Name: Alex Chantavy, Subimage.IO
Uploaded: 2026-07-29T17:35:00.000Z
Duration: 26 min 22 s

Clips
More from theCUBE + NYSE Wired: Data + AI: Turning Data Into Knowledge for Autonomous Systems

Alex Chantavy

CEO

Subimage.io

play_circle_outline Graphs as Ideal Data Structures for AI: Knowledge Graphs to Map Cloud Assets, Attack Paths, and Fixes

play_circle_outline Attackers Think in Graphs: Mapping Cloud Assets to Prioritize Fixes

play_circle_outline Taming Alert Fatigue: Graph Databases and Localized Context Engineering to Prevent AI Agent Hallucinations

play_circle_outline Locality-Limited Graphs for Agent Automation: Narrowing Context Windows to Avoid Hallucination

play_circle_outline From Cartography to Autonomous Defense: SubImage's Graph-Based AI Platform for Cloud Security and Remediation

play_circle_outline Neo4j's strengths: Cypher query language and scalability

play_circle_outline Neo4j's value: Cypher query language, scale, practitioner-focused ecosystem

play_circle_outline Practical challenge: building large ontologies (hundreds node labels, thousands relationships)

play_circle_outline Governing AI Agents: Open Source Control Planes, Identity Standards, and AI-Native Infrastructure

play_circle_outline Industry parallels to cloud-native: need open standards, governance, control planes for agents

play_circle_outline Why Graphs Outperform Relational Joins for Explainability, Lineage, Observability and Deterministic Agent Substrate

play_circle_outline Graph-Based Explainability and Vulnerability Lineage: Prioritization and Least-Privilege Controls for Machine-Speed Agents

19:00

18:08

Info

Alex Chantavy, Subimage.IO

Alex Chantavy

CEO Subimage.io

John Furrier

>> Welcome back everyone to theCUBE here in San Francisco at Neo4j's GraphTalk in San Francisco. Of course, they have a New York City GraphTalk coming up in September. I'll be there as well. And it's where we're documenting and also talking with the leaders who are in the trenches, the practitioners, the engineers, the ones that are doing the work in AI that's powering the new data structure, that's actually creating the ontologies, the brains, the knowledge graphs that are powering the value that's being created by the new graphical user interface, which is now natural language that AI is now driving. A whole new sea change in the stack and in the data layer. It's a really, really big, important point. Of course, we're going to continue to talk about graphs. Alex Chantavy's here, CEO of Subimage.io. Welcome to theCUBE. Thanks for coming on.

Alex Chantavy

>> Oh, thank you so much for having me.

John Furrier

>> So all these communities start out small, very intimate mixture of experts. This is kind of feeling the same way, this event. The talks are great. They're all kind of insiders, all graph friends, colleagues, all love graphs, but also new people are coming in and coming to realize that graphs are perfect fit for AI. What's your reaction to that?

Alex Chantavy

>> About time.

John Furrier

>> It's about time. It's kind of like relief and freedom from in the database world. I mean, why is Neo4j popular right now? Why are graphs so important in your opinion?

Alex Chantavy

>> I think that starting off, I've been using Neo4j for about over 10 years now. My background, I used to work in government, working in cybersecurity. I eventually moved over to Microsoft on the Azure Red Team. And what you do in the Red Team is you look throughout the environment, you try to find your way in and then try to move toward objectives. And there's a saying that attackers think in graphs, defenders think in lists. And as long as that is true, the attackers will always win. And so as long as you have that map, you must have that map in order to stay ahead. And so why not apply that same kind of mindset to our own defense and fixing those problems? Which is the mission of my company where the idea is that we'll take a map of everything that you own, the cloud, identities, workloads, your devices, put that all together into a map. And then that way, because nowadays security teams, they're all inundated with all the different findings you can have, all these other different problems. You don't need another thing that generates findings for you. You need something that puts it all together, contextualizes it in your environment, and it helps you decide, "Which of these problems that I have are open to the internet? Which of them presented a viable attack path over to production to steal sensitive data so I know where to prioritize my fixes?"

