MCP vs CLI: How AI Agents Should Really Use Your Apps | Karan Vaidya, Composio의 전체 녹취록

00:05:18.240 In nirvana, okay, so nirvana insurance, uh which is where uh we wanted to integrate with a lot of TSP providers like Keep Truck in Samsara, etc. 00:05:18.639 --> 00:05:20.079 And it was a core. 00:05:20.160 --> 00:05:35.519 Like uh the first integration obviously has its own creativity, but after that, like engineers start hating doing that extra uh integration, and uh I could see, I mean, like AI automating that whole process. 00:05:35.600 --> 00:05:47.040 Uh, we were pretty early, we had access to obviously the early GPT-2, GPT-3 APIs, which made us feel okay, this is a step change in how we see the world. 00:05:47.199 --> 00:05:51.279 Uh, and uh yeah, that's where we started with this problem statement. 00:05:51.839 --> 00:06:13.680 Um, over time, kind of like while building that, we realized that S uh thing is not working out because the models were still very early back then, but we realized we are building agents, which is gonna be uh the way a lot of people will write software in the future. 00:06:13.839 --> 00:06:23.199 So we are just building one type of agent, but there will be like uh n number of agents, different types of agents that will kind of like come, people will build companies around it. 00:06:23.439 --> 00:06:29.360 And the problem that we are facing while building this agent will be applicable to so many others. 00:06:29.519 --> 00:06:43.199 And one of the problems that we were facing was okay, to build and make this agent useful, we had to connect these agents with different apps uh with GitHub so that it can create a PR linear for issue tracking, etc. 00:06:43.680 --> 00:06:45.600 And that made us very clear, okay. 00:06:45.759 --> 00:06:57.759 Like basically, if the same problem statement is applicable to so many uh other agents and it will be applicable, then that's kind of like like uh the problem statement that we should uh build towards. 00:06:58.079 --> 00:06:58.800 Got it. 00:06:59.360 --> 00:07:02.639 And and how did the first round happen at Composio? 00:07:03.040 --> 00:07:10.800 Uh so the first round uh we were essentially building, me and uh Soham were essentially building with our day job. 00:07:10.959 --> 00:07:12.639 Uh we were just tinkering around. 00:07:12.959 --> 00:07:14.160 Then you were in SF or India? 00:07:14.480 --> 00:07:16.399 No, we were in Bangalore, India. 00:07:16.879 --> 00:07:22.879 And yeah, I think we like uh as I mentioned, we had early access to GPT23 APIs. 00:07:23.040 --> 00:07:30.240 Uh then kind of came I think DaWinchi uh coding models, uh the early coding models uh from OpenAI. 00:07:31.279 --> 00:08:19.680 And it kind of while doing those tinkering, we realized okay, the problem statement that we want to build is like kind of the integration one that I just mentioned, and like we had some prototypes, we did some chats with like obviously Bangalore is also a tight-knit community of people builders, so we did some chat and we had some early design partners, and like uh, like obviously, we have some friends and VCs uh via via the news spread, and uh we got some inbound to chat, uh and yeah, I think it was like like uh we decided okay, it's it's fine to chat, and like like I think in first meeting itself, like uh I think we met with Together Fund and Garish really liked the vision and like like was pretty positive, and that's how it all started. 00:08:19.839 --> 00:08:22.560 We raised first round four at 16 post money, right? 00:08:22.720 --> 00:08:24.560 Uh four and uh I think twenty. 00:08:26.319 --> 00:08:30.160 So, how did Together happen to lead and how did you bring in two leads in the round? 00:08:30.480 --> 00:08:32.240 Yeah, so it was uh interesting. 00:08:32.399 --> 00:08:36.000 So obviously, kind of uh I have huge respect for Garish. 00:08:36.240 --> 00:09:27.840 Yeah, so I think we first had a chat with uh Garish, and he really loved us uh the product and everything, and then like kind of like I think uh we had some chats lined up already because uh we have friends across the ecosystem uh in different VCs, and so we didn't want to uh not do those chats, and that's where we did some other chats, and like it kind of came along that like these two or Elevation and Together were the most excited uh team for us because of like we chatted with the team and the way they were thinking about uh uh stuff, and more than that, the I think uh like kind of like founder first mindset that attracted us uh very much with these two teams, and we were decided to do a non-traditional setup. 00:09:28.159 --> 00:09:30.320 And then you could quit your job, supposed that? 00:09:30.480 --> 00:09:31.120 Uh yes. 00:09:31.600 --> 00:09:34.559 So to 2023, and I believe you would have quit your job. 00:09:34.799 --> 00:09:35.919 Uh in like July. 00:09:36.240 --> 00:09:36.559 July. 00:09:36.960 --> 00:09:37.200 Yeah. 00:09:37.440 --> 00:09:42.399 So so uh till the light speed round happened, what was the progress in the company? 00:09:42.799 --> 00:09:52.480 So yeah, I mean uh I think early we were obviously building uh the the the initial product that I mentioned. 00:09:52.559 --> 00:09:56.399 Uh there also we had uh design partners like Glean, for example. 00:09:56.480 --> 00:09:58.480 So that was like uh going good. 00:09:58.559 --> 00:10:07.360 We had some kind of like good uh uh like like uh demand, but uh over time we realized okay, this is definitely not working. 00:10:07.440 --> 00:10:10.480 We the models are still not there in terms of like coding. 00:10:10.799 --> 00:10:37.039 So, in like I think somewhere 2024, March, etc., we decided that we'll pivot to uh kind of like what we are doing right now, and yeah, I think uh like basically we like started building uh and uh kind of like sharing what we are building in different communities, and I'm talking about 2024. 00:10:37.120 --> 00:11:09.759 Uh, there were these like Langchain, uh, like Lama Index, a lot of like agent tech frameworks, and their communities were pretty

