Mukund & Madhav: How Solving Verification Enabled 7M Apps

1. 0:00 – 1:06

   Intro

   1. MUMukund0:00

      So I think now we are just truly seeing this unlock where people who, who were, like, really close to problem, domain expert and-- but have been blocked by, you know, technology barrier to sort of really express themselves are using Emergent to sort of build these things out.

   2. JFJared Friedman0:12

      There's just so much focus on AI is gonna replace jobs, knowledge work is going away. Like, what's that gonna mean for employment and civil unrest? But [chuckles] like no one's really talking about the fact that actually, like, if you have, like, some agency of interest and you want to start your own business and have autonomy over your life, like, you are empowering that at scale. [upbeat music] Welcome back to another episode of The Lightcone. Uh, unfortunately, Garry got called to jury duty and can't be w- here with us today. Uh, but we are really excited to be joined by Mukund and Madhav Jha. Uh, they're both twin brothers and founders of Emergent, which went through YC in summer 2024. Emergent's a platform that lets anyone build and ship production-ready software using AI agents. You guys are actually one of the fastest-growing companies I believe YC's ever funded. Um, I mean, the statistics you were telling us were mind-blowing.

2. 1:06 – 1:18

   What Is Emergent?

   1. JFJared Friedman1:06

      You have, in eight months since launch, seven million apps have been built with Emergent. Walk us through this, like, incredible growth you're seeing, actually. When did that hit a real inflection point, and how did that, that feel for you guys?

3. 1:18 – 2:09

   Founder Backstory

   1. MUMukund1:18

      So we-- Both are twin brothers. We actually, uh, you know, started programming when we were age twelve. Both of us came to US to do our PhDs. I dropped out of the PhD program, joined Google, and Maddy went on to, um... Was at Zenefits, then went on, uh, to start the deep learning team at Amazon. And, uh, we've been meaning to do a startup together for a long time. And, um, before this, I was, uh, running a startup in India called Dunzo, which was a hyperlocal quick commerce company. Um-

   2. JFJared Friedman1:42

      Dunzo was a big company, actually, right?

   3. MUMukund1:43

      Yeah, it, it was, it was really big. Uh, and, and we... We are almost a verb in India. So when people ship something, they say, "Dunzo it." Uh, and, uh, and I was managing a really large team of three hundred engineers, um, and, you know, and we have been sort of watching the deep learning field for a while, and we knew an inflection point is coming. One of the things that I observed when I was running this large engineering team was that software testing was the biggest bottleneck in shipping fast. Um, so when we started looking at, you know, what we want to build in AI, uh, that was the first idea we actually-

   4. JFJared Friedman2:08

      What

4. 2:09 – 2:52

   From AI Testing to General Coding Agents

   1. JFJared Friedman2:09

      year was this?

   2. MUMukund2:10

      This was '23 end.

   3. JFJared Friedman2:12

      Okay.

   4. MUMukund2:12

      Yeah. And, and so when we applied to YC, like, we applied with this idea of automating software testing. Uh, that was the first idea. In fact, we went to a lot of VCs with this idea. They thought it was too crazy. Uh, you know, and, and now looking back, it, it, it almost looks, uh, funny. And so we applied to YC with this idea, and, um, and when we were building this testing agents, we, uh, realized that if you can solve for verification, which is essentially, you know, you can solve the, the testing part, uh, you can actually automate all the software engineering. That was sort of our key insight, that like, you know, verification is the loop which sort of keeps agent running for a longer, longer period of time. And that's when we pivoted to looking at general coding agent as a space, and we, uh, started building a general coding agent.

   5. JFJared Friedman2:49

      And this takes us into 2024.

   6. MUMukund2:51

      This would... 2024.

5. 2:52 – 4:18

   Getting Ahead of the Market

   1. MJMadhav Jha2:52

      Summer of '24.

   2. MUMukund2:52

      Summer of '24. Yeah.

   3. JFJared Friedman2:52

      Yeah. Tell us what the landscape looked like. Like, how big was Lovable at this point, and just-

   4. MUMukund2:56

      I mean, nobody had started. Lovable had not started. I think Cursor was just, just getting, getting, getting started, um, and very, very early. Uh, I think Devin had just come out, uh, so, so really, really early. And, and we looked at this benchmark called Suitebench, which is essentially a benchmark. Now it's saturated, but at that point, uh, of time, like, that was the benchmark where all the coding agents were getting measured on. And we took on this challenge of, uh, becoming number one on that benchmark. And, like, we sort of packed ourselves in a room, uh, four of us, and said, "Okay, let's just look at this benchmark. How do we crack it?" That sort of set the foundation for Emergent, and we built, uh, you know, sort of coding agents, which became world number one on Suitebench, um, you know, in two months of time. And that was the time when we sort of discovered a lot of the fundamental truths about building with LLM, building with agents.

   5. JFJared Friedman3:36

      And your intended users at this point were presumably engineers.

