So I have certain assumptions of behavior. With AI, there's zero assumptions. So with AI, not only the standard process has to be documented, you also have to document the other behavior. And that other behavior, you can't ask AI to come with it because it can be whatever you want. What I've noticed is in big companies, there is a hierarchy that has been built around humans, and humans have limitations. AI models don't have that problem. One AI model can do everything. Where are we in twenty three? From a technology perspective, I think it's not a question of power anymore. AI will be able to Alright. We are here with Binnie Gill, CEO of Cognitos. Welcome. Thanks for having me here, Akeet. You guys started in twenty twenty. COVID, what was that like? Well actually, I left my previous job, I was CTO at a company called Nutanix, and we did well, I was there for eight years and had a great ride, rang the bell in the Wall Street, and then at some point I was, you know, wanting to do get back to the roots, do something starting from the scratch. I love the startup journey, So I had left and I was thinking about what are the areas where I'm truly excited about working, and then COVID hit, so I was stuck in my house, didn't have the ability to talk to people. Normally, know here a founder as well, meeting people and trying to come up with something that would be good to work on together, it was not possible. So that's where I was stuck home alone, and there was nothing better to do, and I was watching my son learn Python, encouraging him, like, go do that. And then I realized that it took him a lot of effort to write his first program, almost the same amount of effort that it took me thirty years ago. It was a game of tic tac toe, and I wrote it in basic, he wrote in Python. And I realized that the world hadn't changed, it was the same amount of effort in learning. I would argue a little bit more because basic was very forgiving, I could use a go to statement to cheat, kind of. Python, you have to be more rigid around how you pass variables inside functions. So I was like, man, maybe we need to do something better. And then in my sessions with my son, I realized that there was this bad feeling I was getting that thirty years, nothing has improved, I got to do something about it. And then he challenged me with one thing, said, Dad, I don't want to learn this language. I said, Pius, like, you know what, Alexa is better. I'm like, there's something there. Alexa does do one liners. This was only five lines that I was trying to, you know, make him write, and that sort of started this journey for me. And that's all COVID. I mean, if COVID wasn't there, then maybe this company wouldn't have been there. So it's the idea that machines should be speaking in our language versus us having to think in their language. And it was already in front of us, right? We you know, this is five years ago, Alexa was there, and we were getting used to the idea that machines can understand more flexibly what we want, and the mental jump I was trying to make there is that's a one liner. Can it be five lines and therefore can it be five hundred lines? And then what could be the structure of those lines? And it has to be something natural that I don't have to teach somebody. And then I had a realization one day is that that's how humans have operated for millennia. Like, even back in, you know, ancient times, people would write down rules, ten commandments or or the bible or something. Right? Constitution of countries. It's all about how do you manage intelligence at scale where every intelligence is slightly different and sort of uncontrollable, untrainable. You send them to school, by the way even school is also based on written down books. So we as society have figured out that the way you control intelligence is through written down logic. And I'm like, okay, maybe AI also will need that one day, so let me try and build an interpreter to constrain AI to follow logic that has been written down just like Python but in English. So that was super exciting back then. I thought it'll take a long time, but five years later we are here and it's actually possible. How did you grow the business? I mean, starting, and by the way, how crazy would it have been if Claude Code and all this stuff was available in twenty twenty when everyone's locked in their house with nothing to do? You know, imagine the amount of SaaS applications that would have, you know, spawned up with, no distractions. How did the business kind of evolve from, you know, you at home alone twenty twenty to where are you today in terms of in place? About sixty people. Okay. Yeah. How was that did you see a have you seen a ramp up? So actually, so the first two years was a lot of research, and this was just an idea, right, how do you even get this? I spent eight months just by myself trying to convince myself that actually English can be a programming language. Now of course I have been taught in school that English is not context free and all of that's a bad language, it's ambiguous, and therefore we learn computer languages, right? And I was trying to convince myself that no, there has to be a way and there are a bunch of people who have tried, in fact COBOL was a common business oriented language, it was supposed to be natural until it became hard, and now very few people understand, but Claude does now. That's what I heard. So folks have tried, and I realized that there was one key piece that every generation that has tried to do English as code could not do, which now we can, which is now we can reach out to the operator and ask the operator a question. In all of computer science, we had to make code unambiguous because there was no room to ask somebody at run time. Right. And now we have everybody has phones, they have laptops, and we are so connected to computers. And I'm like, maybe we go back to the the way how, you know, as a dad, I deal with my kids, toddlers. Toddlers crying. Like, that's a very ambiguous language if you think about it. Yet the toddler always gets what they want because I say, okay. No. What do you want? Do