active because I think it was still very early in the AgenTech era, and people were still exploring, and like the best part about exploring is you film like build these good communities where people share and kind of like cross-communicate, like transfer knowledge, and that's where I think those communities uh had a clear clear problem statement where people wanted to build these uh agents and wanted to connect their agents with different apps. 00:11:09.840 --> 00:11:15.120 And when we posted, I think we got an immense love from the community, people started adopting, etc. 00:11:15.360 --> 00:11:26.720 And then it was clear that we have some level of product market fiddle, and then it was all about like kind of like iterating on the product with the the love that so what was the problem statement that you got the first PMF on? 00:11:27.039 --> 00:11:34.240 I think it was essentially connecting AI with uh different apps, like basically like in the interaction layer for AI. 00:11:34.399 --> 00:11:38.080 So basically, you for anybody who is building agents, right? 00:11:38.399 --> 00:11:40.240 Uh you are building connectors for them. 00:11:40.559 --> 00:11:50.240 Yes, like building like uh kind of initially it was uh uh building the auth authorization, the tools, the quality of the tools. 00:11:50.320 --> 00:11:54.799 Uh, we had like a very early benchmark on like how quality of tools matters. 00:11:54.879 --> 00:11:57.440 I'm talking about 2024 when the models were still not there. 00:11:57.519 --> 00:11:58.799 I think things have evolved. 00:11:58.960 --> 00:12:07.919 Uh, but like uh we had like one of the first benchmarks of uh uh showing uh that like the quality of tools can improve accuracy. 00:12:08.240 --> 00:12:17.759 Uh it's not just the models, it's like like I think you can get like a uh 20% to 80% jump just by changing the quality of tools, which is like pretty important. 00:12:17.919 --> 00:12:22.960 And I think then we also did some like work on the software engineering benchmark side. 00:12:23.120 --> 00:12:30.000 We were actually fourth on the Suibench leaderboard, and we kind of like I think just like the idea was to show that the quality of tools matter. 00:12:30.159 --> 00:12:31.840 Uh, it's not just the model. 00:12:32.240 --> 00:12:32.639 Understood. 00:12:32.799 --> 00:12:34.639 And how did the light speed round happen? 00:12:34.799 --> 00:12:38.799 I believe you were at 100,000 monthly active developers before the round happened. 00:12:39.039 --> 00:12:40.080 Yes, approximately. 00:12:40.159 --> 00:12:44.080 I don't know the exact numbers back then, but yeah, I think probably around that number. 00:12:44.240 --> 00:12:53.840 And uh uh I think we had some some kind of like a pretty good mix of startups and like some early enterprise uh traction. 00:12:54.320 --> 00:12:57.600 Uh and it was clear that like like people were loving us. 00:12:57.679 --> 00:13:08.240 There were a lot of like great startups in SF that uh kind of like

were loving us, uh, and and we were building out of Bangalore, so it was pretty good pull out here. 00:13:08.480 --> 00:13:35.120 Uh, we used to travel between me and Soham kind of like once in a while to SF to meet the folks, to meet the users, and kind of like have those uh discussions on the product, and yeah, I think it like like uh when we saw that we have like a pretty good pull and we wanted some more kind of like capital to uh grow the business faster, that's when we decided okay, it's kind of like time to probably uh raise the next round to get a much better team. 00:13:35.360 --> 00:13:37.360 And how did how did you meet light speed? 00:13:37.519 --> 00:14:01.600 Uh so uh I mean I was before like like basically I if I kind of like uh kind of see my career, I've mostly been in light speed funded companies very I don't know uh coincidentally uh that like I started my journey with Rubrik, uh which is also like like Lightspeed, Bipple, like the CEO was uh like a light speed partner himself before that. 00:14:01.759 --> 00:14:09.360 Uh a lot of my friends uh ended up becoming a light speed partner after Rubrik because Rubrik and Lightspeed had a pretty good tie-up. 00:14:09.679 --> 00:14:21.200 Uh after that, nirvana actually uh was also led by Lightspeed, and uh basically Raviraj, our investor, was actually the first investor in Nirvana. 00:14:21.360 --> 00:14:29.840 And while I was interviewing Nirvana, I had a chat with Raveraj as well because I was second employee, so um kind of like that helps. 00:14:29.919 --> 00:14:43.759 Uh and yeah, I think that kind of uh like build the relationships and like the relationship like early on in my nirvana journey and which helped kind of like to bootstrap uh got and when did you start monetization? 00:14:44.080 --> 00:14:55.600 Uh we started monetization, uh the first trench of monetization uh we started was just before our series A, uh which was I think kind of like early last year. 00:14:55.919 --> 00:14:56.159 Sure. 00:14:56.639 --> 00:15:09.279 But like yeah, I think we didn't uh like like uh we started like very like nominal tier based pricing, but the major monetization we started was with enterprises early this year. 00:15:09.600 --> 00:15:14.960 So when enterprise decides to to become a paid customer, what do you solve for that enterprise? 00:15:15.039 --> 00:15:17.039 Uh you know, versus the free users? 00:15:17.360 --> 00:15:28.559 Um so I think uh obviously, like uh there are uh different things, like so like we have like two types of users segments. 00:15:28.720 --> 00:15:33.519 One is like a consumer using us directly with like clot code codexes, etc. 00:15:34.000 --> 00:15:44.879 Where you don't need to do much, you go to Composio and uh like you either use CLI, so it's a prompt that you need to put it in your clot code or codex, and or use MCB. 00:15:45.039 --> 00:15:47.440 So like uh that's pretty much the setup. 00:15:47.600 --> 00:15:49.600 Uh that's free of cost right now. 00:15:49.759 --> 00:16:01.120 Uh then we have like a developer segment, which is okay, people building their own agents can use Composio to connect their user-facing agents with the users' apps. 00:16:01.360 --> 00:16:21.279 Uh, and I think that's kind of like like where the actual monetization happens, where okay, if you're building your own product on top of Composio, that's like where you kind of like uh uh decide uh to kind of like like pay on a like like usage basis, and for enterprise, we have like different offerings. 00:16:21.600 --> 00:16:30.960 Uh we kind of like like I think some enterprises want uh that like their tokens stay with them, their own like KMSs of the world. 00:16:31.120 --> 00:16:37.200 So like many kind of like security-based settings which kind of are the label for uh paying more. 00:16:37.519 --> 00:16:37.919 Got it. 00:16:38.080 --> 00:16:43.440 And and which is the most popular product with enterprise today, with majority of the monetization came to come through. 00:16:43.759 --> 00:17:05.519 Yeah, uh so essentially uh the product is the same all across in some sense, uh, but uh uh to give you an abstract of what the product is, uh we have we basically are the connectivity tissue between agents and knowledge work apps where we provide the catalog of different apps, or uh you can bring your own apps on top of composition. 00:17:05.759 --> 00:17:06.400 Uh example. 00:17:06.640 --> 00:17:12.000 Uh so, like for example, Slack, Gmail, Salesforce, like different apps that people use in day-to-day. 00:17:12.079 --> 00:17:21.440 Uh, and where we basically provide like kind of like the tool execution side of the harness, which you can connect with any of your harnesses. 