   6. MUMukund3:39

      Yeah. At that point, we were, like, purely just a research company, just building coding agents. We were not thinking about a product. There was a time when we sort of invented the multi-agent system. We invented memory. We invented, like, how, how do we do agent-to-agent communication? How do you scale up test time compute? Uh, a lot of those things which, like, were sort of coming out. Like, we would, we would discover something, and we'll see three months later something come out in a paper. Uh, you know, a-and that sort of set the foundation for, for us to, to-

   7. MJMadhav Jha4:01

      So we were like Cloud Code before Cloud Code was a thing. Uh-

   8. JFJared Friedman4:03

      Yeah.

   9. MUMukund4:04

      But the, the paradigms like multi-agent, uh, orchestration, how do you use, like, different, different routings, a lot of those things we, we sort of discovered at that point.

   10. JFJared Friedman4:10

       I definitely wanna come back to that.

   11. MUMukund4:11

       Yeah.

   12. JFJared Friedman4:11

       Um, I'm curious at this point in the story, though, when did you sort of pivot into becoming a tool for non-technical

6. 4:18 – 5:22

   The Pivot to Non-Technical Users

   1. JFJared Friedman4:18

      users?

   2. MUMukund4:18

      Yeah. Yeah. So we actually-- Like, once we had this coding agent, uh, we actually went the enterprise route. That was the common wisdom at that point, that, "Hey, like, go to enterprise. Build for enterprise." And we spent, like, two, three months trying to, uh, you know, make our agents work within an enterprise. We found that it was too slow, and at the same time, we were internally started using Emergent's platform to build internal tools, internal software. And at that point, you know, we saw, like, Lovable was growing like crazy. Bolt was growing like crazy. Uh, so we thought, "Hey, why don't... We have this, you know, really strong coding agent. How do we sort of package it and, and, and bring it out in the world?" And we launched a very, like, small beta pro- uh, pilot, uh, almost, uh, in June, uh, last year, 20, uh, 25, and that really took off. And, and since then, you know, like, we-we have been just focused on solving, uh, problem for non-consumers. We in fact thought a lot of technical people would us-use us. But today, eighty percent of users who are on the platform are non-technical users with zero programming knowledge. Uh, and they're building, like, apps that, that run real businesses on top of today. So it's almost been a-

   3. JFJared Friedman5:13

      And they're based all around the world, right? Like, how many countries?

   4. MUMukund5:15

      Yeah, yeah. So they're, they're all global audience, eighty percent. Um, seventy, eighty percent are in US, Europe. Over one ninety countries right now.

7. 5:22 – 9:04

   Why Second Movers Can Win in AI

   1. JFJared Friedman5:22

      Something that we in- have talked a bunch about at YC internally is just, um, how does first mover advantage versus second mover advantage play out in the AI world? Certainly something that we've noticed, like if we look at some of our company, like Lagora entered the legal AI space after Harvey-

   2. MUMukund5:38

      Mm.

   3. JFJared Friedman5:38

      -but is, like, growing incredibly fast. So there was clearly... There wasn't maybe as big of a moat around being a first mover, um, as you traditionally think there is in software. When you guys made that sort of the pivot or the slight change in direction into non-technical users at a time where Lovable and Bolt are growing really, really quickly, uh-How did you think about that?

   4. MJMadhav Jha5:57

      There are like two to three different, different threads that I would want to pull. One essentially is that I think the, the model, uh, every new model generation actually is, is presenting a new opportunity of looking at the world. Like for example, when we started, GPT-4 was the, the first one that we sort of started looking at. And at the end of the day, the biggest problem that everybody was trying to solve was JSON parsing, like a structured output format. And we thought, okay, like the next model is gonna solve for it. Um, you know, like, let's not s-spend time on that. And I think with every new model, what's happening is that you need to s-to start reimagining the world. For example, like Opus is a different class of model right now. It's gonna enable extremely long horizon tasks. It's gonna, uh, enable like multiple agents coordinating together. And so I think like one of the advantages of starting, uh, second, right, is that you can actually, one, like learn from what is, what is not working, uh, for the, uh, current competition, right? And also I think you fundamentally start from a different starting point, right? Like where like your approach of the world is like very different. Like your, your imagination is, is really big, right? And I think a-and, and when we, we were starting, um, uh, Emergent, we realized that like a lot of the users that were going to, you know, um, some of these, these, these apps, they wanted to actually really build an app that works, right? And most of these were actually like really, really optimized for front-end prototyping at that point. So we started fundamentally reimagining that, okay, what would world look like if you could actually ship things to production? And our key insight was that to automate all of software engineering, you will have to build a platform that replicates what, what best engineering team do, like code reviews, automated testing, debugging, deployment, security, hosting. So we reimagined the entire platform from ground up saying, "What would an end-to-end platform look like?" And the real user need was actually to ship the product, not, not just the front-end prototyping. I think second thing is like, how do you sort of get the distribution? Because you're coming from behind, right? So even if your product is really strong. Uh, and fundamentally I think you'll have to enter the market with, uh, a really, really strong product, which is, you know, head and shoulder above what, what, what exists in the market today for people to take notice. Um, we were very confident about the product and, and so a lot of our focus, like in an early days once we sort of launched, was on how do we sort of rapidly scale up distribution. Um, we built out a, a large influencer network, and that was our initial sort of, you know, um, starting point for us. Like we used TikTok, Instagram, and partnered with a bunch of influencers to really, really spread the word out and, and that sort of, you know, kickstarted the whole thing for us.