John Furrier

>> Yeah. I mean, when the perimeter was collapsed over a decade ago with cloud, you had now endpoint protection, the zero trust all came out because of that. The game changed. You mentioned attackers think in graphs, defenders think in lists. How would you advise someone to think in graphs? If someone's not a graph-trained person, what would be the technique? Do I just start learning graphs? Do I just dream in graphs? What's the technique? Is it meditation, technical meditation? What's the approach? How would you advise someone who wants to get up to speed? Because there's a real interest. People are now seeing graphs as the key piece of the puzzle.

Alex Chantavy

>> I think that everybody can think in graphs. If you're familiar with Harry Potter, I solemnly swear I'm up to no good. You have the Marauder's Map, you get to see Professor Snape running around the halls and seeing all that happen. Or even like you watch TV, you watch the FBI, they go looking after their main culprit, you make your own big board about how you connect ideas together. This is your mind map of how you look at the world. And the lists of separate things, sure, in isolation, they can be useful, but everything's become so much more powerful. At least in my domain with security where you have ... I've got my laptop, I own this laptop. If somebody gets malware on my laptop, they can get access to my GitHub account, my AWS account. Once they get access there, what else can they jump? It's all these paths into something a lot more interesting. And I think that that's something very natural for humans to grasp.

John Furrier

>> Yeah. And it's also a lot more paths. If you had to do a trace and like, "Okay, what are the possible paths?" There's a lot more paths now than traditional architectures because of all the new software. And security in particular, you mentioned it earlier about getting all these findings. There's also vendor sprawl. I mean, every CISO that I talk to has the same pain. I don't need another tool. That was about, I'd say 5 to 10 years ago, that was consistent theme now is I don't need another platform. So you have a lot of platforms now sprawling and it's like walking into a hospital and hearing code blue everywhere. It's like there's too many alarms going off. It's like there's too much things to look at, pay attention to, almost you get fatigued. So there's a lot of fatigue, there's tool sprawl, platform sprawl. And then ultimately, to your point, it's like, "Okay, things are happening and I got to look at it." Okay, that's happening. But now, you got AI, you got agents. People are recognizing that there now are ways to automate work. So now, that's here. How do I get from my list-based approach, current situation to getting to that automation? And what does graphs do for you? I mean the map, besides the map, how would you lay that out? How would you instruct an intern or a new hire to get up and running to move from point A to point B, old to new?

Alex Chantavy

>> I think that one thing that 2026 has taught us is that AI agents, they're here and here to stay. There's just so many useful workloads and non-deterministic things that are useful and doable that would never have been possible without it. So in this new world, the key or one of the key problems with how do you make your agents useful and correct and reliable is all context. How do you get the right information that the agent needs at the right place? Because if you are to, at least in my domain with cloud and with devices and technical data, if I was to take a big massive dump of a huge enterprise, throw it all together at the agent, you'll blow up the context window. And then even the smartest agents in the world, there's technical limitations there.

John Furrier

>> You got to feed the agents.

Alex Chantavy

>> Yes.

John Furrier

>> So the feeding the agents becomes like school.

Alex Chantavy

>> Yes.

John Furrier

>> It's a crash course instantly. So is that data pipelining? What's the process? I mean, is it to get the graph, first-party relationships mapped out and then feed data?

Alex Chantavy

>> One of the cool properties or one of the key properties of graphs is locality. So you have, or at least in my domain, I have an AWS EC2 instance. I have alert. An alert has fired. There is malware running on this EC2 instance. Now, how do I spider out from there? What is everything immediately relevant to me? If I feed my entire environment, I don't want the agent hallucinating about something that is unrelated completely. I need to know, "Is this instance, this instance open to the internet? What permissions does it operate as? Who touched it last? What code is running on there?" And all of that is all about locality.

John Furrier

>> That can be a deterministic feature. This is where agents are bringing back deterministic. It's interesting to see the whole first wave of AI was not deterministic, which we saw. Type in a prompt, you get an answer, it's always different. That's generative. But now, you can take non-deterministic workflows and make them deterministic. We're seeing that in robotics, for instance. So agents are bringing back that, but also the scope of the compute. You don't need monster GPUs and context windows if you get it right on the data structure side.