00:17:21.599 --> 00:17:25.839 Uh, if you're using a cloud code, cloud agent SDK, Pi, etc. 00:17:26.319 --> 00:17:43.359 Uh, where uh uh we essentially give those agents uh like capabilities, like essentially it's all five tools which uh give the agents like manage authentication, governance, authorization, uh like like context management. 00:17:43.440 --> 00:17:45.759 So like you can't have like we have 50,000 tools, yeah. 00:17:45.920 --> 00:17:55.039 So you can't put in all to the agent, so like just in time, tool discovery, uh like code mod uh where the agent can write code instead of doing actual tool calling. 00:17:55.200 --> 00:18:00.079 Uh, in the background, we uh essentially all our integrations are built by agents. 00:18:00.240 --> 00:18:03.599 Uh, we have like an agentic pipeline that builds and maintains these integrations. 00:18:03.680 --> 00:18:19.440 So our integrations are essentially self-healing, where uh anytime any tool fails uh because of whatever reason, a mistach case uh can be a reason, or some version update on the background which we were not aware of, uh, and it happened very recently. 00:18:19.599 --> 00:18:20.640 Uh dot dot dot. 00:18:20.799 --> 00:18:25.920 Like even uh the schema is not well understood by the agent, and like it's

trying again and again. 00:18:26.000 --> 00:18:39.839 So we have some a lot of like cases where uh we detect all of these and uh the agentic pipeline gets triggered and a new version of the tool gets created, uh, which happens all in real time, so real-time uh kind of like improvements. 00:18:40.000 --> 00:18:49.759 Uh, then we also do things like skill distillation where we just we have the end-to-end agentic trajectories, we distill them to skills so that uh like like it's kind of like learning. 00:18:49.839 --> 00:18:56.799 So, first time the agent, a bigger model, uh did something complex, it took a zigzag path to reach the outcome. 00:18:57.200 --> 00:19:09.039 The next time we have that learning baked in into the product, so uh something similar if it comes, we'll like like automatically get that skill to the agent, it will be like a straight line. 00:19:09.119 --> 00:19:17.680 Uh and we kind of like are like like distilling those learnings and kind of like to the model level itself itself now so that it's even better. 00:19:18.079 --> 00:19:18.799 Understood. 00:19:20.160 --> 00:19:23.920 I think Composio has been one of the fastest growing developer tooling companies. 00:19:24.319 --> 00:19:24.960 Is that right? 00:19:26.000 --> 00:19:31.920 Yeah, I guess I think specifically in the agentec space, uh kind of like we have been growing pretty much. 00:19:32.000 --> 00:19:37.200 So, what has been the reason reason for such a fast growth and what has the GTM that resulted in such a fast growth? 00:19:37.519 --> 00:19:52.319 So, yeah, I think uh that's a strategy uh kind of we started with like pretty early on, like sort of we decided early this year that we are gonna grow, like like so. 00:19:52.400 --> 00:19:55.920 I think the model that I would describe is very parallel to 11 labs. 00:19:56.160 --> 00:20:10.240 So in 11 laps, you have this kind of buy uh like like two product approach where okay, you have a creator product where anybody and everybody can create voice clones, etc. 00:20:10.559 --> 00:20:59.839 And then you have a developer product uh which both run in parallel and are somewhat like synergetic, and that's what we are like trying to do, where we have a consumer play where anybody can use us with their clawed codes, codexes, open clause of the world, and then there is a developer play, and the idea is developer products inherently are uh somewhat slow moving, specifically because uh when I see like a uh a product, it's not the same time I'll start building because my use case of building might come like later in the journey, like while like if I'm kind of like operating, uh I'm not like right now building something that requires the developer product, uh but the activation, and that's why the activation developer product is somewhat low. 00:21:00.000 --> 00:21:03.599 Uh, while for consumer product it's very high, right? 00:21:03.680 --> 00:21:06.880 Like I can start today uh using the product that I saw. 00:21:07.039 --> 00:21:19.200 So that was the strategy where okay, uh you get the activation from the consumer product, and when that consumer wants to build, uh there is like a transition where okay, they remember because they're using day-to-day things. 00:21:19.440 --> 00:21:19.839 Can you give examples? 00:21:20.559 --> 00:21:23.279 For consumer product where it resulted in such a life cycle. 00:21:23.599 --> 00:21:26.880 Uh like 11 laps is one example that I can kind of think of. 00:21:27.119 --> 00:21:34.240 I'm not like I might have to think more for others, but like, yeah, that's that's where I think like for example, ChatGPT is also one, right? 00:21:34.400 --> 00:21:40.160 Like where you have a Chat GPT where everybody is using ChatGPT and they move to the like Open AI developer platform. 00:21:40.319 --> 00:21:40.400 Yeah. 00:21:40.559 --> 00:21:41.359 Same with Claude. 00:21:41.440 --> 00:21:59.839 And like like a lot of this is happening now more and more, uh, where you see that like okay, like the developer platform is using their APIs to kind of build the consumer offering, and consumer offering is resulting, it's like a growth acquisition channel, which is resulting at like like whenever they want to build to the developer platform. 00:22:00.160 --> 00:22:00.799 Platform. 00:22:01.119 --> 00:22:02.160 Very cool. 00:22:02.480 --> 00:22:09.920 And if you can share the monetization journey in approximate numbers, you'd mentioned last year you started monetization. 00:22:10.000 --> 00:22:11.359 What has it resulted in today? 00:22:11.680 --> 00:22:22.559 Yeah, so last year we started monetization, but after our series A, we focused more on growth because like we could see that we are just starting on the verge of like knowledge work getting automated. 00:22:22.960 --> 00:22:35.200 But we started enterprise monetization again early this year, where we had some amazing enterprises, AWS, Zoom, Glean, all building on top of like uh product uh agent to product on top of Composio. 00:22:35.440 --> 00:22:40.160 So since the start of this year, we have like 7x in our revenue. 00:22:40.400 --> 00:22:42.799 Uh and like yeah, continuously growing. 00:22:42.880 --> 00:22:46.559 I think the goal is like ending this year 20x from the start of this year. 00:22:47.119 --> 00:22:49.519 Ending and end of this year, what what's the target? 00:22:49.759 --> 00:22:51.680 Uh the end of the year target is around 20. 00:22:52.079 --> 00:22:52.960 20 million AR. 00:22:53.039 --> 00:22:53.279 Yeah. 00:22:53.519 --> 00:22:53.680 Yeah. 00:22:54.000 --> 00:22:56.640 You must be adding what like one or two million ARR every month. 00:22:56.880 --> 00:23:00.640 Uh yeah, I mean, like I don't want to go into numbers, but yeah, it's going good. 00:23:01.599 --> 00:23:08.880 And and these are enterprises uh who are coming inbound to you, or how uh how you are doing going to these enterprises? 00:23:09.119 --> 00:23:16.799 Uh so yeah, like basically we have like a pretty good PLG, uh,