   5. JFJared Friedman8:09

      So to me, sort of building the influencer marketing engine is like, um, it's like tactics to land grab.

   6. MJMadhav Jha8:14

      Yeah.

   7. JFJared Friedman8:14

      Like were you also thinking about just focusing on personas and specific sub-types of users you wanted to go after that weren't-- they either weren't being targeted by LevelBall or, or others or, or Emergent was a better fit for them?

   8. MJMadhav Jha8:27

      I mean, our, our thesis was that like there are a lot of users who would want to build serious applications, right? And that was our sort of target audience. And a lot of our marketing, a lot of our initial messaging was around that. Like, "Hey, come and ship, uh, uh, real software." What we did was like little bit broad, broad-based like marketing and, and but users that, um, you know, were coming to the platform that we would convert were users who actually wanted to ship a real, real app, uh, on the platform.

   9. JFJared Friedman8:52

      And was that in the messaging then?

   10. MJMadhav Jha8:54

       Uh, it, it was in the messaging, yeah. So, so we would say, "Come and build real apps." We would also use the common errors that you would see on other platform. You know, like, "Hey, don't, don't see-- don't face this error, uh, uh, on Emergent."

8. 9:04 – 18:21

   Building for Production, Not Just Prototypes

   1. SPSpeaker9:04

      It seems like a key insight for you. Basically, you went very hardcore in terms of being maximalist in engineering from your experience, having run large engineering teams of 300 engineers, having worked on deep learning teams at Amazon. You really knew how to architect the systems. Can you maybe, uh, share a bit how you built it? One of the, uh, cons of all these other big products like Level or Bolt is just that it's, it's difficult to get those into a fully usable. You can get to a prototype very quickly, but yours, you went zero to 100% very quickly, and that takes finesse. It's almost like that 20% gets 80% effort, like the Pareto principle.

   2. MJMadhav Jha9:40

      Yes.

   3. SPSpeaker9:40

      But you, you did more than that. The last 20% of that engineering-

   4. MJMadhav Jha9:43

      Yeah

   5. SPSpeaker9:43

      ... to build production was a lot of work, and that's a lot.

   6. MJMadhav Jha9:46

      Yeah. Yeah, and I think like the, the last mile that you mentioned, right? Is, is always what people neglect, that hey, you need to make sure that not, not only app gets built, it also gets deployed. And this is one of the conscious reasons why we chose to build our own infra on which the agent is like running. So like we provide like, uh, you know, cloud sandboxes. Uh, we don't outsource it to like some third party sandbox provider, which was also pretty popular at that time, right? So we, we built our own Kubernetes, uh, tech stack from ground up, uh, the container tech stack. And one of the insights here is that if you give your, uh, agents the same infra during the build time and the same infra during the deploy time, then the sort of like, uh, during this like deployment phase, you don't, uh, encounter those many problems, right? And the fact that we have our own infra also allows us to give like rapid feedback to the agent. So your agent is only as good as the feedback that you provide. Uh, so we build this like sort of infra and agent like, like sort of co-build it together and from the, uh, from, from day one. And to your point, right? Like, uh, because we, we focused on, you know, sh- building like, uh, ship-ready apps, which, which are production ready, which has-- which comes with back end and, and front end and everything. The tech stack we chose was also pretty unique to us. We have a Python backend, uh, server, we have a React front end server. Like most people would like typically go with like a much more like, you know, Node, Node-focused, Node-heavy, uh, tech stack, right? And, and this like server-client architecture where you can have like background jobs if you want to have background queues. So we knew that, you know, users who would, who would use this app, their ambitions are gonna go bigger and bigger, right? "Hey, I want to run a job which can like do this, uh, asynchronous video processing," you know, and they're gonna prompt it, and we wanted to support it from day one, right? And so it's the ta- same tech stack on which Emergent is built is what we expose to our end users, is what we expose to our agents, right? Uh, on the agent side, we were very early on the multi-agent architecture. Uh, so we knew that you want to be very frugal about your context management. So what you do is, hey, let the main agent, the driving agent, handle the, the main routine, but any delegated tasks that you want to delegate, you delegate to a sub-agent, be it like testing, be it like, "Hey, I want to do a design, uh, search," or, "I want to do like, you know, integration search. Like how do I integrate this unique API?" Um, and along the way when we were like finding or doing all of this, we were able to figure out, okay, all the trajectories that we are generating, we can kind of aggregate over time and like sort of build an, a long-term memory for the agent, which is very unique in the sense that, uh, your agent learns not just from your own session, it learns across the sessions. This is something I would say is one variant of continual learning, uh, that people are like, uh, interested in now. Uh, you would have noticed that people are more interested in skills. Uh, like people create like, uh, skills and, uh, the, uh, there's a new benchmark called Skills Bench, which shows like agent with skills outperform agent without skills. Uh, and interestingly, like those skills cannot be generated by agent themselves. Like if you generate those skills by agents they don't like, uh, match up to the performance. So we were able to do it in a way where the skills get auto, uh, you know, sort of, uh-

   7. JFJared Friedman12:33

      Generated

   8. MJMadhav Jha12:34

      ... yeah, generated based on previous trajectories. And we run it through a CI/CD process and then add it to the long-term memory. Uh, so all of that like compounds for us, right? So if your, if, if your agent was struggling to do a calendar integration three weeks ago, uh, today it is no longer struggling thanks to the, uh, the previous session where it was able to make it happen.

