theCUBEd Awards | Binny Gill, Kognitos

Clips
More from theCUBEd Awards

Binny Gill

Founder and CEO

Kognitos

play_circle_outline Kognitos wins CUBEd Award for most innovative AI product.

play_circle_outline From IBM Research to "English as Code": How Binny Gill Turned Teaching His Son into a Software Infrastructure Revolution

play_circle_outline Executable English SOPs: Deterministic AI Curation for Hallucination-Free, Auditable Processes in Regulated Environments

play_circle_outline Neurosymbolic approach: separate AI planning (creative) from execution (deterministic engine).

play_circle_outline Difference from RPA/process mining: exposes English "code" for business users, not hidden workflows.

play_circle_outline Interpreter asks humans on ambiguity, capturing tribal knowledge and exception handling.

play_circle_outline Governed LLM-Driven SOP Curation and Exception Resolution with 200+ Connectors for API and Developer Invocation

play_circle_outline Primary industries: finance, accounting, healthcare revenue cycle, manufacturing, logistics, retail, banking.

play_circle_outline Knowledge graph versus agentic/unstructured document search; depends on use case scale.

play_circle_outline Vision: empower creators, govern English business logic, be enterprise layer for AI governance.

Info
Transcript

Binny Gill, Kognitos

Binny Gill

Founder and CEO Kognitos

Kognitos presents a neurosymbolic artificial intelligence platform and the English-as-code approach during the 2026 CUBEd Awards coverage. Binny Gill of Kognitos, chief executive officer and founder, joins a theCUBE Research conversation hosted by Dave Vellante to discuss enterprise automation and governance.

Gill explains how Kognitos separates creative planning from deterministic execution, captures tribal knowledge and applies a governed and auditable execution engine to support enterprise automation. They describe treating natural language as deter... Read more

explore Keep Exploring

Why was Kognitos founded? add

What is your background and what inspired you to found Kognitos (including any experiences during the pandemic that motivated you)? add

How does your AI system (Kognitos) document and curate business processes, plan work with humans, and then deterministically execute those processes (including creating processes if none exist) while handling exceptions and avoiding hallucinations? add

How does a system combine a creative planning engine (generative AI) with a deterministic neurosymbolic execution engine to reliably run and document business processes, handle exceptions with human-in-the-loop decisions, avoid hallucinations, and keep operational documentation from going stale? add

Why is using natural language (English) as the basis for coding and automation preferable to visual, drag-and-drop, or RPA-style approaches? add

How can an AI system handle unforeseen edge cases during business-process execution by eliciting demonstrations from humans, capturing that behavior as “tribal knowledge,” and providing manager review/approval or deletion so the system can learn safely and stay up to date? add

How do you use LLMs and your platform (Kognitos) to turn business processes into executable SOPs while handling exceptions, maintaining governance, and keeping critical business logic protected (separate from low-code/“vibe coding”)—including integrations with tools like Claude Code and Cursor? add

In which industries or use cases is provenance and deterministic, auditable AI/LLM output particularly important? add

What prerequisites must customers meet (or what must they provide) to use your product/service? add

What is the future vision for Kognitos and what role will it play for enterprises in managing AI infrastructure and governance? add

bolt Powered by CUBE AI

Binny Gill, Kognitos

search

Dave Vellante

>> Welcome to this special CUBE conversation. This segment is part of our ongoing coverage of the CUBEd Awards, where we recognize the builders and the innovators. They're not just participating in the next wave of technology, but they're actively shaping it. And today we're spotlighting one of this year's winners, the Kognitos neurosymbolic AI platform, which won for the most innovative AI product category. And with me is many time CUBE alum, Binny Gill, CEO and Founder of Kognitos. Binny, first of all, congratulations, it was very competitive, and I really appreciate you spending some time with us.

Binny Gill

>> Thank you, Dave. We are honored to receive the award. I'm just so happy that I can be of use to the community and helping them figure out how to leverage AI in enterprise in production.

Dave Vellante

>> Yeah, you're doing some amazing work. And I just want to say, we have over 50 judges in these CUBEd Awards, and so as I say, it's very competitive. A lot of thought goes into it and so we really appreciate the submissions from everybody. And the winners, really, we'd like to call them out because of the excellent work that you're doing for the community. And we're going to talk about why separating AI, how AI thinks from how AI acts is very important for enterprises. But I want to start with the company. Why did you start Kognitos?