which I was describing the whole consumer then developer play. 00:23:16.960 --> 00:23:31.039 And via that, like we have a pretty good solid inbound where people just start building and like like they come to us when these enterprises need contracts, they need like some security offering, which is like like kind of uh not available in the PLG. 00:23:31.359 --> 00:23:34.559 Got what have been some of the challenges during the journey? 00:23:34.880 --> 00:24:12.799 Um I think I would say it like right now, uh one of the biggest challenges I think most of the startups building in the space is facing are facing is basically around like the uncertainty, kind of like around this like space where okay, you have like the the pace of progress is so fast sometimes that kind of like we have been on the luckier side till now, but I think you have to be always on the constant lookout for okay, are you building for the future or are you getting gonna are you gonna get crushed in the future? 00:24:13.039 --> 00:24:17.359 So that's like a constant like battle that you have to fight in your head, okay. 00:24:17.519 --> 00:24:23.839 Uh like three months on the line when a newer model release, or six months down the line when even better model release. 00:24:24.240 --> 00:24:34.400 Uh, kind of how can I maintain my relevance and how can I like like be more uh actually like more suitable for that world? 00:24:34.640 --> 00:24:38.880 So yeah, I think that's kind of like one of the biggest things that you have to keep on evolving. 00:24:39.200 --> 00:24:39.759 Got it. 00:24:40.000 --> 00:24:49.759 And uh uh let's say uh from India, I think three companies emerged at the same time in your space, right? 00:24:50.079 --> 00:24:57.759 You were there, Port Key was there, I think True Foundry also is in the same space today, right? 00:24:57.920 --> 00:25:02.319 But uh but on the US side, which were the companies which would be in your space? 00:25:02.960 --> 00:25:06.480 Like uh, I mean there are a bunch, right? 00:25:06.559 --> 00:25:27.680 Like there is there are all these agentics framework companies like Langchains of the World, Lama Index, uh then obviously uh I think uh I think now some of them are customers, but like uh the background agent like that emerged like early this year, OpenClaw, uh Hermes uh agent, etc. 00:25:28.400 --> 00:25:46.480 Um I think like these all are in like like I think what I consider like the space in which we are operating is essentially uh like knowledge work agents, where obviously like last two years we saw like humongous companies uh being built around coding agents. 00:25:46.640 --> 00:25:52.079 You had a cursor, you had like like a lovable replet, etc. 00:25:52.319 --> 00:25:54.319 Those all are in like coding space. 00:25:54.559 --> 00:26:17.039 I think uh like this year, next year, I think it's all about like knowledge work, and we have like a lot of these companies, like background agents, clean, which is also another customer, like

all of like these companies are essentially in parallel where we're trying to bring that level of agent tech capabilities uh which were available for software engineering to knowledge work. 00:26:19.440 --> 00:26:25.839 And uh you don't have a thread that you know like all startups have that open AI and anthropic can become competition. 00:26:26.079 --> 00:26:27.680 I mean, yeah, that's what I was saying, right? 00:26:27.759 --> 00:26:36.559 Like that's the constant, like it's not about the comp like competition from them directly, but it's also uh I mean, uh being relevant with what they do. 00:26:36.799 --> 00:26:43.680 Obviously, there is some, like for example, Enthropic have their own integrations in plot code codexes of the world. 00:26:43.920 --> 00:26:46.720 But I think there, I think it's very clear. 00:26:46.960 --> 00:27:16.880 Uh, because it's not a single model world, it's multimodal world that people want a unified layer to own what we own, uh, which is like things around the model, which is like context, the authorization governance piece, uh, the learning piece, like which is like I think Satya had a really nice blog around like every company is essentially in like its own like learning sandbox where you have to own the learning and context. 00:27:17.039 --> 00:27:20.079 So, like essentially, if you think about it, like that's Composio. 00:27:20.400 --> 00:27:23.759 Uh, we are that sandbox, we provide that sandbox around the model. 00:27:24.079 --> 00:27:34.799 And given uh there is for every uh opus, there is a chemical setting out there, uh, we are in a pretty good spot that way. 00:27:35.039 --> 00:27:42.559 Yeah, I think the more open weights and open source model uh get popular, the more you get popular, right? 00:27:42.799 --> 00:27:43.440 Yeah, exactly. 00:27:43.519 --> 00:27:48.319 So kind of like like I think uh that's that's where we are doing a lot of work. 00:27:48.400 --> 00:27:56.319 We have like uh some amazing benchmarks uh that are going pretty viral as well, but like more than that, like uh on Twitter. 00:27:56.400 --> 00:28:06.319 Uh but more than that, I think it's kind of like like uh the audience is moving towards using a mixed mix of models, not relying on one provider. 00:28:06.400 --> 00:28:15.440 Yeah, and that's where I think kind of it's our duty to kind of take tell them uh which model is suitable or for what. 00:28:15.599 --> 00:28:28.640 And that's where we have some kind of like like products which are okay, like like given the use case, we'll tell you which model is the best for that use case, given our older uh set of like like PLG data that we have. 00:28:29.279 --> 00:28:40.880 So so would would you be going for another round of fundraise this year or uh uh yes, that's kind of like the plan, like uh like we are uh like gonna go like probably this quarter. 00:28:40.960 --> 00:28:41.920 It's understood. 00:28:43.759 --> 00:28:49.119 And uh let's say one year down the line, you know what you do you