Alex Chantavy

>> Exactly. So the data structure, I will say that you create it deterministically and then the agent has something reliable to ... There's something reproducible also. I think it's just an opinion of mine. I don't think that creating the data structure itself, I'm not a fan of doing that in a non-deterministic way. You make the map, you make that a really reliable data substrate, keep that maintained. The agent will always have something. There's already enough variables, but the agent-

John Furrier

>> And they have reasoning too. They can actually go in and reason. Okay, so talk about your company. History, where are you guys at? Phase, status. And what was the origination story?

Alex Chantavy

>> Oh, this has been ... It's still a long time coming. So the origin story of Subimage back in 2019, I was working at Lyft and then we had open sourced a tool called Cartography. So very apt with maps and everything looks like that. My mentor, Sacha Faust and I and a team of, shout out Evan Davis and Taya. Also, what we all did was we built out an open source tool in Python, grabs all your data from AWS, Okta, every single different provider that a cloud native company at that time would care about, puts it together in a map. And at that time, nobody else was doing that. And then we've just been so excited about ... We did zero marketing for this. Nobody had ever said-

John Furrier

>> Word gets out fast, believe it. Something's really good if it gets out fast. Okay. What happened next?

Alex Chantavy

>> Oh, and then the idea caught fire. We built out a community out of it. As the years went on, we eventually donated the project to the Cloud Native Computing Foundation over the Linux Foundation. And I decided last a year and a half to go turn into my own company, Subimage, where we are accelerating those ideas, building a security platform around graph, leveraging AI to go turn visibility into something actionable and be more outcome oriented, being that substrate that lets agents take action and fix problems in the environment.

John Furrier

>> And how's that going right now? How many employees? How many people are on the team? Funding? Give us an update on the numbers.

Alex Chantavy

>> So we finished Y Combinator last year, last March. We were in the winter '25 batch. We raised a 4.2 million seed round. We are keeping the team very small. Small team, big dreams as you do. Especially now with agents, you try to accelerate yourself and scale out that way, and it's been a whirlwind of a ride. I'm super proud of the people.

John Furrier

>> And what's the culture like? If you had to describe the culture, say I want to go work there on a potential higher way, what kind of culture do you have? What's the culture like?

Alex Chantavy

>> I don't want to speak on behalf of my whole team too much, but I think that one strong attribute that every single person that we have is an extreme sense of ownership. And then because we really believe that you own your outcomes, you have a belief in something that you want to do, you have conviction, and then we want to be able to empower anybody that works here to see it through to the end.

John Furrier

>> And you're a small team, so you can keep that DNA. And now, you're using agents to scale, sounds like.

Alex Chantavy

>> Yes.

John Furrier

>> So this is a classic small team. Again, startups are so much "easier." It's not easy. No, but it's different. You can actually get a lot more product market fit, focus on the value proposition, keep honing on that, get the customers happy.

Alex Chantavy

>> Yes.

John Furrier

>> What's the customer momentum like? Can you share some anecdotes? You don't have to name names, you don't have to, but what's the vibe? What's some of the outcomes you're seeing? What's been the reaction?

Alex Chantavy

>> Yeah. I'll be the first to say that we're super proud that Neo4j was actually one of our very first early customers. And the partnership together with Neo4j has been nothing short of amazing. Being able to meet the entire security team, push us on these scenarios, and then truly run us at the scale that you're operating.

John Furrier

>> And they had a good year too. I mean, Neo4j was misunderstood for a very long time. Now, the cat's out of the bag as they say. How would you describe Neo4j? I get this question all the time from people that are putting their toe in the water, graphs. I mean, you have kind of a graph community, but now you have a lot of in migration coming into the graph community. And everyone, there's a lot of confusion on what the best graph is for which use case. How would you describe Neo4j from a value prop? I was talking to a friend about this. He's like, "Wow, there's still a lot of graphs out there. There's not much better graphs going on." And I go, "Well, it depends how many nodes you want to turn on." So I'm hearing different things, but how would you describe what makes Neo4j different from others?

Alex Chantavy

>> I'm extremely practitioner focused. I've been in the trenches for a very long time and my answer is likely going to be extremely merity.