Binny Gill

>> Oh, Dave, so this was, previous to this company, I was at Nutanix, which you're aware is a cloud infrastructure company. I had a great run there, and prior to that at IBM Research for another eight years. I've been in the software space, but working in many different layers of the software. All the way at IBM working very close to the flash firmware and figuring out where leveling with caching algorithms to at Nutanix, looking at storage, then networking, and then finally looking at cloud and getting into AI at that point. It was the pandemic time that actually gave me the impetus to go start Kognitos. I started Kognitos about five years ago. And during the pandemic, there was an episode where the schools had all closed. My son was 12 and he was trying to learn Python. I'd encouraged him like, "Please go learn Python because that's all I do programming." And I wanted him to be a good programmer as well. And it's like, "Dad, why should I do it?" I said, "No, you should do it because you can make games with it." And he was excited, intrigued. And two days later, he comes back and shows me that, "Hey, look, play this." And it was tic-tac-toe, and I was super proud dad. Here's my 12-year-old walking on my same footsteps and he's going to be a programmer one day just like me. And I played it, and I slept that night. And Dave, I woke up the next morning realizing I had made the same game 30 years ago.

Dave Vellante

>> Yes.

Binny Gill

>> And-

Dave Vellante

>> That was always the first program we ever wrote was tic-tac-toe.

Binny Gill

>> Tic-tac-toe.

Dave Vellante

>> This game, always.

Binny Gill

>> And I wrote it. I didn't have a proper personal computer 30 years ago when I was 12 also. I had a graphing calculator. It was a Casio handheld PB-1000. I remember the model number as well, eight kilobytes of memory, and it would do basic programming along with graphing. And tic-tac-toe is basically lines and Xs and Os so you can build it. And I realized that there was no internet when I built it. And I built it in a couple of days. I went back to my son and said, "Did you Google it or did you build it from scratch?" He said, "Of course I Googled it. I cut/paste it." I said, "That's cheating. I did not have that. I built it from scratch. I looked at the manual and let's go do that. Let's start with the manual." I gave him a Python book and say, "Let's go build a program." And I gave him a next common one, find out if a number is prime or not. We've done that all the time. And he is drawing a blank. I said, "No Googling, no YouTube." He said, "I don't know." I said, "Okay, let me teach you something." There's something called pseudocode, meaning just explain the logic as if you're explaining it to your younger sister, right? How do you figure out if a number is prime or not? He said, "Dad, that's easy. If it's a one, it's not prime." Why? Because it's convention. Okay. "And if it's divisible by any number from two to the squared of the number," and he made me proud when he said that, "square root a number, then it is not prime, otherwise it's prime." I said, "Go write it down." So we wrote it five minutes done. And then after that, we started converting one line at a time into Python. The first line took 20 minutes if you have to understand the syntax and other things. The next one, the loop was very hard. Teaching somebody how a loop works where the same line executes multiple times with different meaning that does not happen naturally. When you're reading text or reading novels, you don't read the same paragraph and multiple times and it means different things every time. And that's where I was getting frustrated and he was getting frustrated and we got into a fight. Basically he said, "Dad, I want to go play basketball. I'm not doing this." And I'm like, "Oh." And I said, "No, no, no, no. Sit down. I want to solve this problem for you. What is so hard?" I had forgotten what's hard about programming. And he said that, "It doesn't make sense. Why can't it be like Alexa?" I said, "What do you mean?" He said, "Alexa understands English." And this is five days ago before ChatGPT, five years ago, before ChatGPT and English. I didn't have an answer. I said, "Maybe he's right because Alexa is doing one line. This is five lines. How hard can it be? " So I just said, "Okay, you go. You don't need to learn Python." And that's how this company started. I wanted to take English logical instruction that anybody can come up with, and make that the program. English must be code, and that's how we started. And that was five years ago, and now we are sitting at a point where that is reality. Now, there's a whole lot that is going on in the world. There could be English as prompt, there could be English as just guidance, but here we are talking about English as code, deterministic. You mentioned neurosymbolic. That's what Gartner refers to this as it is English, but it runs as if it was deterministic like Python. It cannot hallucinate. And that's what I wanted to build and now we are there.