imagine Composia would look like? 00:28:50.000 --> 00:28:53.119 Um yeah, I think uh that's a good question. 00:28:53.279 --> 00:29:24.400 Uh basically, like I think one year down the line, I would first kind of give my worldview, which is I think a lot of these enterprises will have a lot of self-improving, self-learning loops around how they operate, which is the the whole Satya's blog around like kind of the like self-learning, self-improving agents in uh different enterprises. 00:29:25.119 --> 00:29:49.839 And the idea is you would like the enterprises will want to own the context, will want to own their learnings, and will want to in some cases own the weights, which is like kind of like like not use the the uh open AI's entropic of the world, but host their own models so that they own end-to-end stack. 00:29:51.680 --> 00:30:04.319 Like I think the and the the kind of like the reason for that is like avoiding lock-in, avoiding the data flowing through the the edge that they have flowing through different of these like labs. 00:30:04.640 --> 00:30:37.039 I think Composio is essentially trying to avoid uh like like build that worldview without like with avoiding the lock-ins to the the big labs where okay, if you have uh the rest of the stack on Composio, today you are using OpenAI, like with kind of like all the apps that you have, uh all the authors in Composio, all the context sets in Composio, all the learnings that you have because of the whole agency traces set in composure. 00:30:37.440 --> 00:30:51.680 Tomorrow you can move from like all your skills are in Composio, you can move from like open AI to a Kimi very easily because the the things around it, the learnings have been distilled, which Kimmy gets out of box directly. 00:30:55.680 --> 00:31:01.119 So your one worldview is that you are saying is enterprises would be more sovereign. 00:31:01.519 --> 00:31:04.400 Yes, they would be using their own models, uh right. 00:31:04.799 --> 00:31:08.559 But do enterprises have capabilities to to build that? 00:31:08.960 --> 00:31:18.160 So that's where I think like like we have like a pretty good like forward-deployed uh team which helps these enterprises to kind of like get those capabilities via us. 00:31:18.400 --> 00:31:20.720 But that's like a services arm of Composio? 00:31:21.119 --> 00:31:38.720 So I won't say it's like fully services arm, it's more around I think like initial setup or bootstrapping of the whole setup uh for the bigger enterprises to make sure that they are kind of like AI first, uh like like uh like whatever like learning loops that we have to set up, we can help them set up. 00:31:38.960 --> 00:31:45.039 But let's say the current customers that you shared, like AWS, Glean, and Room, they're all all AI first. 00:31:45.519 --> 00:31:46.160 Yes. 00:31:46.960 --> 00:31:56.000 Uh yeah, I mean, but still kind of like they are I think there is still like some uh information or submit tree and you still need help. 00:31:56.319 --> 00:31:59.759 Okay, so you are uh providing these companies also FDs. 00:31:59.920 --> 00:32:00.880 Yes, yes, yes, we do. 00:32:01.200 --> 00:32:03.839 Today, how many paid customers would you have approximately? 00:32:04.160 --> 00:32:08.799 I think overall we might have like around somewhere between 500 to 1000. 00:32:09.119 --> 00:32:15.359 And these include startups, paid startups to large enterprises or how many Fortune 500 you would have today? 00:32:15.519 --> 00:32:20.400 Uh I think that will mean like kind of like somewhere uh double digit thing. 00:32:20.880 --> 00:32:23.759 So your goal would be to to go it make it triple digit. 00:32:24.079 --> 00:32:24.960 Yeah, yeah, yeah. 00:32:25.119 --> 00:32:28.960 I think over like end of this year, next year types. 00:32:29.279 --> 00:32:29.839 Understood. 00:32:31.680 --> 00:32:38.720 And um let's say uh the priority for next 12 months or 24 months, is it revenue, is it growth? 00:32:38.799 --> 00:32:39.680 What is the priority? 00:32:40.160 --> 00:32:48.799 I think it's mostly like like growth because we are still in that phase of like knowledge work agents have just like gotten the inflection point. 00:32:49.200 --> 00:32:57.119 I think uh the the the market there is like much, much, much bigger than coding market. 00:32:57.680 --> 00:33:01.680 And we could all see from the coding market outcomes, right? 00:33:01.759 --> 00:33:10.720 Like kind of I think the initial set has to be like just getting the people to use you and get hooked onto your product. 00:33:10.880 --> 00:33:14.079 I think the revenue is very easy to tap in after that. 00:33:14.400 --> 00:33:19.119 So what would your estimate uh put a number on size of the market? 00:33:19.680 --> 00:33:29.839 Uh I mean it's like kind of I would just say it's like probably, if not more, at least like 20 to 50x of the coding market. 00:33:30.160 --> 00:33:30.640 Why? 00:33:30.960 --> 00:33:35.920 Like, like I think at the end, like coding market, even like though it looks really huge, right? 00:33:36.720 --> 00:33:40.960 It's almost like what today, 100 billion in revenue combined everything? 00:33:41.279 --> 00:33:42.319 Probably more, I guess. 00:33:42.400 --> 00:33:45.039 I'm like, yeah, I'm not very good with these numbers. 00:33:45.200 --> 00:33:47.200 Uh like VCs are better there. 00:33:47.599 --> 00:33:52.079 Uh but yeah, I think coding is still like enabling the knowledge workers, right? 00:33:52.319 --> 00:33:54.880 Like most of the companies are enabling the knowledge workers. 00:33:55.119 --> 00:34:18.320 So I think knowledge worker knowledge work revenue like market includes like a finance re like market in New York to insurance market in like like uh like uh like the whole of US to like legal market to like like the whole kind of like economy is like probably like software is a big part of it, but it's still like minuscule compared to rest of it. 00:34:18.559 --> 00:34:19.679 Yeah, understood. 00:34:20.000 -->

00:34:25.599 And uh today, you know, uh as you are growing, how have you built the team? 00:34:25.679 --> 00:34:27.920 Like what is the team size today and how are you adding? 00:34:28.400 --> 00:34:34.400 Yeah, we are around like I think 30, 30, 35 homes overall. 00:34:34.960 --> 00:34:39.519 Yeah, I think majority still like engineering in product. 00:34:39.760 --> 00:34:42.800 And how have you between you and Soham split responsibilities? 00:34:43.039 --> 00:34:50.000 So I'm focusing more on the product and tech side of things while Soma's focused, like he's the CEO, so he's focused a lot on like sales, growth, etc. 00:34:50.239 --> 00:34:50.559 as well. 00:34:50.639 --> 00:34:52.960 Obviously, like deeply involved in product as well. 00:34:53.760 --> 00:34:57.119 And are you seeing people from Compodio start their own companies? 00:34:57.440 --> 00:35:10.639 Yes, uh, we are, and that's kind of like really, really kind of good to see, specifically kind of like I think in India where we built our initial team, like like kind of like a like a couple of folks have given. 00:35:10.960 --> 00:35:11.840 Can you give examples? 