John Furrier

>> Yeah, that's okay. That's okay. We like that.

Alex Chantavy

>> I know every detail about how Neo4j can fail. I know every detail about how it can succeed also. I think that for me at least as a end user of the product, I think that the reason why you would go over Neo4j over everything else is the Cypher programming language. So the query language, being able to take all of your nodes and understand them in Husky art. It's a declarative way of describing the path that you want to match and you just come back and get it. Yes, there's some quirks if you want to do some aggregation, but this construct is just incredibly powerful.

John Furrier

>> And I heard scale too is another big point with Neo.

Alex Chantavy

>> Yes.

John Furrier

>> So talk about that result, the benefits of graphs. Folks who are trying to really kind of jump in the deep end of the pool, so to speak, they want to go all in. What are some of the things that they can get out of this? What are some of the benefits you would describe for graphs? What do they get? Obviously speed, you get the bigger first-party citizen with relationships. That's a key. Is there a community support? What's some of the things that make the outcomes shine?

Alex Chantavy

>> There's a strong ecosystem, strong first-party support in open source along with all the tooling put together with Neo4j there. And I think that it's been tried and true and battle tested basically. I've run it at scale at ... My friends have run it at much, much, much larger companies, like things that are Fortune 100s basically.

John Furrier

>> I love this event, because when you get these kind of events like the GraphTalk, it's like the agent conference that we had ... You get the practitioners who are actually building out the infrastructure side of agentic. They're basically cloud native meets data engineering nerds and they're in the weeds and rightfully so, because you're talking about infrastructure at this point. And they're small communities and they all talk to each other and a similar vibe here. One of the things I heard here was is that AI workloads that move into production the fastest are most have graphs. What's your reaction to that? Do you agree? Was that just hyperbole, or is that real?

Alex Chantavy

>> I think that for us, I'll add a couple of caveats there that there is a decent amount of engineering work that needs to make the context engineering work. So being able to teach it. "This is my node, this is my semantic layer, this is my mapping between all these concepts." And then especially if you have a data model that is very large with, at least for us, we have over 500 different node labels and maybe about a definition of 1,500 different relationships. And again, the relationships are really what it is. And so teaching the brain or teaching your whatever agent that this is how to look at it, that's still a challenge.

John Furrier

>> It's an architectural exercise really.

Alex Chantavy

>> Yes.

John Furrier

>> It really is. I mean, you're laying out the map. You mentioned the Harry Potter example. So that's a really great metaphor because you got to know the rooms. You got to know the relationships. How hard is that to do? I mean, people are just now trying to understand. What does it take to do all this? How much architecture's involved? What's the work? How would you answer that question?

Alex Chantavy

>> It's an open problem right now. I think that it's something that I believe that we have ... I'm pretty proud of the solution we've made so far. But that said, I mean, there's a lot more improvement to be had. And then this is something that the entire industry is just fighting over right now. What is the smartest way to supply the right context and when you can figure that out?

John Furrier

>> I like context engineering. It's a great term, because you really are engineering the system. And it is a system, it's not a solution. It's a system. And the context is the key. Philip always says that on theCUBE. I'm like, "Yeah, I totally agree. I'm not going to disagree with that." But what is context engineering? How would you define what is context engineering?

Alex Chantavy

>> It's taking a very disciplined process about getting the right inputs to an LLM and then making sure that the outputs align with your expectations. And typically, this is done with the evals or having some kind of evaluation set where you have your expectations and making sure that everything is behaving the way that you expect it to.

John Furrier

>> Talk about Subimage. What's on your plate? What do you guys do? What's the value proposition? Put a plug in for the company. Obviously, love what you did with the cloud native. Makes total sense. Love the Linux Foundation. We've been to every KubeCon since existence. So we watched that grow, that community. By the way, Lyft was an amazing contributor. They were a huge partner. Intuit was as well, they were on earlier here. That community really built cloud native.

Alex Chantavy

>> Yes.

John Furrier

>> And now, cloud native is the foundation of AI.

Alex Chantavy

>> Yes.