Dave Vellante

>> At the time, were you plugged into what was happening at DeepMind and transformers and diffusion models and the pre-chat heard around the world? Were you plugged into that or were you in an infrastructure mindset? I mean, were you-

Binny Gill

>> I would say luckily not.

Dave Vellante

>> Why do you say luckily? Explain that.

Binny Gill

>> Because see, if you know that AI can actually start understanding English and execute it in its mind, then that is the easy way out. That's what we would've done, that's what I would've done.

Dave Vellante

>> Right.

Binny Gill

>> And we would be in a point where we would be worrying about, yes, it does it most of the time, but sometimes it hallucinates. Have we ever said that Python hallucinated or Java hallucinated?

Dave Vellante

>> I always did when I sent it out to you. It was always my mistake, not its mistake.

Binny Gill

>> Exactly, right? So I did not have that in mind, but I started off with from first principles. I looked at what are the projects that people have done in the past to make English a programming language. One of the professors who wrote book, Donald Knuth, I deeply respect his books and his work and all of that. He had also tried. He called it CWEB. It didn't quite work. And I was like, there are other people. In fact, COBOL was common business oriented language. It was supposed to be as close to natural business language as people speak in business. In fact, if you think about computers were invented because humans made mistakes. And where was it applied? In payroll and finance where you had to do things logically in a deterministic fashion. And therefore, I just wanted to take natural language English and make it the programming language. Now, I know my professors taught me at IIT Kanpur and UIUC that English is a bad language for programming. And one day on a walk, I realized that the reason why English is bad for programming is not because of the language itself, it's because of the interpreter of the language. I'll tell you the example. Say a recipe for making apple pie, 10 steps, maybe grandma wrote, somebody's grandma wrote it. And then those are 10 steps. If you follow that, you get apple pie. That's a program. It's written in English, but there's a difference between that program and a Python program or a Java program that I write. Computer programs keep on bloating over time. The recipe does not bloat over time. If somebody runs out of sugar, nobody has to tell grandma to fix the code because we found a bug. And then she writes, "Okay, please go to Trader Joe's." And now that Trader Joe's closes down, then I have to say, "Oh, no, no. Now it's a different way to handle that same problem." So things get loaded and loaded. I realized that we are missing something big in computer science. Natural language can be ambiguous, can be incomplete, yet it works when it runs on the human mind. When we are following process, a business process before automation comes in, is step-by-step instruction. How do I pay for invoices? How do I go through procurement, order to cash, procure to pay? All of those things are processes that humans implement and we are able to handle ambiguity in completeness. The reason is our interpreter knows that whenever there's ambiguity, ask, ask a human, ask for help, and then learn from that so that next time you don't ask. That's what interns do, that's what new employees do, and computer science never did that. And I'm like, okay, let me give that a shot. Can I build an interpreter that's going step by step? And whenever it has a question, it asks, and it asks a subject matter expert, not a developer, because this language is English, so subject matter expert will be able to do that. And when I started working on that, I realized that that's the missing piece. And now you can actually get eventually deterministic behavior just by asking so that you know what the problem was and how does the business work. That's my answer. We need to get to this natural language as a deterministic programming language by asking a human anytime there is a problem and not guessing, you should not be able to guess. Yeah.

Dave Vellante

>> Okay, so this is the big idea. I mean, I've seen that you've said every business runs on English. I mean, not every... We are using English here, of course, as a metaphor for natural language, but what do you mean by businesses running on English or any other native language? That's how we're going to communicate to them. I want to put a finer point on that, but help us understand why it's the starting point for enterprise AI. Is it because it's intuitive, it's natural, but there's some missing pieces. As you mentioned, there's got to be an interpreter. I have a number of other questions as to how it all works, but pick it up from there, if you would.