   9. SPSpeaker12:51

      So fascinating. So it learns on its own because I think one of the challenges of all these, uh, vibe coding app

   10. JFJared Friedman12:58

       platforms is at some point, the applications will get so complex that if you build it very simply, you would run out of, uh, the context window-

   11. MJMadhav Jha13:05

       Yes

   12. JFJared Friedman13:06

       ... for all the models because that seemed to be the, the bottleneck, and I think you guys architected your way out, so you kind of built a lot of, uh, what the state of the art is now, but way back a year before.

   13. MJMadhav Jha13:16

       Our coding agent is so powerful that we basically internally use it, uh, as a replacement for Cloud Code as developers, right?

   14. MUMukund13:22

       Mm.

   15. MJMadhav Jha13:22

       So we, uh, we are so proud of that, and, uh, but yet we don't want to expose that, uh, sort of ex- you know, power tool to our end non-technical user. And so we, even though we have this VS Code editor, we kind of hide it, uh, because what we've noticed is that non-technical users, they even get panicked as soon as they see-

   16. JFJared Friedman13:38

       Mm

   17. MJMadhav Jha13:38

       ... a diff, you know? Uh, we, we, we had a, like a fairly technical PM in our team, and, uh, like, he doesn't like, like JSON. You know, he's like, "No, don't show me, you know, I, I get intimidated." So building that user empathy, where you have that user empathy and building that agent empathy, you also have to empathize with your agents. What is, what is a- what is agent feeling like, right?

   18. MUMukund13:59

       We internally have a term called agent experience, right? That we measure that how, like, how, how is agents experience on the platform.

   19. JFJared Friedman14:04

       Actually, a really important point I think people don't realize is you guys actually, you actually started out essentially as sort of Devon cursor-

   20. MJMadhav Jha14:11

       Yes

   21. JFJared Friedman14:11

       ... in like the actual, like, coding agent world for engineers. You just made the choice to package it up for non-technical users.

   22. MJMadhav Jha14:19

       Absolutely.

   23. JFJared Friedman14:19

       So you're sort of like, moving almost in the opposite direction from like a Lovable. Like you have like-

   24. MJMadhav Jha14:23

       Right

   25. JFJared Friedman14:23

       ... the power. You have all of the actual, like, power. You just need to simplify the user experience.

   26. MJMadhav Jha14:28

       Right.

   27. JFJared Friedman14:29

       Whereas they like sort of have, they start with the user experience, and they're gonna have to develop the power over time.

   28. MUMukund14:34

       Right. And I think fundamentally it's, it's, like unless you start from, you know, a, a starting point which, which, uh, sort of solves all of these problems along the line, the whole software development life cycle, it's actually really hard to come from the other side and solve these problems, because you'll, you'll make some architectural choices which are very hard to reverse.

   29. JFJared Friedman14:49

       Do you have any more, I'm really curious, like any more examples of where sort of as you were engineering the system, you just sort of just trusted in the model? Like you mentioned JSON parsing, but was there anything else where you're like, "Let's not invest time in that, um, because like Opus four point five will, will solve it?"

   30. MUMukund15:05

       I mean, some of them has, has been, for example, um, you know, like library definition, some of the integrations that we have sort of built. Like, you know, we think that, you know, the next sort of models are solving for us. Similarly, like how do you generate unit tests? Some of those things that we, we actually like would have heavily prompted before. And the other thing that we are very conscious of is that how do we give more and more autonomy to the models as they, uh, the next generations come out? And the more autonomy you're able to give to the, the models, the, the better they perform. Like initially, like our harness was very strict and, you know, like we would, we would tighten it up, um, and, and slowly, like what we're observing is that as these models are getting larger and larger, more, more, more, uh, efficient, like, you know, like the more control you give to the model, uh, decision-making, the better the, the harness gets.
9. 18:21 – 24:40

   Live Demo: Building Apps with Emergent

   1. MUMukund18:21

      Could we see a demo of Emergent? Oh, yeah, sure. Yeah, so this is how, what, uh, Emergent interface looks like. And, uh, I'm gonna like put a prompt where like, because we were coming for this podcast, we-- I thought like, you know, there should be an app which lets you practice- [laughs]

   2. MJMadhav Jha18:32

      ... you know, uh, podcast questions, or maybe you are going to a job interview and you want to practice questions, right? So you can build a full stack app on Mo- on Emergent, you can build a mobile app. Our prompt engine is smart enough that once you give it a prompt, uh, it will figure out that this is talking about a mobile app. So it'll figure out like, hey, the, the right agent to use is, is a mobile app builder, right?

