Building An Agentic Operating System For Cybersecurity And Beyond

Agentic AI is no longer a side experiment running in a lab. It’s starting to look like the next operating layer inside the enterprise, and that raises a hard question: do you want a scattered collection of point tools, or a standardized agentic operating system that your teams can actually run?

I sit down with Anurag Gurtu Chief Executive Officer @ AIRRIVED to unpack what “agentic OS” means in practical terms. We get specific about the three pillars that make agentic AI useful at scale: adapting a language model to your enterprise data, adding deep reasoning so it can synthesize and rationalize like a real analyst, and then deploying autonomous agents to take controlled action. We also dig into real enterprise cybersecurity needs across security operations, identity management and governance, risk, compliance, vulnerability management, and the growing challenge of shadow AI.

We zoom out to the messy reality of adoption: too many pilots, too many vendors, and too many tools designed for developers instead of practitioners. Anurag explains why objective-driven automation beats brittle playbooks, why governance and auditability have to be built in, and how fast proof-of-concepts can turn “AI hype” into measurable ROI. We also touch on open source momentum and why Arrived is building in a more secure, governed direction with Etherclaw.

If you’re building an enterprise AI strategy, leading a security program, or trying to prove value beyond demos, this conversation will sharpen how you think about standardization, productivity, and control. Subscribe, share this with a colleague, and leave a review with the biggest hurdle you’re facing in adopting agentic AI.

Transcript

Why Agentic AI Now

SPEAKER_00 0:02

Hey everybody. Fascinating discussion today. As we know, Agentic AI is quickly moving from buzzword to operating model. And Arrived is positioning itself right at the center of that shift. Anurag, how are you?

SPEAKER_01 0:15

I'm doing really well, Ivan. How are you?

Anurag’s Path To Arrived

SPEAKER_00 0:18

I'm well. Thanks so much for joining. Really exciting times. Maybe tell us about your background and journey to Arrived. And what's the big idea? How do you describe the mission these days?

SPEAKER_01 0:31

Yeah, so I had been uh in the cyberspace for over two decades, and I would say over a decade in the AI space. I moved into Classical ML back in 2013, was acquired by Splunk. And then since then, I have been uh in the AI space doing my uh first NLP company in 2017 and then doubling down on NLP again with a second company in 2019, uh, doing that for about five and a half years. The vision of Arrived is to democratize AI and make it accessible uh to everyone in an organization. Uh as my journey was progressing through classical ML towards NLP, towards NLU, um, towards generative AI and then agent tech, I could see the potential of what AI has to offer. Um, even though there are some doubts of what it can do and what it cannot do. But if you're a really believer in AI, the potential is tremendous. And then the vision was very straightforward is how can we democratize AI so that everybody can harness the power and solve different business problems? Now we open up the market with cybersecurity as a first vertical. Um, and we have apps in the security operations space, in the identity management space, identity governance space, third-party risk, shadow AI, um, vulnerability and exposure management, um, audit compliance, and so on and so forth. But the platform is designed uh for customers to connect the dots, be creative, and solve many different problems, not only in IT, but beyond uh cyber and IT and beyond.

Fine-Tuning, Reasoning, Autonomous Agents

SPEAKER_00 2:12

Brilliant. Well, we'll we'll dive into all those use cases. Uh you describe at a high level arrived as a Gentec OS. Uh, what does that look like inside an enterprise in a practical sense?