Binny Gill

>> Yeah. The fundamental observation is written down rules is how humanity manages intelligence at scale. Think about it, right? If I have two people in my business, I don't have to write down process. It's okay. I can talk. Five people, 10 people, 50 people, I say, "Okay, you have people, you need process." Why do we say we need process? Because how do you manage so many people at scale and have them do something that you would want to control? So humans write process so that other people follow it and they don't do things that are not approved. Businesses do that, countries do that through constitution. Even at a global scale religion is like that, you write down the rules and everybody must follow, and there's a penalty for not following. If you don't follow religion, God is going to punish you. If you don't follow the constitution, you're going to jail, the law is going to punish you. If you don't follow business process, you might lose your job. There's a consequence for taking intelligence and saying, "Follow this rule, otherwise something bad will happen." And now that intelligence can come and go, meaning a new employee can come in, they follow the same process. Imagine, let me take a different example. McDonald's with fries. The fries taste the same no matter where you go. How? It's not by training cooks in a kitchen, it's through written down process. Now, if you want to change how it tastes, you just write down the process and everybody does that. So that's the overall philosophy. And I think if that worked for humans, that will work for AI. That's number one. AI is churning today faster than humans churn in a business. AI leaves us in two months, right? A new model comes in. So if I was training the previous model, guess what? It's useless. The model is already there. So the most important, the goal that an organization has is actually not what they have trained AI, but what they have documented as how their business runs. That's how you've done it with humans, and that's exactly how you need to do it with AI. We say, "Let's write down your standard process, but also just write down the tacit knowledge, that tribal knowledge that is in people's heads also needs to be written down." The moment you have written down these two pieces, then AI can come and go. As long as it is forced to follow the written down rules, how do you force it? You cannot make AI do stuff out of fear. That's what we do mostly with humans, like it or not, that's how we have civilized society. There is always consequences if you don't do it, AI does not have that. The other way of doing it, which is what we say, remove the planning of AI from the execution part. Let it be creative when it's planning what to do, but when it executes, just give it a deterministic engine so therefore it is forced to follow because it cannot be creative when it is doing something that was already written down. Does that make sense?

Dave Vellante

>> Yeah. So you know what's ironic listening to you speak is in your previous company, the premise was infrastructure as code and now it's English is code. You're making a case that this is a better fit for business process than shoving everything into a traditional software abstraction. And I think that's what's got the markets freaked out in a way. Not that software's going to go away, but it's going to evolve for sure. But you're also making the point that AI, in Kognitos specifically, you separate the planning from the execution, and you argue essentially that AI should be both curator and executor of business process. So why is that, and what does that look like in the real world and in practice with your product?

Binny Gill

>> So yeah, so pick any business process, right? Let's talk about, I have to do a three-way match or a four-way match for paying for an invoice that came in. So invoice has to match with my purchase order with the goods received and sometimes even a certificate analysis and make sure everything is aligned. Now, I can talk to a VP of finance and say, "Okay, what is the process?" You say 90% is the same. Every business has to do it the same way, but we do have our own codes. We do have our own tolerances like if the invoice and the purchase order, the line item is mismatching by a few cents, we have our own tolerances, maybe less than a dollar, maybe $10. God knows. I mean, it depends on which company. So there are some things that are specific to a business. Now, all that can be written down as English code. Now, what we say is that you go and talk to our system, log into our system, and you have a few options. If you already have documented standard process, because you were mature enough, you were running at scale, and you handed off that standard process to a new employee that joined your finance department. Give that same handbook to Kognitos and say, "Go read it and understand it." Ask me all the questions which this is still ambiguous for you to execute deterministically because humans do make some assumptions that other humans are aligned and this correct assumptions because most humans are aligned with each other. We interview humans to make sure they're aligned with how we are culture is how business thinks. AI is not aligned, you don't interview AI. So we let AI look at the standard process you're giving and then curate a more detailed standard process and ask you to review it. That's one thing. If you don't have a standard process, we help you build one. And that's what I say, that AI must be the curator as well, meaning what? So you say, Hey, I normally do this. AI interviews you through a chat and say, show me an example. I say, "Okay, here's one document, here's another document, these are the fields I compare." And you explain to AI and AI is going to say, "Okay, I understand your process. By the way, you're missing these two or three best practices. Let me go create the process for you and document it for you." I believe AI is much better at documenting than humans are. And we have seen that. It's not lazy and it can write things in a more clear manner. And that's what I say is the planning phase. Let AI work with the human to plan what the business needs to do. Now, once it's planned in a language that is reads well for a human or for AI, then we say, now we enter the execution stage. For execution, we have built a deterministic neurosymbolic engine that takes that English and executes as if it was code, just like Python. And that removes the worry that most regulated use cases have is like I don't want something that has not been approved to run in production. And I don't want hallucinations and stuff like that. And that's what we are providing where we have a neurosymbolic engine that goes and runs it deterministically. Now, anytime it needs to ask another question because there's something unusual happening, it goes back to the first engine, which is the planning engine that's highly creative, that goes back and talks to the human, "Hey, I was running this second step or the fourth step, and I see that the invoice has a purchase order number that's not available in SAP. What should I do?" And the human says, "Oh, okay, maybe nobody has entered it there. Maybe somebody forgot. Can you go check this email and see if I can find it?" So that's tribal knowledge, right? You don't want to encode that into the main process. The system does that and says, "You know what? I'm going to keep a note of it as a tribal knowledge. It's a troubleshooting thing that later on I might refer to if you allow me to do that." So that's how you're gathering knowledge. At the same time, you're documenting what the business does and execute it. Now, imagine this, if that is the only way you're running a business, documented code is running the business and all functions of it, then that documentation is always the correct documentation. It never goes stale. Today, what happens with all SOPs and all the knowledge discovery, process mining and all of that, the moment you've done it starts getting stale and this won't become stale because that's running at best.