   3. MUMukund18:50

      So even though you have like selected the wrong tab-

   4. MJMadhav Jha18:52

      Yes

   5. MUMukund18:52

      ... it's just like, ah, like-

   6. MJMadhav Jha18:53

      Yeah, yeah. The behind the scenes-

   7. MUMukund18:54

      I got you

   8. MJMadhav Jha18:54

      ... auto-- Yeah, I got you, right. So while, while this is, uh, running, let me quickly also, uh, show you a few, uh, user apps. So this is by somebody based out of Illinois. Uh, he's a sort of has a business of audio-video setup, uh, that they do like on a-- as manually, right? So basically, whatever this kind of like intake form they would have taken through spreadsheet and, and other calls, they basically build this out without any, uh, coding background knowledge, right? Like, hey, this is the kind of AV setup I want. Um, so you, you, you go and you build your room, and then you, you get, uh, it's a lead gen sort of a form, but this is a fairly full stack app.

   9. MUMukund19:28

      One thing I noticed about that is like the design is really good, like the icons, like it just like it looks like a well-designed app.

   10. MJMadhav Jha19:34

       So we, we have actually spent a lot of time on like-

   11. MUMukund19:36

       This one, yeah

   12. MJMadhav Jha19:36

       ... making sure design is actually good. And-

   13. MUMukund19:38

       Yeah

   14. MJMadhav Jha19:38

       ... our, our, like, so earlier there used to be a big trade-off between design and functionality. Like if you're optimizing for design, like your functionality w-would not be that strong. Uh, and so we had to figure out like how do we sort of, you know, share the context in a way where design also gets better. There's another sort of person based out of Norway. He, he sold his, uh, previous business to a PE and, and realized how much lawyers have to struggle with spreadsheets and other things, so he built a CRM for lawyers. He, he describes himself as like business developer. I, I like the word he used, like, "I'm a business developer," and he, he has-- doesn't have a programming background. So a lot of CRM-related apps, we are seeing small businesses. It's your second monetization a-avenue, right? And so like one of the unique things with Emergent is that before the agent goes off to build things, it asks you for some clarification because agent wants to make sure that it understood your, your, uh, requirements properly.

   15. MUMukund20:23

       Mm-hmm.

   16. MJMadhav Jha20:23

       And, uh, another thing is that non-technical users probably don't know the concept of API key. How do I get an OpenAI API key? So in this particular case, I can just say, "Hey, use Emergent LLM key," so you don't have to worry about getting-

   17. MUMukund20:34

       Mm

   18. MJMadhav Jha20:34

       ... API key from third party.

   19. MUMukund20:35

       Mm.

   20. JFJared Friedman20:36

       This feels like a good example of what you were saying.

   21. MJMadhav Jha20:38

       Mm-hmm.

   22. JFJared Friedman20:38

       Um, 'cause this is sort of like the ask user question skill in Claude Code, but you just like abstract that away.

   23. MJMadhav Jha20:43

       Yes.

   24. JFJared Friedman20:43

       But it's like built into the experience for someone who had-

   25. SPSpeaker20:45

       No idea about

   26. MJMadhav Jha20:46

       Absolutely. I can be very like casual here. I can say, "Hey, uh, the-- for the first one, use Emergent API key. Rest assume good defaults and then go." This is the first time I hand off the agent, and like at this point, I can just like close my laptop. We also have a mobile app, so you can like-

   27. SPSpeaker20:58

       Mm-hmm

   28. MJMadhav Jha20:58

       ... on the go-

   29. SPSpeaker20:59

       Mm-hmm

   30. MJMadhav Jha20:59

       ... keep trying to prompt agent if, if agent requires a-additional, uh, thing. Once it's done, uh, you see a preview of your a-app. So here, for example, in this case, I can practice what is my origin story. Uh, I can record, uh, what my origin story is, and I can keep going to, you know, various questions, uh, eventually-
10. 24:40 – 29:04

    How Emergent Hires and Runs a Lean Team

    1. MJMadhav Jha24:40

       GitHub.

    2. SPSpeaker24:40

       To talk about how you run your team, the way you hire must be very different. I mean, you're a very lean and small team. How do you hire for engineering?

    3. MUMukund24:48

       Yeah. So we, we actually from, from day one have been very conscious of the kind of team that we want to build. And essentially like we index on two things. One is problem-solving, like how good are you at problem-solving? Uh, and second is ownership. Like we think that people who can like really take ownership, um, you know, like we index on that. And a lot of our early sort of hires were people like, you know, we, we were really obsessed with like top hundred ID rankers. So we had this like program going on where like I told, you know, our team that, "Hey, we must hire like top hundred ID rankers." Uh, right now I think we have like ID rank one, ID rank 12, um, all, all of those people working with us. And a lot of the initials had also came from Danzo, so I-- because I was able to build like a really, really good team. We were able to get some, some initial folks from that. The focus that, that we have is, is essentially like one or two people doing work of what a company would be doing. For example, our deployment, which almost mirrors what, what Vercel would look like, is done by two people, right? Our memory, like where you have like multiple startups solving for memory, is just built by one person. So I think like we, like we give way more responsibility to people, and I think people are generally attracted towards harder problems that they want to solve.

    4. SPSpeaker25:50

       Where is your team located?