you want this? Do you want that? Do you want the balloon? So I'm trying to extract precision rather than force the toddler to speak the precise language. And I'm like, Maybe we should do that with computers. So English is not bad. It's the interpreter that was dumb. The interpreter needs to be like a parent and say, did you mean this? Okay. I'm gonna do that, but not force the language to change, and that's how we operate, you know, almost everything. So that was the core insight. It took me about a year to come up with the first example of that. The next year I filed a patent for it. Now we have a patent for the core fundamental thesis here as to when there is a problem in your English interpretation, you can reach out to somebody smarter and then change your plan and continue. So that's what we did. With the whole idea of once it hits a roadblock, it thinks, uh-oh, I may not be on the right path here, it can reach out to human, pull human in the loop, get the answer, rewind back to where it started Yeah. And then go from there. Yeah. But it doesn't have to rewind back to this starting the first step. Where it got to because that's what yeah. That's what computers do Right. Today is like if your program was not programmed with certain logic, it'll only crash. It cannot wait there and ask a developer to come and say, Now you changed, because over the last three, four decades, the security experts have told us it needs to be immutable if it's running because virus is going to change it and all of that. So we have done so much computer science work to make things immutable at runtime, but guess what, humans are the complete opposite. I ask you go get me coffee, exit the door and then call me, okay, where? Right? We are completely mutable at runtime, so it wasn't possible with the current engines that we have, so we built our own. For two years, we worked on our own English interpreter. We initially wrote it in Python, but that's an interpreter written in Python that interprets English. Now we have made it more secure and fast, now it's a rust based engine. It's pre LLMs. It is pre LLMs and also even today what we are talking to our customers, our customers are doing finance and accounting process end to end, order to cash procure to pay, they do not want creativity at runtime, right? They do want creativity at compile time, like when I'm discussing how to do my invoice processing or vendor onboarding or contract management, yeah, I want creative insights as to what is the best process for my business, but moment I have I've decided the process and I'm running it, it must be run the same way every time. So the idea we call it neurosymbolic. So there is a neural aspect of it that is creative when needed, planning, and also when you hit a problem. But when you're running it where it's supposed to be the same thing as pre decided, it does not use LLMs at all. So you save on tokens, you save on time it takes, and more importantly, doesn't hallucinate. So it's sort of the meld of the old computer science and the new computer science and building something that's useful. I would imagine, I mean, you're going after, you're working with enterprise businesses, that's probably a big selling proposition of hallucination is not something you have to worry about here because we're not even using an LLN at that stage. Yep. How has the adoption been or or the, you know, acceptance of the idea of AI even being in the workflows of enterprise businesses gone? Have you felt like there has been resistance to it? Not with our stuff. So the way we pitch it is, in fact, we are telling people there is AI only if you need AI, right? So if you have to summarize a document or if you have to look at a picture and figure out whether it's, you know, what kind of document it is and classify, you do need AI for that. It could be machine learning or it could be more modern LLM or visual model. But if you have a workflow that is supposed to be run deterministically, then you do not need AI. So now a year ago, that would not be sexy enough. It's like, you know what, so you're not selling me that. Today, they're hungry for that. They're like, yeah, because it aligning with how normally businesses run, right? We write down rules and say, everybody, you'll follow these rules, even developers, you have to follow this, you have to get it reviewed, then commit, then look at this thing, we come up with rules all the time. If you look at McDonald's when they make fries, they write down how to do it and then that's how millions of people around the world know how exactly to make the same taste work out, right? McDonald's wants to change the taste of fries, they write, change the written down rules and disseminate. They don't retrain cooks around the world. So it's a very natural thing and what's happening as a side effect of what we're doing is we are gathering the institutional knowledge, if you will. So the standard process in English, which is where we start, but all the edge cases, when they are handled by humans talking to AI, we are also having AI author tribal knowledge, also in English. So in the end, after using this for some time, you end up with knowledge of how the business runs, which is a level separated from the intelligence that runs on top of this knowledge. So just like when humans leave the company, we say the tribal knowledge leaves with them, ask them to do a knowledge transfer, ask them to document what they did. We need to do the same thing with AI. When AI is going to leave today, Claude is going to leave me and chat GPT or some Gemini is going to come in, I need everything documented. Yeah. So fundamentally, that's the direction where we are talking to the customers and they appreciate that knowledge engineering is very key going forward. Who are you typically selling into in these enterprise organizations? Like, who's the buyer? Is it CTO? It's more on the business side, the CFOs, and also the CIOs who are deciding, or the CAIOs now, chief AIO, yeah, that's a new thing. So one of these three folks are the starting point, but before they make the final call for, because