John Furrier

>> Especially in large scale enterprises. So what's the product you guys have? How do people engage? What is it? What's the value? Give a plug.

Alex Chantavy

>> Yeah, absolutely. Well, first of all, Lyft is like, I would say so much respect to all of the cracked engineers there, the infra team, the absolute world-class, the absolute best. So for us with the company, we are going ... As I mentioned earlier, with Subimage, we map the entire cloud and entire environment for the security team and doing things like understanding your identity, understanding your SaaS sprawl, understanding your cloud, how that all maps together. And the differentiators that we're seeing is that how can we take all of the issues that you see in your environment and prioritize them so that an overwhelmed security team knows what to surface first and what to work on first. Because as I mentioned earlier, we can look at it from an attacker's point of view. I'm going to look at where is the weakest point to go in to go and make my attack and then get something valuable. Well, why can't we take that approach as defenders? As defenders, I'll just be frank and say it sucks to be a defender.

John Furrier

>> Yeah.

Alex Chantavy

>> When you're a defender-

John Furrier

>> You're always losing.

Alex Chantavy

>> You're always losing.

John Furrier

>> There's a lot of L's on the scoreboard.

Alex Chantavy

>> You have to be right all the time. The attacker just has to be right once. And then what you get, the experience of a defender, typically, you'll have a big, massive list of all the stuff that you have to fix. That's not fun. How does this beleaguered security team justify to their leadership, "All right, these are the immediate risks that I have to fix now. These can be safely deferred. Here's the evidence to back all of that"?

John Furrier

>> The explainability becomes a huge deal. This is where graphs shine, in my opinion.

Alex Chantavy

>> Yes.

John Furrier

>> All right. So you mentioned Lyft. Matt Klein and Vicki Chung had been on theCUBE. Vicki then wound up going out to co-chair.

Alex Chantavy

>> Amazing.

John Furrier

>> That's one point. And Matt had been a legend.

Alex Chantavy

>> Yes.

John Furrier

>> I think it was around early days of KubeCon. It taught that Envoy was a huge success project. They had pretty much dominated day zero for multiple years and then it became big. What was the learnings from that? Because Lyft, like Uber, you guys built first generation columns or data store. I mean, I've followed in depth around some of the stuff you guys have done. You've had the perfect graph scenario, because you had to manage people, places and things, third-party drivers, real-time communications and streaming data. I mean, I don't think there was any Kafka at that point when you guys were doing that. So a lot of that stuff emerged at open source. Is there a similar experience happening in AI where you're starting to see the end user contributions? And what should people pay attention to? Because open source is the area for talent acquisition, community growth, innovation comes from open source. What are the hot areas if you had to look at ... In fact, it has the AI Agentic Foundation now, the AIIG I think it's called. Agentic AI Foundation, the AAIF.

Alex Chantavy

>> Yeah, that's the equivalent of CNCF, right?

John Furrier

>> It's the new CNCF. Yeah. What's the hot things that might come out that you might envision? Because I think there's going to be an AI native community that's going to be very complementary to cloud native.

Alex Chantavy

>> Absolutely. Well, first shout out to Matt Klein. Taught me everything I needed to know about open source. I followed everything, all the advice that he gave about how to grow Cartography as an open source project. This is such an excellent question. I think that it's very similar now. 2026 feels a lot like 2018, 2017 in the sense that I spoke about this at Microsoft's BlueHat Conference where back then microservices were all new. We were trying to govern them. We were trying to understand. We built up so much microservice tooling at Lyft because none of it had existed yet. What does your off story look like? How are you doing RBAC? How are you managing and governing all of these systems? It was the wild west. And so that's why Lyft invested so heavily in open source because there were no vendors for any of these solutions. There was no control plane. We built our own. This is where Envoy got built. And because of that, we're just making those primitives. I think that it's a very similar moment in AI right now because in AI you think, "Okay, well, there's no ..." AI agents are the new microservices. They operate on compute with a given context on identity. And then there's nothing to really govern the entire blast radius or control plane about what an AI agent can reach. And misunderstanding that blast radius, well, we just saw last week or recently about with OpenAI and the Hugging Face where the ... It's a disaster.