    5. MUMukund25:51

       So most of the team right now is in Bangalore, uh, in India office. Uh, we have a very small office in SF, like three to five people here.

    6. SPSpeaker25:57

       And you guys yourselves, you're kind of like split across both countries. Can you maybe just explain-

    7. MUMukund26:03

       Yeah. So, so in that-

    8. SPSpeaker26:04

       ... how, how the setup works?

    9. MJMadhav Jha26:05

       Yeah. So I mean, I, I, I live here in SF. I've been in like, uh, you know, Bay Area for like last 10 years.

    10. MUMukund26:10

        I split half my time in SF, half my time in Bangalore, uh, constantly jet lagged.

    11. SPSpeaker26:14

        I think you guys are probably the most successful AI company that's... It's not fair to say you came from-- like it's an Indian company, but that's got like significant presence in India. Um, why is that?

    12. MUMukund26:25

        I mean, I think it's like when I went back to India, uh, you know, after Google, and I always had this thought that why is there no Google or Facebook from India, right? So like from day zero, I was thinking, you know, even though I started Danzo, it was an India, India-focused company at that time. And when I was starting, uh, the second company, I, I always thought like, "Hey, there has to be... You know, like we have so much talent. We have, you know, so a lot of now capital available. Everything is available India. Like, why is-- are people not building globe-- truly global tech-first companies from India?" And, and that was the ambition that, that we started with. And in my opinion, I think a lot of it is with, you know, like just your ambition. Like if you, if you just dream big, if you're able to sort of really, really, um, think, uh, global from day zero. I think now because internet i-is sort of fully penetrated, people, people can actually get understanding knowledge from everywhere. I think every single, you know, country has an opportunity to build for global audience. And if you have that sort of mindset, that ambition, I, I, I think, I think lot-- uh, we'll see a lot more companies coming out of India doing the same.

    13. SPSpeaker27:16

        I'm curious to hear what it's actually like sort of on the ground running this sort of like split country

    14. JFJared Friedman27:23

        ... company where the team is mostly in India, but the product is overwhelmingly used in the US and w- Eastern Europe. It's not a product for the Indian market at all. What is it like running this company? How would it be different if you had built a normal Silicon Valley style company that was all based here?

    15. MUMukund27:39

        Internally, we have like really, really set really high standards, like as a, as a, as a global sort of product. I mean, both in hiring, both in like the way we sort of develop product. Uh, and I think us spending sort of time here also, also helps. Like one of the things that we do really religiously is everybody talks to a customer once a week, twice a week. Everybody-

    16. JFJared Friedman27:54

        Like everyone in the entire company.

    17. MUMukund27:55

        Every-everyone in the company, right? Uh, they talk to a customer. Everybody does customer support. So like we were like a really, really small engineering team, like 12 people team, and one person was always on call for customer support. It, it was really hard decision for us because, you know, you are a really small team, you need to ship really fast and then move like one of your best engineers out to do customer support was really hard. But I think that really, really helped us build the customer empathy from day zero. And I think given that like a lot of our distribution happens online, like, you know, like the teams are able to learn, uh, from digital things and build for it. But I think us building that customer empathy from day zero, like talking to our users, like really, really helped us bridge the gap, uh, you know, uh, in terms of like what our users want, uh, today. And, uh, it's funny because like when we launched, my first like five days, I was just glued to a desk doing customer service, uh, support, uh, only. And most of the customer requests were coming in, in a different language like, you know, French, German, because a lot of, lot of the users were global. And thanks to AI, like we were able to understand that, reply to that, and I think that, that, that, uh, you know, like is also helping, you know, us bridge the gap there. Yeah.

    18. MJMadhav Jha28:51

        And we are hiring here in SF.

    19. MUMukund28:53

        [laughs]

    20. MJMadhav Jha28:53

        So, uh, if anybody's, you know, interested in, uh, you know, joining, uh, in various positions like be it res-research across the board, like backend engineers, frontend engineers, we are hiring here in SF and in

11. 29:04 – 34:04

    Is SaaS Dead? The Rise of Personalized Software

    1. MJMadhav Jha29:04

       Bangalore.

    2. JFJared Friedman29:04

       I'd love to go back to what we were talking about sort of regarding personalized software. And what do you think the appli- implications are for SaaS in general? You know, like I guess the provocative question [laughing] is, is SaaS dead now? I mean, you guys essentially killed Asana for yourself.

    3. MUMukund29:17

       Yeah.

    4. JFJared Friedman29:17

       Like, is that bad for Asana [chuckles] and other SaaS companies?

    5. MUMukund29:20

       I mean, I definitely think that like the current, um, way, uh, the SaaS is existing today needs to change, right? I think like I feel there are two like sort of massive headwinds. One is more and more of these SaaS workflows are gonna get consumed by an agent, right? Like, so, like, um, you know, unless your SaaS company pivots into like an agent first company, uh, you know, I, I think, uh, that's going to be, uh, hard to sort of survive. And second headwind is obviously like, you know, like people would want more and more customized software like which they can build on Emergent, just like we built, um, you know, um, our own DoIT, uh, project management tool. And we are seeing a lot of these people, um, you know, building these internal tools, uh, the software o-on, on platform like ours. And like I feel the nature of software itself is changing. I think a lot more so software will, will become agentic in nature. Um, a lot of people who are building on Emergent today, like roughly twenty percent of them are actually agentic apps. So people are actually, you know, embedding our own e-Emergent agent inside those apps to sort of pow- you know, power much of the workflows.

