Ash Kulkarni

>> Yeah.

John Furrier

>> Where's the momentum? Where's the traction? And what's your vision as we look at this moment in time? I think it's going to be a game shift here for sure in a good way. What's your vision and how are people using Elastic today?

Ash Kulkarni

>> So the crux of our vision is that it's all about AI applied to unstructured data. It's the last frontier. You have lots and lots of systems that for decades now have figured out, have mastered how to deal with structured information. But when you deal with unstructured data, and you just think about all the unstructured data, it's like you probably have a document of information to prepare for this interview and you probably had that for every interview that you're doing this week, you are probably going to do a write-up afterwards, there's so much information that we create that's unstructured. How do you get value out of that? That's the crux of what we do. That's what we are going to specialize in. My dream is every app that automates processes that require unstructured data will be built on top of Elastics technology today, tomorrow, and for many years from now. And then every problem that involves that same kind of messy unstructured data and observability and security are perfect examples of that, but there will be more in the future. And that's our path to continued growth. And so we believe that we are well on our way to building a multi-billion dollar business this way. And in the process, be a key infrastructure provider for hundreds of thousands of customers.

John Furrier

>> Well, you have the marketplace for AWS. That's a great way to get the integration. You have the competency, you have the essential relationship. That's key for your go to market.

Ash Kulkarni

>> We are building out that scaffolding, if you will. Like the partnerships, the hyperscaler relationships, the breadth of the platform itself. And our enterprise selling motion, we are now about a 13-year-old company, so we are still reasonably young, but just we've achieved the scale that our team feels very proud of and we are really well set up for growth.

John Furrier

>> You guys have a great tailwind. And like I said, you guys have always been data-centric. You had search, Elasticsearch was well documented, open source. We were a customer, a user in 2011 for our CUBE platform. Ash, thanks for coming on. Great to see you. Congratulations. Appreciate it. All right.

Ash Kulkarni

>> Thank you very much. And I think in 2011, so it's about time that we brought you on as a customer now.

John Furrier

>> We're working on it. All our notes are on unstructured data in our cloud.

Ash Kulkarni

>> There you go.

John Furrier

>> We'll get back on track. Thanks for coming in. All right. I'm John Furrier with theCUBE. Live coverage continues after this short break.