Dave Vellante

>> Okay. So first of all, I love the term, the phrase tribal knowledge. I use it all the time. In fact, on your website, you get this cool section of the site where it has this palette of fade-ins like ask Joe or ask Fred or hold for legal review, freeze for legal reviews, things like that, which is how things work. But I'm interested in how what you're doing is different than say RPA started to, they evolved into process mining. What's different between what you're doing and the traditional RPA and process mining?

Binny Gill

>> The language of coding, right? It's English and basically everything that you do, testing, debugging, troubleshooting, you're just looking at English. You're not looking at a workflow diagram, you're not looking at Visual Basic Python or anything. And it's not through no code, it's not through hiding the code. It's actually by exposing the code, but the coding language is English. The realization I had was I've been a software engineer for three decades and I've never used visual programming, Visual Basic. My kids have used it, drag and drop to make something, but we developers always like language. We do not like drag and drop. And when I'm talking to you, I'm not dragging a hello or a thank you or whatever, right? I just use language. The benefit of using language is language allows for nuance without a lot of friction. I can be nuanced. And in programming, we need the same thing because a business really cares about nuance. Otherwise, you just get cookie cutter things, right? And then that's how you don't differentiate with your competition if you cannot do nuanced processes in your business. So therefore, language is needed. Now, RPA never actually succeeded. Think about how much of a business runs on RPA drag and drop versus how much of it runs on something that is coded using a programming language. That's a lot. Programming language gives you true power. And all I'm saying is that it's still true, but that programming language must be natural language so that finally you bring in the business folks into the fold of programmers and which is now happening. They should be the people who troubleshoot as well and debug. Now, when you have to debug, when RPA is not working, today you have to go and get an RPA developer who has to look at the logs and figure out and look at the graph. In our system, when something doesn't work, you go back to the very person who used to do it manually before automation came in because they know exactly how to handle it, and they're not confused because what they're seeing as a program is the SOP that they anyway would have run if everything was manual, and that is the core differentiation.

Dave Vellante

>> Yeah. And so a couple things there. One is, I mean, you are expert at understanding the unwritten context that lives in people's heads. When the process mining folks try to bolt on natural language, they're going to run into some challenges that you don't necessarily face. The other thing, and just again, reading and preparing for this interview is you are able to learn from the reasoning traces of humans on exception. And then that gets built back into the system so that the next time it's executed without having to get the human involved. Is that correct interpretation?

Binny Gill

>> Yeah, absolutely. And in fact, that is the core distinction between just trying to do a mining on the side or trying to do an interview of humans. If you ask somebody, how do you tie your shoelaces? They may not be able to exactly explain, but they can do it.

Dave Vellante

>> Yeah. Watch me. Yeah.