SPEAKER_01 2:27

Yeah, so we are uh we are positioning ourselves as an agentec operating system. Think of it like how VMware came out with ESXi to democratize and leverage uh full-blown potential of the underlying hardware, CPUs, and memory uh and I. They all are trained on external data. They're not trained on enterprises data. So uh the first thing that an enterprise would maybe want to do, or the last thing that they would want to do, is they want a personalized language model on their own entire data set. So they want to adapt the language model to their data. Uh and the second thing that they would want to do is once they have adapted, they want the system to think and respond and rationalize on that data. So if there is this question, they want to be able to not just answer on it, but look at all the data points and reason on it, just like how a human or a practitioner or an analyst or a business leader would do that. Like you get all this data, and now you think about it and you come back with how do we solve a certain problem. Uh, so that's where our second element within our platform kicks in, which is our deep reasoning systems. So the customers not only can fine-tune language models to our system, they can start building deep reasoning systems. And they can build these deep reasoning systems in different aspects of cybersecurity or IT. They can build a deep reasoning system for your security operations team team, they can build a deep reasoning for the identity management team, they can build a deep reasoning for their governance team, their risk team, their compliance team, and so on and so forth. And the last thing that you typically want to do is run automation, right? You you have the data that is adapted to your knowledge, you have a system that can think, but then you have to act, right? You have to take that final set of action, that closed loop system. And this is where you can build autonomous agents on a platform. And using these three core pillars uh of fine-tuning, building deep reasoning systems, and building autonomous agents, uh, you can create and solve different problems by either leveraging our pre-built apps or composing your apps uh from crowd up.

SPEAKER_00 5:11

Interesting. So, as you know, so many companies are experimenting with tons of different AI tools, maybe uh too many uh in-house. Um, do you see yourself consolidating some of those, or what what other problems uh can you solve that maybe existing platforms aren't solving?

SPEAKER_01 5:31

Yeah, they are uh they are, I would say, a few problems that uh enterprises uh typically run into that we help solve. Uh first, uh organizing a lot, a lot of organizations have AI initiatives and they're focused on developer-centric tools, right? And a lot of developer-centric tools are really amazing if you look at cursor or co-pilots and all those technologies, like really powerful technologies, right? But think about um the persona, the business persona. Think about your your security team, think about your IT team. They don't have developers, they have practitioners, right? They have architects. So they're not, they cannot write code and they are not expected to write code. So they need a solution which can amplify their productivity, right? Uh, similar to how the productivity of developers gets uh amplified using uh clot code or or cursor like technology stacks, right? Um, so the second aspect is uh if you are moving away from these developer-centric tools to practitioner-centric tools like ours, uh, to systems where you do not need uh highly experienced skilled people to operate AI. First, because there's a massive shortage. And uh if there's a talent, then I'm sure Meta and OpenAI are doing these crazy bidding to hire these people. So uh a lot of organizations are pretty much struggling to hire that talent. Uh so the question is how can you still leverage the full potential of AI? Um, you need to pretty much uh expose the technology in a way that a layman can can do, can use, right? So that's the second system that where our platform comes in is well, we don't expect uh developers to use our system. We expect business leaders, business practitioners, architects to come in onto our system, but we make it so easy for them to use that uh they don't have to crunch through and and wait for a specific skilled person to be hired to operate a technology uh to harness the potential of AI. And the last thing that comes in is automation-based tools. There are a lot of automation-based tools that are spinning and saying, well, we can do, we have been doing automation uh RPA systems uh for many, many years, and now we will throw in agents to the mix. Uh, but that's uh not really solving a problem. Uh they're trying to uh uh leverage their uh, I would say an outdated legacy architecture. Uh the older way of doing things uh was building flowcharts. Uh the newer way of thinking is not building a flowchart at all. It's just giving an objective and letting that objective execute uh in a controlled, in a governed, in an audible manner. And that's what our uh system helps them with. It's uh uh helps them move away from uh writing these uh complicated playbooks, uh setting up a team who would do that, uh figuring out if your technology stack has changed, changing these playbooks and so on and so forth, right? Uh so spending years to build them out, uh spend uh years and a lot of money to manage and maintain them versus going with a new way of automation, right? Where you don't even build playbooks at all. So those are the core things of how we uh help customers.