00:35:11.920 --> 00:35:31.519 Like who are uh so for example, I think there's like this uh person called Pratik, yeah, who was building uh kind of like uh inside the company was building uh this thing called Agent Orchestrator, and like uh kind of that got pretty good adoption inside the company, and he got pretty excited, and we were like very supportive there that if you are interested, I think love to support you. 00:35:31.679 --> 00:35:34.639 I think uh he started his own company around that product. 00:35:34.880 --> 00:35:35.280 Got it. 00:35:35.519 --> 00:35:41.199 So, but Agent Orchestrator was a product inside Composio, it has 9,000 GitHub stars. 00:35:41.440 --> 00:35:42.079 Yeah, yeah, exactly. 00:35:42.159 --> 00:35:43.920 It was for internal productivity mostly. 00:35:44.079 --> 00:35:52.639 Okay, and it got like insane uh adoption uh in the team and outside, and like kind of like that was what he wanted to kind of like perceive around. 00:35:52.719 --> 00:35:53.679 So understood. 00:35:53.840 --> 00:35:55.440 Understood, got it. 00:35:56.239 --> 00:36:00.159 And you you are seeing more and more of people kind of now start up. 00:36:00.480 --> 00:36:02.159 Uh yeah, I mean it's a mix. 00:36:02.239 --> 00:36:07.039 Obviously, people want to stay and kind of like get like see the growth journey completely. 00:36:07.360 --> 00:36:15.199 But uh obviously, when you get so passionate, like pretty, obviously, we are kind of very supportive and like uh like in whatever you want to do. 00:36:15.519 --> 00:36:19.119 But you could have also like the agent orchestrator could have become one of the product line. 00:36:19.199 --> 00:36:20.000 So, how did you decide? 00:36:20.320 --> 00:36:28.960 I mean, it was more around like I think as a startup you have limited resources, you have to be focused on what you're doing and like doing really well at that. 00:36:29.519 --> 00:36:34.400 God, you're careful about the kind of battles that you have everything. 00:36:34.800 --> 00:36:35.440 Yeah, yeah, yeah. 00:36:35.599 --> 00:36:36.719 Uh you have to be, right? 00:36:36.800 --> 00:36:40.400 Like as a startup, you always are kind of uh fighting against the odds. 00:36:40.480 --> 00:36:45.760 So who are the best external folks besides the founders that you brainstorm with? 00:36:46.000 --> 00:36:46.320 Sorry? 00:36:46.559 --> 00:36:49.039 Who are the best external folks that you brainstorm with? 00:36:49.199 --> 00:36:56.320 Uh I think it's a lot of like like great advisors that we have, obviously, that helps uh in addition to the investors. 00:36:56.480 --> 00:36:56.719 Yeah. 00:36:56.960 --> 00:37:04.480 Um I mean Example uh like like Glean founder, uh like Arwin is an advisor and like mentor. 00:37:04.639 --> 00:37:04.800 Okay. 00:37:05.119 --> 00:37:09.119 More so from my rubric times, he was founder of rubric as well. 00:37:09.360 --> 00:37:11.199 So he's really helpful. 00:37:11.280 --> 00:37:18.400 Uh kind of like I think uh like Zoom is a customer, Zoom CEO Eric is really helpful as well. 00:37:19.199 --> 00:37:22.960 Uh kind of like helping us on advising. 00:37:23.280 --> 00:37:29.199 Um kind of like uh one of my senior is like founder of Turing, who is uh like Vijay. 00:37:29.280 --> 00:37:31.920 Uh Vijay is uh also an angel in the company. 00:37:32.159 --> 00:37:40.159 Uh he like I think I met him very late when our rounds were not kind of growing, but I think I'm gonna get him to the like next round. 00:37:40.400 --> 00:37:45.920 Uh and there are like a lot like Soham Azumdar, like kind of founder of Rubrik again. 00:37:46.000 --> 00:37:51.920 Like I think all these like kind of like founder uh relationships from past, I think are definitely very helpful. 00:37:52.559 --> 00:37:55.199 So these are the folks that are helping you on the growth path. 00:37:55.440 --> 00:37:55.920 Yes, yes. 00:37:56.000 --> 00:38:03.199 Yeah, I think I might be missing so many more names, like Guilermo, uh founder of Vercel, is like again, like uh one of the angels, and like he's also really helpful. 00:38:03.360 --> 00:38:08.079 Dharmesh is well, founder of pot, uh Manish. 00:38:08.320 --> 00:38:11.119 Yeah, Manish, obviously, like the like one of the first angels. 00:38:11.199 --> 00:38:13.760 Now he's kind of working on building the company as well. 00:38:13.840 --> 00:38:14.239 But yeah. 00:38:16.559 --> 00:38:18.639 And and where do you see the market headed? 00:38:18.880 --> 00:38:21.840 So AI gateways are a lot popular with large companies. 00:38:22.000 --> 00:38:28.960 They have their strategies to to tuck in these products, like Port Key got acquired by Paulo Alto. 00:38:29.119 --> 00:38:29.599 Right, right, right. 00:38:29.760 --> 00:38:32.320 Galileo got acquired by Cisco for 400 million. 00:38:32.559 --> 00:38:33.679 Oh, yeah, makes sense. 00:38:33.840 --> 00:38:34.639 I

think so. 00:38:34.800 --> 00:38:40.400 I think kind of like AI gateway is a very broad term, in my opinion. 00:38:40.639 --> 00:38:48.159 So, for example, what Port Key was doing was somewhat different from what we are doing, and same with Galileo was different from both of us. 00:38:48.480 --> 00:39:11.039 Uh I think like like kind of what I would say we are doing is more around like infra for AI agents to build on top of these, like knowledge work agents specifically, uh, to build on top of these different apps and uh whatever it takes around it, like like improvements, self-learning, uh kind of like like model routing, etc. 00:39:11.280 --> 00:39:27.599 Like, but the idea is kind of like the focus is on that like knowledge work agents need to kind of like operate on these apps and like like kind of like have that like improvements on top of it. 00:39:27.760 --> 00:39:29.039 So that's kind of the idea. 00:39:29.280 --> 00:39:45.440 Like I think obviously, like as I mentioned, like uh AI gateway can be just like a routing layer, it can be uh like like Galileo is more around like uh the eval side of things, so it's like anything and everything, while we are more focused on like developer infra, particularly more. 00:39:45.679 --> 00:39:46.159 Understood. 00:39:46.239 --> 00:39:50.880 Which are the other developer infra companies in agent space that you respect? 00:39:51.199 --> 00:39:56.719 I would say uh I think it's not exactly agent space, but like turbopuffer is like in which conflict? 00:39:56.960 --> 00:39:57.920 Turbo puffer, yeah. 00:39:58.079 --> 00:40:00.800 It's in like embedding space, which I deeply respect. 00:40:00.960 --> 00:40:20.320 Obviously, in the uh like like agents, like again, somewhat uh uh like like not exactly agent space, but used by agents, like there are a lot of like DB companies like Superbase, uh Neon, which kind of like are uh like pretty good DX around agents. 