Name: Victor Jakubiuk, MisaLabs & Shiv Agarwal, Singulr AI & Sri Viswanath, Sycamore
Uploaded: 2026-09-29T17:50:00.000Z
Duration: 12 min 59 s

Clips
More from Dell Technologies AI Leadership Symposium 2026

Victor Jakubiuk

Founder

MisaLabs

Shiv Agarwal

CEO

Singulr AI

Sri Viswanath

Founder & CEO

Sycamore

play_circle_outline Enterprise AI platforms enabling GPUs/models/tokens-as-a-service and custom internal apps

play_circle_outline Discovery, Governance and Scaling of Enterprise AI: Agentic Apps, Distributed Deployments, Centralized Control Plane, Adoption Challenges

play_circle_outline VPC Sandboxing, Cost Controls, and Runtime Monitoring vs Static Governance for Autonomous LLM Agents

16:06

11:00

10:43

Info
Transcript

Victor Jakubiuk, MisaLabs & Shiv Agarwal, Singulr AI & Sri Viswanath, Sycamore

Victor Jakubiuk

Founder MisaLabs

Shiv Agarwal

CEO Singulr AI

Sri Viswanath

Founder & CEO Sycamore

In this interview from the Dell Technologies AI Leadership Symposium, Victor Jakubiuk, co-founder at MisaLabs, joins Shiv Agarwal, co-founder and chief executive officer of Singulr AI, and Sri Viswanath, founder and chief executive officer of Sycamore Labs, to talk with theCUBE's John Furrier about the shift from AI experimentation to production-grade enterprise infrastructure. Viswanath notes that Fortune 500 CIOs are eager to move fast but lack a clear path to finish the "last mile" of deployment. Jakubiuk describes how MisaLabs' software layer connects GPU... Read more

explore Keep Exploring

What does your company (Singulr AI) do and how does its platform help enterprises manage and govern AI? add

How are enterprises currently adopting AI (including agents), and what governance, management, and infrastructure challenges are they facing? add

Are organizations effectively doing "test/dev in production" with AI agents and LLMs, and should enterprises require those agents and LLM calls to run inside a private VPC/VPN sandbox to maintain governance and data control? add

bolt Powered by CUBE AI

Victor Jakubiuk, MisaLabs & Shiv Agarwal, Singulr AI & Sri Viswanath, Sycamore

John Furrier

>> Hello, I'm John Furrier, host of theCUBE. We are here in Silicon Valley, actually in San Jose, as part of Dell Technologies Symposium. There's going to be a New York Symposium on October 12th, and that's going to be a lot more financial. It's about startups, it's about the innovation in Silicon Valley. We got 3 great startups here. Victor with MisoLabs, good to see you, thanks for coming on. Shiv from Singulr AI, been on, been a friend of theCUBE, and Sri with Sycamore, both covered on SiliconANGLE. Victor, we'll get to you first as we haven't covered you on SiliconANGLE. What are you guys doing? What's the focus?

Victor Jakubiuk

>> MisoLabs provides an enterprise AI platform that enables enterprises to deploy GPUs as a service, models as a service, tokens as a service, but also build custom internal AI applications.

John Furrier

>> And you're a NeoCloud?

Victor Jakubiuk

>> We're not a NeoCloud. It's just a software layer.

John Furrier

>> Okay, got it. Okay, we'll get into that. Go ahead.

Shiv Agarwal

>> Thank you. Thank you for having me today. So we are Singulr AI. Our tagline is we put enterprise in control of their AI. And we have a platform that does discovery. It discovers all forms of AI, internal, external, embedded. So enterprises know, the leaders know what form of AI is running in the organization. We classify it as sanctioned and unsanctioned. And then we help put guardrails around it.

John Furrier

>> Governance.

Shiv Agarwal

>> Governance.

John Furrier

>> We'll get into that. We've had conversations. Sri, your company.

Sri Viswanath

>> I am Sri Viswanath, founder and CEO of Sycamore Labs. We build an enterprise AI platform to be able to build and govern enterprise-grade agents. And we call it agentic apps, which is modern kind of apps which has agents and has traditional applications too, and cross-departmental workflows.