Standardizing A Fragmented Security Stack

SPEAKER_00 9:04

Brilliant. Let's talk security on the heels of RSAC. Top of mind to many of us. You're looking at unifying many different domains, SOC and identity and and IT ops and et cetera. Uh, how hard is that to pull off in an enterprise that, gosh, a bank might have 60 different vendors, uh, lots of legacy environments. How do you uh add value to that kind of mix?

SPEAKER_01 9:30

I would say uh every uh executive um has run into that problem of fragmented technology stack. And uh and you can see uh a lot of uh very established and and big players are moving in the same space, be it Cisco, be it Palo Alto, uh be it CrowdStrike uh or Zscale or saying, guys, we are the platform that you need to standardize on. So you can see there is headwinds in that direction. The question then boils down to uh do we see standardization uh beyond that? And we do see that too. Like if you look at uh your IT service management, then ServiceNow is pretty much the standard, be it uh the ticketing system being used by your customer service team or your developers or your HR team or finance team, they are on one standard platform. Uh be it Salesforce as CRM, they are on one standard platform. Uh if a customer is looking for uh routing, they are on one standard platform, Cisco. If they are looking for switching, they are on another standard platform, maybe Orista or something else. So you can see standardization across the board. The question is very simple. Do you want to run a fragmented AI strategy or do you want to standardize on an agent tech system? And we position ourselves as that agent tech operating system that they can standardize on. So that's how we um uh have a conversation with our customers. Uh, we align to their standardization philosophy, and um, and that's how we lead in the market.

Adoption Hurdles And AI Skeptics

SPEAKER_00 11:04

Brilliant. And so my little business, I'm running three different agents as we're speaking. Uh, brilliant. I mean, it's just game-changing for me and my business. Of course, I'm not a big enterprise with all the issues that uh have to be dealt with. So, what are what is the biggest friction in the enterprise adoption of agenc? What hurdles do they need to overcome?

SPEAKER_01 11:26

I would say there are still two schools of thought um that we uh experience. Uh folks which are very pro uh to AI, uh Gen AI, Agentic AI believers, uh, and other set is uh folks that uh just think uh it's not real. There is uh there's no, it's just like uh it's a hype and there is no substance behind behind the hype. Um I think uh for the folks who uh are believers, it's a very easy conversation. The conversation then moves towards uh a very simple question. Um, do you want uh to go towards the standardization route or do you want to go through this fragmented strategy? Because if you have 80 products, point product solutions for doing different things with this agent tick wave, your count is going to go up from 80 to maybe like 160 or 200, right? Or so on and so forth. So it boils down towards uh very simple discussion about standardization. Um and uh that's about it. Uh, the folks who are not believers, uh, it becomes very difficult for them to convince uh themselves. I think they have to self-convince themselves. Um I I would say I just joke around as uh I was very young uh during the uh the internet creation days, like the 1990s. So I'm sure there were two schools of thought running at that point in time where folks, uh some folks were pro-internet and they said, yeah, this is the this is gonna change the world. And there were some people who said, no, this is all uh this is all noise. There's nothing real behind it. So we know with side one. I am on the believer on the AI side as well that this is uh, I think the I would say the industrial revolution of our of our century.

Fast POCs And ROI Proof

SPEAKER_00 13:12

Um that's a great way to put it. I think I'm gonna steal that from you. That that's a great uh message, I would agree, just on my own personal firsthand practical experience working with these agents and others. Um one challenge the industry has, again, is is uh there's so many project science projects, as if you will, trials happening. It was the Gartner uh summit last fall. One bank talked about 75 different pilots that they're running. Uh this was the uh hopefully some of those became came into production. How quickly can you deploy and and make an impact? Are we you know, if these things take a year, multiple quarters, it kind of loses the uh immediate ROI.