John Furrier

>> Yeah.

Alex Chantavy

>> So if you misunderstand that they're building those governing control planes and understanding like, "What are the standards?" And this is all yet to be hashed out, but my main thesis here is everything that is old ... Well, everything new ... Old is new again.

John Furrier

>> Yeah, old is new again. It's the same pattern, but it's a different demographic too because a lot of the AI native developers aren't ... I won't say plumbers, that's not the right word, but cloud native was also a slower build. I mean, a lot of brute force, self-building, Uber, Lyft, Intuit, Netflix. These are the companies that said, "Hey, we did this because no one else did it." That went into the pile and open source. That lifted all the boats up. And that's why I think CNCF was so popular. I think AAIF's going to be the same way. I'm just trying to figure out, because everyone's trying to be the control plane right now. Every company, like, "We're the control plane for AI." Okay, another one. Maybe that's not possible. Maybe that becomes more open like what Kubernetes was. I mean, Kubernetes, I remember they tried to do a venture-backed company around Kubernetes and no, no, that's orchestration layer. No, you can't. No one can own Kubernetes. So I think we might see a control plane that's more open and stable, boring.

Alex Chantavy

>> I hope that it plays out that way. I think that right now there's a war for who is going to own agent identity, for example. And then right now, I really hope that it gets turned into the standards bodies instead, like your IETFs and whatnot. 'Cause again, the agent operates on storage, operates on running on compute as a context of a given same-

John Furrier

>> It's like giving someone access passwords. They go play. Straight off the reservation. Like, "What are you doing over there playing around the wrong area of the map so to speak?" How do knowledge graphs fix all this? Because what's coming out is, again, the lineage, the traceability, the observability is actually pretty phenomenal. I mean, you mentioned the explainability is all there. You can actually go to your boss and say, "Hey, here's a reason why we're taking these priorities versus these 20, and here's the reasons why." And you have all that data. Now, the alternative would've been grind through some queries. I mean, how much work would've gone into that? I mean, scope the way that the alternative, no graphs. How do I justify the 20 top things? I mean, you prioritize, I see what your value prop and you can prioritize and then give context to the opportunities that you have to nail down or challenges. What's the alternative? What would it take to do that? Just kind of riff that in your mind. Share the thought process, assuming we're on a team together. Tell me why.

Alex Chantavy

>> Select star from users, join on computer select, join on blah-blah-blah, join on blah-blah-blah.

John Furrier

>> 2,000-line join statement.

Alex Chantavy

>> Basically. Sure. I mean, that's tractable up to a given point, but then-

John Furrier

>> Yeah, then people quit like, "Where did he leave off?" It's not a fun job.

Alex Chantavy

>> And we've literally seen that happen from firsthand. Yeah. So I think that the value proposition of graph, especially for this case, as I was mentioning, you have your agent running on some compute, you have an ontology to exploit that, those general concepts, and map out the entire blast radius. You want to represent those many to many relationships faithfully. You just can't really do that in relational land.

John Furrier

>> Yeah. And also, you can actually ... It's smart for the security teams too, because then they can actually look at the blast radius, segment the network, quarantine, and do all kinds of segmentation quickly on policy.

Alex Chantavy

>> In this case, a principle of least privilege is even more important than it ever was before because the agents, they operate at machine speed.

John Furrier

>> Alex, thank you for joining our GraphTalk here on theCUBE. Appreciate it. Thanks for coming on. Congratulations on all the great work you're doing. And again, it's just the wild west again. Welcome to the party. Come on in. Everyone else is joining too. It's going to be very popular. Thanks for coming on.

Alex Chantavy

>> Thank you, John.

John Furrier

>> I'm John Furrier at theCUBE. We're at GraphTalk in San Francisco. We got one in New York City. I'll be there as well. This is a very fast emerging area, graph databases, whatever you want to call it, this layer of semantics, harmonization. The data feeds the AI and the computer science match between graphs and AI. Super compatible, recursion. Everything's in line. The semantic layer all kind of happens. And again, this saves a lot of time and opens up the freedom to prioritize and do all kinds of cool things. Doing our part to bring you the coverage here with theCUBE. Thanks for watching.

