A Product Market Fit Show | Startup Podcast for Founders
00:13:46 · Pablo Srugo
[00:13:46] S1: And then how did you decide? Like, that's kind the high level problem, which I, you know, completely agree with and it's been a problem for for a long time. Obviously, with AI, it all of a sudden became a possibility of solving a huge component of that. When do you realize that that that's possible, and and what are the some of the early steps that you take to see that there's a real opportunity here?
00:14:02 · Surojit Chatterjee
[00:14:02] S0: Look. In late twenty twenty two, ChadGPT come came out and there's a lot of talk about LLMs. I started playing with it. I was deciding to step out of Coinbase. That was early twenty three. I started playing with it myself. Right? And and one thing in my head was, can we actually use this technology as a platform, as almost like an operating system to create ability to dynamically plan, use tools, and use memory? I mean, those are the unique human capabilities. We use tools. We have memory that it means we learn from feedback from the environment, and we adapt automatically. Without that, this technology will not work because you will keep building the fixed software which has been the root cause of this problem, which is, you know, we have a lot of software, but they are fixed. Right? So they they don't change. You have to go and keep tweaking, keep doing things for them to kind of adapt to your business. So for me, the first step was I started tweaking, writing my code, some code myself, which I haven't done for a long time. To be honest, I was an executive running large businesses and so on. But my roots are, you know, as a builder, as an engineer, computer science, bachelor's and master's. And then I started I started reading a lot of research papers at that time that came out in early twenty three on how can we use these LLM frameworks, LLM systems to start dynamically planning and and taking actions. This was really the root of the whole agentic thinking. The word agent was not involved if you read the papers. Nobody was using that word. We called it AI employee because we I just really wanted to model it after how humans work. That was the genesis and got together with few other, you know, ex colleagues of mine from Google. My co founder is also ex Google, ten plus years there. And we got very excited about this idea, started building some rudimentary prototypes. We worked a lot with early days, some of the PhD students who were writing these papers, like, almost as volunteers. Some of them joined us back later.
00:16:12 · Pablo Srugo, Surojit Chatterjee
[00:16:12] S1: You're such a selfish person. I actually can't believe how selfish you are because, like, you've been listening to the show. You listened to this episode. You loved it. You've listened to a bunch of other episodes, and you haven't told anyone about it. You haven't told any of the many founder friends you have about it, think about how many founders have helped you out when you're building your startup. So don't be selfish. Tell your friends about this episode, tell them about the show, and help me help them. With these big ideas, on the one hand, they're obvious in the sense that if you can say to someone, hey. If I could create AI employee that costs 10% of what a normal employee does and works twenty four seven, would you buy it? Yes. Of course, I would. Today, people are like, maybe you could do that. You were starting, it was more far fetched. And so how do you go from that huge idea to something tangible? When you start prototyping, what is the first version that you build? What's the first thing you wanna test out? [00:17:01] S0: Yeah. Great question. See, the technology has moved and improved a lot in the last couple of years. But at that time, a lot of things did would not work. So you have to have a belief in where this technology will be in a few years and start building with that future. It's like, you know, early days when e commerce got built. People would have a lot of doubts. I can only say they'd probably buy some books. Will I ever buy, you know, dress online? [00:17:29] S1: Let alone a mattress. Yeah. Things have changed.
00:17:32 · Surojit Chatterjee
[00:17:32] S0: Yeah. I mean, look, for me, I'll give you a little bit of context. I've always worked on things that are in the frontier. So when I was working on mobile, now it is very obvious. Right? I mean, my kids, they have not seen a world before mobile. So mobile phones, so they're like, you know, how did it look like? But when I was starting, like 2007, mobile advertising at Google, then I remember, like, first few years was was struggle to convince even internally, this will be big. Forget externally. I had a discussion, I remember early days, this was still like 2010 or something with like, you know, CMO of Macy's advertising customers. I'm like, you know, mobile advertising will change the world. You need to make sure you have a mobile app and or a mobile site that actually works and people can transact. And he was like, yeah, I see it. But I don't think anybody will buy a dress on a mobile phone. They may just browse, maybe see the picture, but they'll come to the store. So I like your enthusiasm, Suri. I still remember it. He's because I was kind of much junior. I was given 10 in the end of a, like a two hour meeting. The last guy comes in or he's kind of mobile guy, but it's still early. And he's like, I like your enthusiasm, but I don't think it will be true. Crazy. That's crazy. Right? I mean, now you're like, do I ever go to a store unless I really want to, you know, give some time or something. Right? We felt I felt the same way about this technology when we are starting. Right? It was not working, very broken, a lot of things will go wrong. But the trends were there. We started building actually early on some customer support agents. And there's a lot of problems, lot of challenges, right, to make it work, particularly in taking actions. See, answering questions was still possible, but taking actions was super hard. Integrations with with applications, making kind of creating dynamic plans. And the biggest thing is humans trusting AI also. That, oh, I'll open up my system of record for you to kind of come and do some surgery identically. That was a big leap of faith. But I think today, are all at that future already. We are seeing that it's very much possible, and we are seeing attraction of a different kind.
00:19:54 · Pablo Srugo
[00:19:54] S1: So then what was the the first thing that you built to show that this could be possible, to take a step in the right direction?
00:20:01 · Surojit Chatterjee
[00:20:01] S0: Yeah. Actually, I mean, you know, we always talk about things that work. I also talk about things that didn't work. Thing we built, it didn't work at well. Our initial idea was like, we wanted to build a horizontal platform, but we were very clear. We want to showcase that this platform works by building actual applications, actual kind of solving some problem. Right? Like what we are doing today with HR or finance. So we were initially trying we tried for a couple of months like analytics area. You know, the spreadsheet jungle every every enterprise has today. Can you actually declutter it? At that time, it was very difficult. It it was just I mean, today, it's much better in getting there. And we thought, okay, this is not probably the right application to build today. We have the platform. We're building the platform, but we should go and build some of the other applications first. I think the two applications where we saw a lot of good resonance, one was HR, which again, we didn't know or I mean, we saw it in other companies. HR is very fragmented. Right? There are like think of just onboarding an employee. You have to touch like twenty, thirty systems to onboard an employee, giving access to different areas, you know, l and d, you know, compliance training, all kinds of stuff that you have to do for that employee. It's a in a large company, it's nightmare. Look at every stage of an employee's journey from like, you know, recruiting, sourcing, recruiting, onboarding, day to day just employee experience, all the way to maybe like offboarding or any change, like the change of role, change of location, right, all the way to like, you know, offboarding. And even post post offboarding, employees have questions, oh, I wanna get my tax form from last year. Right? Can you give me employee verification letter that I worked for you for last ten years? I mean, every employer has to fill those requests as well. We felt this is that there's just so many broken windows, right, in this area. Agents are our AI employees at the right solution. And we started seeing a lot of traction. We are seeing same thing for say something like finance where similar broken windows, lots of different things you are touching. The other big thing here is every enterprise is unique. And this is a kind of a realization, right, which is, look, I'll give you an analogy.