    6. JFJared Friedman30:15

       Do you have some interesting-- That sounds really cool. Any interesting examples that people do?

    7. MUMukund30:18

       Yeah, I mean, like the, um, uh, app that Maddy was just showing, uh, you know, the, uh, CRM for, uh, lawyers, that is an agentic app where, you know, an agent can take a workflow and, and run, run through the process. The software itself is now morphing into, you know, agentic, like a lot of, lot of people who just want to, you know, build agents that can actually just do-

    8. JFJared Friedman30:34

       Mm-hmm

    9. MUMukund30:35

       ... you know, lot, lot, lot more of the work, uh, on its own.

    10. SPSpeaker30:37

        Where do you think this goes as, uh, agents', uh, horizon for tasks gets longer and longer? I mean, one of the, the Meter-

    11. MUMukund30:44

        Meter chart, yeah

    12. SPSpeaker30:45

        ... chart is one of the ones that was very shocking recently.

    13. MUMukund30:47

        Yeah, I think that's the chart of the year, I would say, right? Like the, the Meter's, um, exponential growth and, and like four, four point five was at like, I think four hours and four point six is at ten hours. Uh, and we are internally sort of now like, you know, experimenting with agents forms where agents can actually like work, uh, for a much longer horizon and multiple agents can sort of coordinate on a single task. Um, early results are like pretty, pretty exciting. Um, you know, we'll see. I think, I think by end of the year you'll have, you know, agents which are running twenty-four hours, uh, and like maybe hundreds of agents collaborating on this single task. Um, and that's where, that's where we sort of see the future going right now.

    14. SPSpeaker31:22

        How are you building for that?

    15. MJMadhav Jha31:23

        People's ambitions are increasing, right? Like, and so like we, we wanna like give agents more autonomy, right? And so like the, the, the main thing is to make sure that the trajectory doesn't get derailed. So you always want to have like an overseeing agent, right? Like, so it's like let's say a few agents are collaborating, then there is an overseeing agent as well, which is like parallelly like monitoring the overall task, right? So, so we are experimenting with many different architectures, right? Like something even as simple as like just, uh, you know, you would have heard of this Ralph-Wiggin loop kind of a phenomena, right?

    16. MUMukund31:51

        Yeah.

    17. MJMadhav Jha31:51

        Like, so the idea that, hey, like just keep poking the agent, "Hey, continue until it's done." And all of that is only possible if there is a good verification loop, right? So it comes back to, hey, are you able to give autonomous verification feedback to the agent? Like, was the job done? So a lot of our work internally right now is in fact still going on, on building be- best verifiers. There we are actually, uh, doing some custom fine-tuning as well. So, uh, we are very careful about like not directly competing with the models in the sense that we don't want to like build a Opus four point five alternative right away, but we do want to augment it through our custom fine-tuned verification layers. Uh, so, so some of the fun stuff we-- on the research side we are doing is on, on that side.

    18. JFJared Friedman32:29

        How do you think about some movement in the opposite direction? I mean, we talked about sort of like the models themselves maybe getting more powerful and what does that mean for everyone building on top of them. But how about, uh, at least some of the model companies are explicitly trying to build applications and own the application layer themselves. If one of those companies decides like, you know, Claude code for non-technical users is a really valuable application to build, what implications does that have for you and for startups in general?

    19. MUMukund32:53

        I think, I think eventually, eventually, I think like, uh, do you understand your customer's requirement really, really well? Are you building closer to them? I think, I think all of those fundamentals of like startup building remains the same. And I think, you know, like for us, like as long as we are focused on like really, really understanding our users need really, really best, I think, you know, we'll, we'll compete on the product side.

    20. JFJared Friedman33:08

        Do you think-- I mean, maybe do, do you think about all the model companies as like the same or are there differences between them?

    21. MUMukund33:14

        If you look at the models themselves, right, like they're very different. Like, for example, you know, um, Opus is obviously a workhorse. Um, you know, like, um, Codex is really good in backend debugging. Uh, Gemini is really good in front-frontend. So I think all of these models have their own behaviors and, and, and one of the like-- A good thing for us is that we can actually utilize these, uh, spikes that model have like, to, to provide the best experience to the user. Um, and I think eventually, like at least my worldview is that most of these models are gonna get, get really commoditized, like where all of these models will have similar behaviors. Uh, they'll have, you know, price, price competitiveness, um, be-between them and, andYou can already see like, you know, like open source is like maybe three to six months behind, right? And, and there's enough optionality for us to sort of really, really build the layer on top where we really meet the user where they are and, and sort of support them in, in sort of their, their journey. Who understands the customer needs really, really well and, and is able to build for that i-is going to sort of win the space.

12. 34:04 – 39:32

    The Future: Niche Apps, Solo Builders and AI Agency

    1. SPSpeaker34:04

       Users have built seven million apps with Emergent. What are all these apps? Who, who are the users, and what surprised you seeing what people do with it?