Binny Gill

>> Watch me is super easy for humans, because a lot of the things are just habit. So what we do is, so you define the standard process, 80/20 rule. Standard process, somebody who is the process owner can easily describe AI can help them create it. Now you're executing the standard process. It hits an edge case that was not foreseen. Instead of crashing and burning, we have an interpreter and that part is patented. It actually reaches out to a human and says, "Hey, in this case, what do you do?" Now, this is going to the person who knows how to do it manually. In that case, they say they chat with the system, "Oh, in this case, I do this." And system says, "No, show it to me." Then they show it to them. Now the system says, "Okay, you know what? This particular contract that we were doing, it has been resolved. Thank you." When the system comes back and says, "Hey, I actually see a pattern in what you're doing. I have learned this from you. Do you think it is right? Is it something that I should note down as tribal knowledge?" And the person could say, "No, no, don't learn this. I just did it because CEO told me." Or they might say, "Yeah, of course, learn this and don't bother me next time." So that goes into our system as tribal knowledge. It bubbles up to the manager and the manager says they have a chance to review. Is that tribal knowledge something that should be done across the board or that is just some, one person's style? So they could say, "No, I want to approve it for across the board usage." Now the next time the system hits the same problem, it will leverage that tribal knowledge without bringing in a human. That's the whole flow that we have done. What it does is the quality of the learning has to be high in this scenario because it was the real business process. It was not a simulation and it actually finished. We can see the outcome and this is kept up to date because the business is running through this. The next time, say a month later, a manager comes and says, "You know what we have been doing? We have been approving this particular example. We have changed our mind." Okay. So you go chat with a system, "We are changing your mind." System will point out all the troubleshooting tribal knowledge that you have and say, "You want me to delete these?" You say, "Yeah, delete them." So that gives you full control. Compare this with either vibe coding where everything is buried and you don't know what's happening, or compare it with AI model training where everything is in the training data set. And again, you don't know what to delete, what not to delete. So the way I think the future of AI is we will govern AI by first solving the documentation of what we want AI to do. The moment you do that, AI will become useful for businesses.

Dave Vellante

>> Okay. So you started the company, there was no ChatGPT at the time it was there, but nobody knew about it. There were no LLMs. You didn't have coding assistants or agents or R3, and of course you didn't have OpenClaw, it's just been weeks now. How are you leveraging these innovations that came out well after you did a lot of the core work and your first principles? How are you taking advantage of all this?

Binny Gill

>> So initially, we had created a version of our vision where English pseudocode was the thing that we were running as software, meaning you have a business process, you write it down as pseudocode and it'll execute it. Ever since LLMs are there, we use it heavily in the SOP curation phase and on the exception handling phase. Those are highly creative. We use the frontier models with governance. We make sure that when it's super creative, it cannot do stuff in your business. It can only come up with solutions or come up with the process itself and ask you to approve. Once you approve, it goes into deterministic execution, which we've already had. So what we've been able to do is to make it more and more approachable for the business users to use. And also, now we have created integrations into Claude code, into Cursor, where people are able to, even developers are able to say that, "You know what? I'm going to take business logic that usually was buried in software and I worry about vibe coding. I'm able to take that out and put it in Kognitos and get APIs to invoke those SOPs." But the rest of the stuff, how it looks, I want to vibe code a visual representation of that. We have also allowed that to happen. And that is accelerating the rate at which you can bring in AI adoption into enterprises because you are not worried about a bad commit because of vibe coding. We're not saying don't do vibe coding. Absolutely do vibe coding, but make sure that the business logic that your business really cares about and wants to audit it and it needs to run the same is actually outside of that vibe code and that's protected by the Kognitos platform. So yeah, there's a lot of things that are helping accelerate how much of this gets adopted here.

Dave Vellante

>> Okay. So very exciting times. You've basically put forth the premise that you code in English, you're going to separate the planning from the execution. How is this going? Give us some examples of what's happening in the field. I mean, that was definitely part of your submission. And I know the judges put a lot of weight on that. Some of the proof points that you've seen out there.

Binny Gill

>> Yeah. A lot of the companies, especially in the areas where they care about regulation, the finance and accounting, even healthcare customers where they care about provenance, they care about who made the decision why. That is also sometimes more important than the actual decision because you need to make sure that you don't get sued. Finance and accounting is leveraging this capability fundamentally because they don't want a finance department that gives you different answers if you ask the same question again and again, because that's what an LLM does. We want deterministic output anytime money's involved. Manufacturing, logistics, retail, CPG, banks, we are broad in terms of the industries where this is applicable, but going deeper into finance and accounting use cases, and healthcare revenue cycle management use cases where it is extremely important to make sure that the logic is approved and upfront .