John Furrier

>> Yeah, you guys are a great example of the evolution we're seeing in AI factories, AI in general. NeoClouds have been a great use case. We've seen the rise of CoreWeave. They got an event next week. We'll be there. Nebius, Nscale's going public. So these are now the new hyperscalers. Great. Now the enterprise realizes that, hey, I can get in this game too, but it's distributed computing. They don't have to spend all this CapEx. I could just get an AI factory, put it on-prem, use a little NeoCloud, a little hyperscaler. So you're starting to see the formation of the architecture. What are you guys seeing now in the enterprise adoption? Are they leaning in? Obviously coding opens the door. Agents, they're running wild. So we know that. We know people are blowing their token budgets. So there's uptake for sure. There's definitely people who love the AI. They want more of it. What are you guys seeing in the enterprise?

Sri Viswanath

>> They are still early. I talk to Fortune 100, Fortune 500 companies, CIOs and CTOs. They're all thinking about AI, they all want to move fast, but they don't have a mechanism to finish the last mile. And their only option right now is to try to prioritize and reskill their existing team, or go with consultants, or try to figure out if there is SaaS, but there's no SaaS that has been built recently that actually solves their problem.

John Furrier

>> Yeah, they had their fraud detection, they got machine learning playing around, but Shiv, talk about the governance piece because, they're all adopting data lakes. We saw Databricks is doing great, so good for them, but that's just the beginning of the story. What are you seeing around understanding, okay, now I got my data, I got to watch it, I got to understand the movement?

Shiv Agarwal

>> Yeah, no, great question, and it goes beyond the data. What we are seeing in enterprises is the right model, the right provider, the right platform for the right use case. It's a very heterogeneous environment and that's where our value prop comes in or the governance value prop comes in. When public cloud happened, there were 2 or 3 big providers, right? Even then, people needed consolidated, centralized management for security and cost and all that. Case of AI, every enterprise is running 5, 10, 15 different kinds of agent platforms or providers, right? They come with their own set of controls. And then we have the security layers, all the big mega security incumbents providing controls. There's a lot of controls there. How do you know those controls are working at the end of the day? Someone needs to really do some rationalization between the enterprise intent and the actual runtime state, right? So governance is becoming very important, but it's becoming hard as well in this world, just the pace at which the AI is getting adopted.

John Furrier

>> I want to come back to that control plane or that control layer around context. But Victor, first I want to ask you on the software stack, And this comes back to the software stack. 2 years ago when AI factories were kind of getting into gear, we were scratching our heads, I think, what do you run on these things? You don't just load Linux. It's like there was really no maturity or even formation of what the hell's going to run on it. Now you're starting to see visibility into things because of the agents. You're starting to see things like context layers, semantic layer or control plane. Those are words too we've kicked around, but these concepts, what's your view on this?

Victor Jakubiuk

>> Yeah, absolutely. I think we're still very much in early innings of this process, of this development. A lot of companies today have deployed the chatbots like the ChatGPTs of the world for their marketing teams. Of course, the coding tools for their software engineers, and those are great tools. But oftentimes then they start thinking, okay, what's next? How do I control that? How do I, assess the ROI on even the relatively basic tooling that's already deployed today. But taking that next step, how do I go from just the chatbot, just the coding tool, to something much more specific to my enterprise, to my business? That's one. And second, how do I then connect it back to that infrastructure? Sure enough, you can buy the GPUs where it's GPU as a service by the hour from an expensive cloud vendor. Maybe on-prem in a more cost-efficient manner, but still connecting that infrastructure back to something more than just a coding agent, something more than just a chatbot, I think, and we're early stages. The software stacks are evolving very rapidly. The memory context. Yeah, it's just the beginning, right?

John Furrier

>> You know, it's early. I agree. We all agree it's early. You know, it's early when you have the kind of experimentation. I was talking with a friend of mine, I won't say his name, He works at Stripe, and he said, John, the AI craziness is so crazy. They're running essentially test dev in production. I'm like, what do you mean? Well, they're using tokens. They're paying for tokens. I go, I'm not following the bouncing ball. Slow it down. He goes, that's like Stripe running our test dev in our production payment rails. I'm like, oh, I get it now. So what he's saying is that a lot of the stuff that's going on now is They're using tokens, they're paying for them. But he's saying that there should be an inside-the-wall sandbox. To your point about governance, you don't yet know what's going to break. We saw that with all the early agent days. What do you guys think about that comment? Is that something that's just a part of the evolution? Do you agree with it? Are you seeing similar things?

Sri Viswanath

>> Yeah, we are seeing a lot of CIOs asking for all of the AI agents running inside their premise. In their VPC or VPN. And Sycamore Labs, we support inside their VPC. So for them, not even just the agents running, the LLM calls also, they don't want data going out to LLM providers. So they're also saying, can the data stay within the VPC? And there's a lot of momentum.

John Furrier

>> Yeah, so they're trying to understand the control layer.