00:40:20.559 --> 00:40:22.800 If you have to redesign MCP, how would you do it? 00:40:23.039 --> 00:40:26.000 So obviously every protocol has its own gaps. 00:40:26.239 --> 00:40:42.000 Uh kind of like I think MCP, for example, uh because of being remote, kind of like like doesn't natively have some of the capabilities uh like code execution. 00:40:42.079 --> 00:40:44.400 Uh we have to kind of build around it. 00:40:44.639 --> 00:40:48.880 Uh the file like sharing uh is not the best. 00:40:49.039 --> 00:40:55.360 Uh like like we have kind of like I think uh Composer natively has some better primitives than directly using MCPs. 00:40:55.760 --> 00:41:02.400 I think like it's more around like obviously when you start something, you kind of like don't prioritize everything as a protocol. 00:41:02.559 --> 00:41:18.400 So like that's just like limitation, it's not like uh uh I think uh per se like like a kind of like I think over iteration, over the next few iterations, I think we are seeing a lot more things coming up in MCP uh that will make it better. 00:41:18.559 --> 00:41:25.679 And yeah, I think like some of the progressive disclosure stuff like that CLI has uh is difficult to build on top of MCP. 00:41:25.920 --> 00:41:32.719 So so how in in a month now, what or uh on also in a day, how many active developers would be using Composio? 00:41:33.519 --> 00:41:46.000 Um I mean uh kind of like I I don't yeah, I think I have to recall the exact numbers, but it will be somewhere in like uh I think right now we are having 100k plus active. 00:41:46.159 --> 00:41:59.039 Uh like probably I think like the overall developer segment we have is around 800k, 900k, out of which I would say like like somewhere 150k to 250k will be active on a monthly basis? 00:41:59.280 --> 00:42:04.239 Uh yeah, I I yeah, I guess like I think uh like we have like 100k plus daily. 00:42:04.480 --> 00:42:04.960 Okay, wow. 00:42:05.119 --> 00:42:07.679 Uh so over monthly it will be somewhat higher. 00:42:09.039 --> 00:42:11.599 What are the technical bits that you are taking right now? 00:42:11.840 --> 00:42:36.719 So yeah, I think um like the a few important technical bits are one of the things is like we have this agent tech pipeline that auto-builds and maintains all the integrations, uh which has access to uh because we have like thousand plus apps, we are gonna get to like 10,000 with this pipeline in next month, is the plan. 00:42:36.960 --> 00:42:58.079 Uh and the idea is like like kind of we have like test sandboxes for all these apps and different edge cases, use cases, kind of like that we have seen in real-world usage on these apps covered in our testing harness, which essentially kind of like like like as soon as we see an error, we need to fix it, is the goal. 00:42:58.239 --> 00:43:17.440 So that like agent loop is uh kind of like one of the important technical bets for the company because that's where our PLG uh like advantage shines compared to one company because we have so many companies building on top of us, we'll see any and every type of error faster and collectively solve it for our users. 00:43:17.840 --> 00:43:43.599 Uh, then the other one is like around kind of like we have so many agent tech trajectories across different companies, uh, which is another like data fly wheel that we have, kind of using that maximally to improve agent tech reliability, where okay, distill the learnings from those trajectories into skills, uh, distill the issues that agents are running into, like common pitfalls that we can kind of surface to the agent beforehand. 00:43:43.760 --> 00:43:59.440 Uh, because again, like like we have so much, like we have at this point approximately uh 20 million tool executions happening a day uh on our cloud product, which we can leverage to kind of create those learnings. 00:44:00.000 --> 00:44:21.440 So I think those are like basically uh that is the second important bet that we are doing, and the third important bet is where okay, we distill all these learnings directly into the weights

uh where uh we use things like load adapters or like RL to improve the post-training of the uh model itself on top of these uh skills, knowledge work skills in our apps. 00:44:21.760 --> 00:44:22.480 Got it. 00:44:24.559 --> 00:44:28.400 And and how are you let's say context has become really popular word? 00:44:28.800 --> 00:44:33.199 How are you capturing millions of contexts across an organization? 00:44:33.840 --> 00:44:41.360 Essentially, it's around I think if you think about it, context is mainly a mix of two things. 00:44:41.599 --> 00:44:56.159 Uh I like to kind of like call it like nouns and verbs, where nouns it's the actual thing around okay, this is the customer name, like kind of like uh the value of the customer, dot dot dot. 00:44:56.559 --> 00:45:02.159 And then the verb is essentially how are things done in that particular company. 00:45:02.239 --> 00:45:03.840 Uh, what are the processes? 00:45:04.079 --> 00:45:10.559 Uh like like you have to maintain these particular fields for a particular customer uh in the CRM. 00:45:10.800 --> 00:45:17.599 So I think the the nouns is captured in memory, yeah, while the verb is captured in skills. 00:45:17.760 --> 00:45:27.119 Uh so kind of like the the agentic trajectories converting to skills is where we have like that distillation happening at different levels. 00:45:27.360 --> 00:45:43.039 One is across the globe, like for all our users, where we distill like the common generalized skills on particular like apps like Salesforce or across different apps into common pattern skills across the like user segment. 00:45:43.199 --> 00:45:55.840 Then there are like org-specific segments which kind of capture the verbs of a particular company on okay, when uh sale is done, these are the steps that the sales agent should take. 00:45:56.079 --> 00:45:57.920 So, this is a skill for the sales agent. 00:45:58.000 --> 00:46:01.760 It's particular for your company because another company will operate it differently. 00:46:01.920 --> 00:46:08.239 Uh, so this is how like the skill destination helps in kind of capturing those learnings at a particular company level. 00:46:08.559 --> 00:46:20.000 And and right now, you know, as as Composio is evolving right now, uh what does your leadership team look like and what are the new additions to your leadership team in the next six months to one year? 00:46:20.239 --> 00:46:36.320 Yeah, I mean uh we have some kind of like pretty like uh uh seasoned folks on like the GTM side, like I think uh kind of just mentioned like money issuers or First Angel joined us recently kind of to lead that. 00:46:36.480 --> 00:46:46.320 Uh we have uh uh folks, ex folks from like Samsung, Cred, etc., on the product side uh that are leading the charters. 00:46:46.480 --> 00:46:58.639 Uh and on the engineering, we have folks from Shopify, uh to kind of like like people from like Google, like AI mode, uh that are leading uh the charters of the engineering. 