Dave Vellante

>> One of the things that we talk about a lot on theCUBE Research, and George Gilbert's done a lot of work on this, is the whole idea of a 4D map. And you're using a knowledge graph. Essentially, we talk about businesses increasingly are going to need capabilities of this 4D map, the knowledge graph, et cetera, to be able to converse in things that humans understand. This is the fundamental premise that you have. Not store things in strings that databases understand, but things that humans understand, people, places, activities. And this is exactly what you're talking about, so my question is, how hard is it to build that knowledge graph? What do I have to do to be successful? What are those prerequisites for your customers?

Binny Gill

>> Yeah, Dave. I realized that... So this is how we started. We built a knowledge graph at Kognitos when I started it initially, right? But the LLMs are becoming quite smart to the point where the data could even be unstructured, and it could glean information from that. So it depends on the use case that you're going for. You need it to be a knowledge graph or it could be unstructured and you do a agentic search or agentic understanding for that. I am thinking that the way software is evolving, and I use software in terms of whatever this computer does for you, right? It could be English as software. The way it is evolving, it's going and mimicking more and more what humans do rather than what computers have taught us. So humans don't necessarily do knowledge graph on a daily basis, but we do say, "Okay, I have some books and if I have a question, I'm going back pre-technology. There are books, I can go to a library, I can flip to the... I search in the index, I flip to the page and read the whole paragraph. Now I know the answer." AI is just doing that now, right? Agentic search. You give Claude access to a bunch of documents and it does precisely that. It does a search, reads the relevant paragraphs, and it knows. I think we are going in the world where there'll be not much distinction between how humans operate and how AI operates, but the guardrails for humans are based on society. The guardrails for AI have to be based on technology, and that's what we are providing, the guardrails, but the behavior will be the same. So answer is what we are saying is we are going to document in process documents for the whole company. The knowledge graph, you could build on top of it, but I think it remains to be seen. Will that be the thing? Because at the end of the day, you always want something that humans can also understand and search and do. So if a knowledge graph, if it's a small knowledge graph, great. But if it's a million node knowledge graph, it's useless for humans. And I think we won't even show it to humans. It'll be there somewhere inside the system.

Dave Vellante

>> Got it, thank you for that. I want to ask you about SaaS, the term SaaSpocalypse, everybody's talking about that. You see the stock market, what it's doing. I know you have connectors to, I think about 200 on different tools and applications. There seem to be three schools of thought here. One is SaaS is dead, the market's overreacting to that. The second school of thought seems to be, well, it's not dead, which I think most people believe it's not dead, but the value, the price to earnings value, maybe not as great before. The discounted cash flows, the future value maybe is not as exciting. And then there's a third school of though, which is really what Jensen puts forth. Of course, Jensen is the perpetual optimist, but he's proven that he's right more than he's wrong. That SaaS, they're tools. And agentic is exploding, and millions of agents are going to use SaaS and it's going to increase the usage, and so perhaps it's more valuable. What are your thoughts as somebody who's been in software for decades? What are your thoughts on the SaaS business?