Alex Chantavy, Subimage.IO

John Furrier

Alex Chantavy

>> Oh, thank you so much for having me.

John Furrier

Alex Chantavy

>> About time.

John Furrier

>> It's about time. It's kind of like relief and freedom from in the database world. I mean, why is Neo4j popular right now? Why are graphs so important in your opinion?

Alex Chantavy

John Furrier

Alex Chantavy

John Furrier

Alex Chantavy

John Furrier

>> You got to feed the agents.

Alex Chantavy

>> Yes.

John Furrier

>> So the feeding the agents becomes like school.

Alex Chantavy

>> Yes.

John Furrier

>> It's a crash course instantly. So is that data pipelining? What's the process? I mean, is it to get the graph, first-party relationships mapped out and then feed data?

Alex Chantavy

John Furrier

Alex Chantavy

John Furrier

>> And they have reasoning too. They can actually go in and reason. Okay, so talk about your company. History, where are you guys at? Phase, status. And what was the origination story?

Alex Chantavy

John Furrier

>> Word gets out fast, believe it. Something's really good if it gets out fast. Okay. What happened next?

Alex Chantavy

John Furrier

>> And how's that going right now? How many employees? How many people are on the team? Funding? Give us an update on the numbers.

Alex Chantavy

John Furrier

>> And what's the culture like? If you had to describe the culture, say I want to go work there on a potential higher way, what kind of culture do you have? What's the culture like?

Alex Chantavy

John Furrier

>> And you're a small team, so you can keep that DNA. And now, you're using agents to scale, sounds like.

Alex Chantavy

>> Yes.

John Furrier

Alex Chantavy

>> Yes.

John Furrier

Alex Chantavy

John Furrier

Alex Chantavy

>> I'm extremely practitioner focused. I've been in the trenches for a very long time and my answer is likely going to be extremely merity.

John Furrier

>> Yeah, that's okay. That's okay. We like that.

Alex Chantavy

John Furrier

>> And I heard scale too is another big point with Neo.

Alex Chantavy

>> Yes.

John Furrier

Alex Chantavy

John Furrier

Alex Chantavy

John Furrier

>> It's an architectural exercise really.

Alex Chantavy

>> Yes.

John Furrier

Alex Chantavy

John Furrier

Alex Chantavy

John Furrier

Alex Chantavy

>> Yes.

John Furrier

>> And now, cloud native is the foundation of AI.

Alex Chantavy

>> Yes.

John Furrier

>> Especially in large scale enterprises. So what's the product you guys have? How do people engage? What is it? What's the value? Give a plug.

Alex Chantavy

John Furrier

>> Yeah.

Alex Chantavy

>> When you're a defender-

John Furrier

>> You're always losing.

Alex Chantavy

>> You're always losing.

John Furrier

>> There's a lot of L's on the scoreboard.

Alex Chantavy

John Furrier

>> The explainability becomes a huge deal. This is where graphs shine, in my opinion.

Alex Chantavy

>> Yes.

John Furrier

>> All right. So you mentioned Lyft. Matt Klein and Vicki Chung had been on theCUBE. Vicki then wound up going out to co-chair.

Alex Chantavy

>> Amazing.

John Furrier

>> That's one point. And Matt had been a legend.

Alex Chantavy

>> Yes.

John Furrier

Alex Chantavy

>> Yeah, that's the equivalent of CNCF, right?

John Furrier

Alex Chantavy

John Furrier

>> Yeah.

Alex Chantavy

John Furrier

Alex Chantavy

John Furrier

Alex Chantavy

>> Select star from users, join on computer select, join on blah-blah-blah, join on blah-blah-blah.

John Furrier

>> 2,000-line join statement.

Alex Chantavy

>> Basically. Sure. I mean, that's tractable up to a given point, but then-

John Furrier

>> Yeah, then people quit like, "Where did he leave off?" It's not a fun job.

Alex Chantavy

John Furrier

Alex Chantavy

>> In this case, a principle of least privilege is even more important than it ever was before because the agents, they operate at machine speed.

John Furrier

Alex Chantavy

>> Thank you, John.

John Furrier