SPEAKER_01 13:58

It's it's very quick uh with our platform. Uh and I think um I think even uh in your experience as you're using these agents, you can see the value instantly. Is there value or is there no value, right? Uh it's a very similar thing in the B2B space too. When you go into a very specific team, be it identity, be it SecOps, be it Threat Intel, um, they know what the outcome is, right? Uh so the proof is are we gonna get to that outcome fast? And is it gonna beat your expectation? Because they're expecting a certain outcome. And the question is not about uh can you beat my expectation on what I was thinking I would get. Um so the POCs are extremely fast, the value can be realized very instantly. Um and if the customer relates to the value, then your your buying cycle also is actually accelerating. Um and which is very obvious if you if you look at how the how the new age AI companies are accelerating to 100 million or beyond in ARR, they are beating the numbers of what SaaS companies do. The benchmarks were set to five years, it was set to seven years, it then moved down to five years, hitting 100 million ERR. And now you're seeing these AI companies are hitting 100 million ERR in in about 18 months or so. Uh so there is real value that is being delivered at a very fast uh rate.

Open Source Pressure And Etherclaw

SPEAKER_00 15:26

Fantastic. And you know, the open source momentum is so exciting. I haven't configured OpenClaw yet, but that's been just a phenom, the uh fastest growing open source project ever, including Linux, and as you kind of mentioned. Um, how do you see that playing into your strategy? And are are you planning to open source your solution?

SPEAKER_01 15:48

We uh we not yet. We are not planning to open source uh our our solution uh yet, uh, but uh as you saw with OpenClaw, it really showed the potential of what AI was designed to do. And uh very similarly, we were architecting a very similar technology stack uh and we launched it at RSA called Etherclaw. It was uh a very similar technology from a concept perspective where you can give very complex objectives and you can get a mix of agents uh and API calls or MCPs uh execute that objective, but under full governance, uh compliance and auditability built in. Um so not yet to open source our technology, uh, but we we do are building things which are in the open source world just in a more secure and governed fashion.

SPEAKER_00 16:45

Makes sense. So if we were to fast forward a couple of years, do you have a vision of what a truly open uh agentic enterprise looks like in practice? What what do you see your vision uh uh looking like in practice and in practical terms?

SPEAKER_01 17:01

Yeah, I I would say this agentic uh journey that these organizations are going towards is I think the right bet that they would have to uh make and they are making. Um and they would uh definitely see massive productivity boost, and they are already seeing a lot of productivity boost and efficiency gains. But I think the conversation will uh, and there's not much to prove over there, that's already proven. I think it will boil down to really quantifying the ROI investment, right? For every dollar that they're investing. The question would be how many X number of returns are you getting for that from that agent tick investment that you did, right? Um, and that will boil down to a much efficient uh way of running an organization, right? Uh getting to IBITRS really faster, better outcome for the investment firms, uh, and so on and so forth, including your customer base. So I can see the journey, not just uh being in cyber, the journey will uh be in IT, be in HR, in finance and our operations, and pretty much every aspect uh of an organization through this agentic transformation.

SPEAKER_00 18:14

Brilliant. Can't wait to see that come to reality. What are you excited about over the next few weeks, months, as you roll out your product, any releases or maybe events? What's on your radar?

SPEAKER_01 18:25

Yeah, we are uh very excited. Uh, and we are about to power the first uh data center with our operating system uh to make uh uh GPUs uh accessible uh in a way that uh has never been experienced before. So that will be uh one of um the first hyperscalers uh in the world that will get powered using our operating system uh in very, very soon. So we are super excited about that.

SPEAKER_00 18:54

Well, I can't wait for you to come back and talk about all of the goodness coming from that project. Thanks so much for joining. Really learned a lot. It was a pleasure.

SPEAKER_01 19:02

Uh have an amazing time.

SPEAKER_00 19:05

Thank you. And thanks everyone for listening, watching, sharing this episode or post. And be sure to check out our TV show, techimpact.tv, which is on Bloomberg Television and Fox Business. Thanks, everyone. Thanks, Henry.

SPEAKER_01 19:16

All right, thank you so much.