Binny Gill

>> Yeah. SaaS is software as a service. Now, if making software becomes trivial, then providing it as a service is questionable. And that's what people are questioning. Why do you have to provide it as a service? Now, if Claude comes up with the next version, Mythos, and I can just prompt it, make me a CRM and it actually builds me Salesforce, then do I need SaaS? That's the thing that people are asking. Fundamentally, I don't believe that SaaS is dead because of just that reason. I mean, that's not sufficient. There's a difference between writing code and engineering a solution. So yes, I can tell AI to write me all the software for a CRM or an ERP, but how will I run it? How does it scale? What happens when it hits an issue because some server went out or there was, our data center was down. Am I doing backups? I'm doing recovery. What is the SLA on it? Do I have audits? Is it SOC 2, HIPAA-compliant? All of those things. It's not about writing code. It's about process. That is hard. And there are some decisions that need to be made that actually humans decide. It's an appetite for risk taking. How much penetration testing are you going to do on that platform? I mean, I'm sure Salesforce does a whole lot of expenditure trying to make sure that happens. If I ask my OpenClaw board to write in CRM, will it do all of that for me? No. I mean, and will it go too far? God knows. So a startup has certain appetite. As it becomes bigger and bigger and mature, it does different things. So fundamentally, it will not be about software. It'll be about the experience that you're providing and how you're providing it to the businesses, whether they care about what you're doing. It's almost like the restaurant business. Anybody can build a restaurant because they can cook, right? A meal, but that restaurant is not about that. It's not about the meal itself. It's not about even the cook. It's about the other logistics of making sure it is successful and software business is the same so that's one explanation. I don't think just writing software is becoming easy and therefore SaaS will die, but there's another reason that is far more important right now. What's happening is that the businesses, our customers are saying, look, we have to spend on AI, and that budget has to come from someplace. And therefore they're looking at every SaaS vendor and saying, "Hey, you know what? I just saw somebody just vibe code this and I can get the same thing internally, then do I need that?" A lot of the cases they can actually consolidate and they don't need those things because you can collapse the stack and that is... So that is definitely... But the key focus is where do I get money to spend on AI? So either it comes from the SaaS wallet, but unfortunately more and more, it is also coming from payroll. That's what we are seeing is happening. And at some point when the payroll thing becomes big and people are basically getting rid of people, and that's where you get the money to spend on AI, then yes, you do need SaaS as tools that are high quality, high security, vetted, but mainly data stores for agents to write stuff to read and write stuff. So that's what I think the future looks like.

Dave Vellante

>> Yeah. Which was, ironically, it was Satya's premise when he was trolling Benioff when he said that SaaS is just going to become agents talking to crud databases, I guess he didn't realize at the time that the biggest exposure was personal productivity software where all the formats were open sitting in a crowd database and Claude could just come in. And so you've seen Microsoft pivot off of that. But basically you're saying as we scale without labor, I would also say there's going to be fewer, well, let's say maybe not fewer seats, but less seat growth. And so the business model changes, and the value proposition comes from the value that you get out of tapping intelligence or producing intelligence through tokens, your whole COGS model changes. And as I was listening to you speak, I was thinking, okay, you're describing essentially all kinds of new services that could be developed, spun up, new processes that can be improved, new services delivered as software. So whole new industries will likely be born. Do you buy that vision? And I'll ask you to close with what's the future for Kognitos? What do you want to accomplish in the next 12 to 18 to 24 months?

Binny Gill

>> Yeah. I think what's going to happen is people who are creative will now become empowered and they do not need to wait for a lot of funding and they do not need to wait for getting a great software developer, co-founder of anything. If you have a business idea, you can start really quickly. Right now we are looking at a revolution on the software side, but you'll also start seeing this on the hardware side. I mean, I can maybe vibe code an iPhone in maybe five years or whatever. It might be that powerful and I just need to have an idea. Now, that kind of thing will happen where there's anybody with an idea can start, go and build something. Now the challenge will be how do they sell it? And this is like the same thing, restaurant business. Every second shop is a restaurant. Go figure out how you sell it. But people will be profitable. The small restaurants servicing the same hundred people can be profitable because your COGS are very low. You just need few tokens and you don't have to hire a hundred people. It's all robotic, very efficient, and you can get it done. I think overall it'll be good boost for small business where they can figure out how to eat from the larger business who's still working like a dinosaur cannot figure it out. That's one. But on the other hand, infrastructure folks who provide the real estate on which AI runs, they will expand. Future for Kognitos, I believe just like AWS created infrastructure for you so that you focus on building the app and you don't worry about racking and stacking the hardware, which is what my previous company did, right? It's like, you don't have to worry about it. I believe the same is true and Kognitos can help enterprises to go just focus on the English business logic. Rest, we will take care of it. The governance of it at scale, your workload's going up and down, AI churn happening, new models coming and going. You don't need to worry about it if you're worrying about what the business needs to achieve rather than how does this version of AI understand that. That's our job. I think that's the layer we would like to become for the enterprise and allow everybody to innovate very fast.

Dave Vellante

>> Well, Binny, congratulations on getting not only the company off the ground, but it seems like you have had some great success and maybe the best is yet to come. I look forward to seeing you at various events and hopefully we can have you at the New York Stock Exchange sooner rather than later. Thank you.

Binny Gill

>> Always a pleasure talking to you today.

Dave Vellante

>> All right, thank you for watching this special CUBE conversation. I'm Dave Vellante, and we'll see you next time on theCUBE.