this is a platform purchase for the company, we are a platform where you could automate any process that you want. Good thing is that you can avoid purchasing twenty different point solutions because AI has to be everywhere. You can't say, I have one problem, say I have a three way match problem, I'll get it from this finance point solution, and I have another one under contracts and procurement, I get another one, and in the end you get so many, and so they are trying to come up with a standardized platform, so the chief AI officer or the CIO looks at it from an AI governance perspective. Like do I have a trail and audit of what AI did and is there repeatability so I can just review one percent of it and know that I understand the rest, right? That's what they look at. The CFO looks at is like how quickly can I automate stuff And at the end of the day, is it becoming a black box or does it become documented? That documentation is the key. That's what you're saying is like, that's the value prop that we need to instill with everyone is if you're not documenting, whether it is LLM to LLM, whether it is people processes, then there's going to be a knowledge gap somewhere in there, whether it's just on the human side or the computer side. Yeah. I mean, businesses work basically based on if you haven't documented, they say, you're missing a very important aspect, people process, right, and product. These are the three pillars on which you stand, and then there's culture on top, that's how we look at running a business. People are there, process, if you have not documented process, I mean it's Wild Wild West, you know, when John leaves, maybe you've lost a big part of your business and now your business is to operate differently. At scale, you have to document. But one thing that is even more important, which with humans you don't need to, is the tribal knowledge, the tacit thing that they come up with. Whenever you hire somebody in your company, you interview them. Why are you interviewing them? Because you're trying to figure out what is the tacit knowledge that they come with. What value are they gonna provide that we don't have today? With AI, you know, it's the same model every business uses. Now that same model can be a pirate, it can be a priest, it depends on what you ask it to do. So with humans that's not the case. I look at the resume, so this guy is a finance major from Harvard, I know, and then he grew up in this place and this is the background and all of that. So I have certain assumptions of behavior, right? With AI, there's zero assumptions. So with AI, not only the standard process has to be documented, you also have to document the other behavior. And that other behavior, you can't ask AI to come with it because, you know, it can be whatever you want. So you have to generate it yourself, and the place where it is stuck today is in people's heads. And you can't ask people to write it down. They won't do that. What you can do is have a system that tries to do it itself and asks for advice once in a while and therefore figures it out and authors it for them. And then human reviews and says, yes, no, yes, no. And then there's an approval, like, sometimes managers are sort of surprised, like why are people doing it this way? I didn't know how approve yet. And it's okay, so that way you can go and tell the humans, please don't do it, you should do it in a different way. So now it's exposing all of that, I think this is the most important thing with AI going forward. Speaking of going forward, it's two thousand and thirty? Far, but not that far. It's only four years. Not only four years, but with the way technology's been moving, could feel like light years when it comes to what it can What's your outlook on that? Where where are we in twenty three? From a technology perspective, I think we're almost getting there to a point where it's not a question of power anymore. It's like AI will be able to do whatever you want it to do. Whatever humans do, it can do. I think we'll stop talking about technology as something that we're innovating on. What's gonna happen though by two thousand thirty is on the people side. That's where the focus would shift. It's almost like nuclear power today, it's been decades. We don't talk about how powerful it is. Right? Like, okay. We know it's powerful. Okay. The challenge is our option. Right? Look. Would you put a nuclear power plant in San Francisco? That's the challenge. Right? Okay. So by two thousand and thirty, power is there. We will be talking about, hey, are we going to put AI there? What's gonna happen to the people? Are they comfortable? Are we re architecting the whole organization? You know, companies like Accenture today are eight hundred thousand people. It's like a city. That transition and transformation of human architecture is gonna take a decade. It's not gonna be easy. There'll be a whole lot of politics, whole lot of cultural issues, and all of that. We'll be sitting there. Now from a company's perspective, the older companies, I predict like ninety five, ninety nine percent would be gone because their architecture is just old. Right? There'll be newer companies that are agentic from day one. The people architecture is based on a different machinery. Let me explain through an analogy. Right? The older companies would look like I have a bunch of tailors with sewing machines. Everyone has a sewing machine, a farm of them. Copilot, everybody has it. Right? The newer companies would look like an assembly line for making shirts. The role of humans is very different. Yeah. They oil the machines, they, like, purchase a new machine, put it there. Provide context Yeah. Document. And they look at quality, but they are not tailors. Right? Yeah. And how many do you need per thousand shirts that are going in? So this newer form of companies that are built on a different architecture, assembly line architecture, they're gonna get hugely profitable. The older ones, I I think they'll be in the midst of a lot of, turmoil. Alright. Bini, thank you so much for coming by. Thank Appreciate the time. Nice meeting you.