00:46:59.039 --> 00:47:02.400 And we have like like many more on the research side that are opening up. 00:47:02.559 --> 00:47:09.119 I think we are looking for someone to lead our research engineering division, talking to quite a few folks there. 00:47:09.360 --> 00:47:12.800 And then as we come towards the conclusion of the podcast, right? 00:47:13.119 --> 00:47:15.119 Uh, can you share? 00:47:15.199 --> 00:47:18.079 You know, obviously, I would love to know yours and so on's background. 00:47:18.159 --> 00:47:19.360 Where do you come from in India? 00:47:19.679 --> 00:47:20.000 Right, right. 00:47:20.639 --> 00:47:24.480 Growing up, uh which college did you guys go to the companies that you worked in? 00:47:24.559 --> 00:47:26.239 Obviously, you've shared some of those same things. 00:47:27.519 --> 00:47:39.440 And uh uh in your journey of of building Composio, what have been the biggest resources that has helped you to grow as founders, be it books, be it blogs, be it podcaster videos? 00:47:39.679 --> 00:47:40.000 Right. 00:47:40.159 --> 00:47:49.679 Yeah, so I I I come from Jaipur, uh and Shoham comes from like Amdabad, so like like northwest part of India, both of us. 00:47:50.000 --> 00:48:01.840 Um kind of uh we met at uh like Physics Olympiad camp, uh, where like I think 30-35 folks from all over India come to kind of prepare for the internationals. 00:48:02.239 --> 00:48:12.559 Um and yeah, I think after that we were college batchmates and even wingmates in IIT Bombay, uh, which is where we connected the most. 00:48:12.639 --> 00:48:20.719 We did a bunch of things together and kind of like like a lot of projects and formed that level of like camaraderie on like building stuff. 00:48:21.039 --> 00:48:30.000 Um yeah, I think post that I joined this company called Rubrik and then like kind of joined this company called Nirvana Insurance as second employee. 00:48:30.400 --> 00:48:33.280 Um yeah, I think what was the second question? 00:48:33.440 --> 00:48:36.400 And and uh where did for which companies did for whom joined? 00:48:36.639 --> 00:48:38.639 Yeah, I think he started on his own. 00:48:38.719 --> 00:48:43.519 Uh kind of he started the entrepreneurship uh like just out of college. 00:48:43.679 --> 00:48:47.679 Uh he built this company that got exit uh to exotel. 00:48:48.239 --> 00:49:00.320 Uh and then uh I think we both started in between uh something in uh like finance space, uh kind of like medium frequency trading, yeah. 00:49:00.480 --> 00:49:06.559 Um and like back in like 2018-19, but uh that didn't pan out. 00:49:06.639 --> 00:49:18.079 Uh kind of we quit, and then I joined uh this company called Nirvana Insurance and Soham joined uh Bureau, which is another kind of like fraud detection company based out of India. 00:49:18.559 --> 00:49:19.519 Got it, very cool. 00:49:19.679 --> 00:49:22.559 And what are the resources that has helped you grow as founder? 00:49:23.199 --> 00:49:28.159 Uh I think it's like a lot of things. 00:49:28.239 --> 00:49:34.079

Uh honestly, it's obviously like like the great set of advisors, uh, kind of past relationships there. 00:49:34.400 --> 00:49:37.280 It's books as well, obviously. 00:49:37.360 --> 00:49:48.639 Like I'm kind of like I think uh zero to one to like hard things about hard things, a lot of like great books that like founders have written, like uh specifically built like as a founder. 00:49:49.519 --> 00:49:58.559 Uh and I think at the end it's everybody's own journey that is different and that teaches the most. 00:49:58.719 --> 00:50:05.119 So, kind of like I think we have a great team, and every day I learn a ton from the team itself, yeah. 00:50:05.360 --> 00:50:08.639 Which is like like the most touch point that you have is with your team, right? 00:50:08.719 --> 00:50:12.639 If you have like really great folks surrounding you, you just learn a lot from them. 00:50:12.880 --> 00:50:16.880 So, for Soham and you, what does a week on a usual week look like? 00:50:17.519 --> 00:50:26.960 Uh, I think it's like for me, it's a lot of like kind of meeting uh with the tech team, thinking like building um like at this point. 00:50:27.039 --> 00:50:29.760 I think my like Claude is always on. 00:50:29.840 --> 00:50:38.159 I think one of the companies that I kind of really like right now is exc.dev, for example, uh, which provides really good DX around like machine for agents. 00:50:38.320 --> 00:50:41.679 So my like Claude is codex is always on on that machine. 00:50:41.760 --> 00:50:44.320 Like I'm just kind of like chatting uh with it. 00:50:44.400 --> 00:51:02.639 So it's like a lot of like like building new stuff, thinking through that, um, chatting with the team, showing how we can move faster, uh, kind of like what are the blockers that are affecting everybody's thing speed, and then thinking through like like having like product discussions and like what is next. 00:51:02.880 --> 00:51:05.119 So I think those are the things on my plate. 00:51:05.280 --> 00:51:20.960 I think uh on Sohoms, it's kind of like a mix of uh obviously product discussions with like a lot of GTM stuff, planning on how like that looks like, and kind of obviously uh going forward it will be some like uh chats with BCs, etc. 00:51:21.119 --> 00:51:21.599 as well. 00:51:22.320 --> 00:51:31.360 And uh uh can you name any top three founders that haven't raised funding that you really respect from your circles? 00:51:34.079 --> 00:51:45.519 I'm just thinking, I think I mentioned like Praty, for example, uh kind of like uh from X Compucio segment um who haven't received funding. 00:51:46.320 --> 00:51:47.760 That's a good one. 00:51:48.719 --> 00:51:56.639 Yeah, most of the like names that come to my mind uh kind of have already received funding or early in their journey, yeah, they have received funding. 00:51:56.719 --> 00:52:07.519 Uh kind of like uh uh I think another like like person ex uh composure like like Wenki started out recently. 00:52:07.679 --> 00:52:09.039 I think that might be one. 00:52:09.840 --> 00:52:12.239 Uh yeah, I I think those are the ones that are. 00:52:12.400 --> 00:52:19.760 Yeah, yeah, because any founders or investors uh sorry listening to this podcast that can uh reach out and benefit them. 00:52:20.079 --> 00:52:21.039 Yes, yes, for sure. 00:52:21.199 --> 00:52:25.440 Um yeah, I I think those are the two that comes to my mind that come to my mind right now. 00:52:26.079 --> 00:52:26.400 Yeah. 00:52:27.039 --> 00:52:28.000 Thank you so much, Kerani. 00:52:28.079 --> 00:52:29.679 It's been such a pleasure hosting on the podcast. 00:52:29.840 --> 00:52:30.320 Yeah, thank you. 00:52:30.400 --> 00:52:31.119 Thanks for having me. 00:52:31.840 --> 00:52:35.679 I enjoyed the conversation and loved it coming from you. 00:52:35.920 --> 00:52:36.639 Yeah, thank you.