Sri Viswanath

>> They're trying to understand the control layer. They're also trying to architect in a way that all of the data stays there and they don't trust sending it outside because then we don't know where it goes.

John Furrier

>> How are you seeing it, Sri?

Shiv Agarwal

>> We see all these different kinds of agent behavior in the organization where these boundaries that we tend to put around software are not working because the boundaries are not about what's allowed or what's not allowed. It's actually what at runtime the agent will do. As simple as, an action where it reaches out to the internet to get some data or sends out an email to a private account or something. That could be deeply hidden inside an agentic action, right? So it's moving away from deterministic set of controls and governance actions to actually observing runtime behavior and then really understanding whether what the enterprise intent is, is it happening as per that intent or not?

John Furrier

>> You said runtime, it triggers me because we know what that means, but aren't agents always running? It's runtime all the time. So because that's what's happening, they're out there and they can make decisions, touch resources. So you have this new state run all the time.

Shiv Agarwal

>> Run all the time, yeah.

John Furrier

>> Victor, what does it mean? You're smiling.

Victor Jakubiuk

>> Well, that likely requires the change to your business operational workflow too, right? Human employees only work 8 hours a day, not 24/7, right? You have to probably adjust your business workflow to take advantage of that. That's one. Second, you have to trust that agent, you have to trust that agent where that agent actually takes action. It's one thing to accept sort of the recommendation of the agent and have the human in the loop approve that. But it's a different thing if you're fully enabling that agent to make autonomous decisions 24/7 without that human oversight. That's where the security sandboxing safety are extra critical.

John Furrier

>> Well, we don't have a lot of time left. I want to get one more question, then go into what you guys are working on right now. I gave a talk on sovereignty, which has expanded the definition from country, cloud, where do you put the data, to basically enterprise. Enterprise AI sovereignty. So I had a quote, I said, "No one is asking whether to adopt AI, that's pretty much a done deal. They're asking who controls it, where it runs, and whose law it operates." Now the caveat is, take away the geopolitical for a second, 'cause, whether it's France or Canada or whatever, enterprises have laws. It's called IT governance. So you got a sovereign equation happening in the enterprise. So the sovereignty has kind of expanded from country, region to enterprise. What do you guys think about that and how would you describe that opportunity? Do you see an opportunity for an organization to look at sovereignty as their laws. What are their laws?

Sri Viswanath

>> Yeah, so every company had governance in IT for the deterministic world. Now they're trying to figure out what it is for the probabilistic world. So any software that they built, by the way, all software doesn't have to be probabilistic, right? People are realizing you don't want to be all probabilistic because you max out your tokens within 2 months and then you don't have anything left for the rest of the year. But They are trying to figure out if there are probabilistic agents running, then what is the law for that? So they're still trying to write it. And when we build our agents, we build it with some laws in place and then they figure out, okay, does that make sense? How does it align with the company's goals?

John Furrier

>> So you're in the governance business, it's your wheelhouse.

Shiv Agarwal

>> Yeah, I think what is happening is that every region has some local standards that they are following, the EU AI Act, the NIST RMF, the OWASP Top 10. And that is what is governing a lot of these guardrails and policies around it. But AI doesn't behave differently just because it is running in the US versus Europe, right? And so we need some level of standardization across these. That's what we are saying. It'll help solution providers like us as well as help the underlying enterprises.

John Furrier

>> Here's the last word, 'cause we're squeaking into the presentation now. Go ahead, your final word.

Victor Jakubiuk

>> Yeah, I think all these different enterprises have their own rules, right? And they want to follow their own rules. So having one standard governing body, whether it's the large closed-source lab or otherwise a government perhaps, is always going to be limiting, right? And as you said, AI has no limit to the speed of development, speed of progress. You want, as an enterprise, as that AI lead, you want to be able to take the latest and greatest and stay on top. And that's really hard if you have to follow just one very specific standard or

John Furrier

>> regulation.Guys, we'll have to bring you back into theCUBE. Now the presentation's going on. Congratulations on the funding. And congrats to Dell Technologies Capital for funding great startups. Thanks for coming on.

Shiv Agarwal

>> Thank

John Furrier

>> you, I'm John Furrier for theCUBE. We're kicking off the program here back at Dell Technologies Symposium in Silicon Valley. All the innovation happening here, the hot stars defining the next generations coming online, AI factories in the enterprise, Intelligence factories. And that's what the demand curve wants. And more and more software stacks emerging and the picture is filling out in real time. We're doing our best here. We'll be back after this short break.