    2. MUMukund34:11

       The users who are coming to platform for us are generally people who want to build a serious apps, people who re-like really have a business use case that they want to automate or they have a business idea that they want to launch. Um, primary users who are coming to us are small, medium business owners. They're running their business today on, on email, WhatsApp, spreadsheet, uh, and would have gone to a dev shop to sort of build a custom software, um, to run-- automate their business. They're coming to us. And if you look at the price point that, you know, we are bringing down, it would have costed you like five hundred thousand dollars to build the software. Now you can build it for five thousand dollars completely on your own. Um, and, uh, that is the kind of, you know, like unlock that we are sort of bringing to the world right now. Uh, second, for example, this morning I was talking to a user, Christie. She's based out of Alaska, uh, and she built this-- She's a clinical psychologist. Uh, she's also, uh, a sports coach for equestrian, the horse riding. And she wanted to m-marry these two fields like, you know, like that. She has a lot of insights on psychology side. She has a lot of insight on, on, uh, horse riding side. And, and she said she looked around everywhere to find an app that does that, and she couldn't find one. So she wanted to build one. She actually went to a dev shop, uh-

    3. SPSpeaker35:14

       That is definitely the intersection of where she is [laughing] .

    4. MUMukund35:17

       Yeah. And, and, and she went to a dev shop in Nova Scotia and tried to find somebody who can build it. Uh, they were charging her a bomb, so she, you know, discovered Emergent, started building, uh, out, and she, she just launched her app like a couple weeks back. It's called EquiMind on, uh, on App Store. Uh, and it actually marries, you know, like her insights in psychology and, and, and, uh, into this, this, uh, sports coaching. Um, she has like hundreds of users right now using the-- using the platform. And I think that is a lot that we're trying to build. Like, you know, people who would have been, um, who have had an idea for a long time, people who are like really, really domain expert, very close to a problem, uh, can now go and build, build things up. Um, we also have like a lot of solopreneurs building on platform like who would have had to go and hire a technical CTO, uh, to, to build these apps. And the success that we are seeing on the platform is like recently somebody pinged me that, "Hey, like this company has raised like four million dollars, uh, on an app that was built on Emergent." Uh-

    5. SPSpeaker36:05

       Really?

    6. MUMukund36:05

       Yeah, yeah. And, and I need to get their permission to, to share more. But yeah. And so I think now we are just truly seeing this unlock where people who, who were like really close to problem, domain expert and but have been blocked by, you know, technology barrier to sort of really express themselves are, are, are, you know, like using Emergent to sort of build these things out.

    7. MJMadhav Jha36:23

       A-and also like one thing, uh, these people tell us that like, uh, it's not just about money. Like, hey, I can give money to the dev shop, but a lot of-- a lot get lost in the translation when you're trying to express your idea to the-- through a developer and they say, "Hey, I know what I want to build. If I could just say it out my-- out loud myself, I would, I would do a better job." And so the, uh, the Norwegian, uh, person I was talking about, like he said that, "Hey, in my team, I am the only builder. I don't even bring in anybody else because I know exactly what to build and like others focus on the business aspects of it." So this like single solopreneur sort of attitude of like, I'm gonna do it myself. I have the domain expertise. Nothing is lost in translation. Uh, that kind of agency is what people are looking forward to with these kind of platforms.

    8. JFJared Friedman37:01

       Yeah, I think it's a really important story that doesn't get told enough actually, is like what you're building is really necessary for society. There's just so much focus on AI is gonna replace jobs, knowledge work is going away. Like, what's that gonna mean for employment and civil unrest? But like no one's really talking about the fact that actually like if you have like some agency of interest and you wanna start your own business and have autonomy over your life, like you are empowering that at scale.

    9. SPSpeaker37:26

       It's so cool the like amount of human creativity that you're unlocking. Like who would have thought that the thing that the world needs is an app that marries clinical psychology with horse riding.

    10. JFJared Friedman37:34

        Yeah.

    11. SPSpeaker37:35

        Um, and in a world of limited software, that app would never have been built. But in a world-

    12. JFJared Friedman37:39

        Yes

    13. SPSpeaker37:39

        ... of unlimited software, you can build that and seven million other apps that like nobody would have ever gotten to build before.

    14. MUMukund37:45

        Yeah, we are getting to the niche of niches-

    15. SPSpeaker37:46

        Yeah

    16. MUMukund37:47

        ... today. Yeah.

    17. JFJared Friedman37:47

        I mean, so, Pete, this is like just an extension of the trend PG wrote about a while ago, right? And sort of like maybe coming out of the Second World War, you had sort of like a few big companies and people like built whole careers, hopefully staying at like IBM or whatever for a couple of decades and then retire. Then the startup wave came along and suddenly like the world becomes higher resolution. People are like, "Oh, maybe I should start my own company or at least join a smaller company and work at multiple companies or found multiple companies." And like the next extension of that is just everybody like runs their own-

    18. MUMukund38:18

        Mm-hmm.

    19. JFJared Friedman38:18

        ... like business that's at the intersection of like clinical psychology and horse riding-

    20. MUMukund38:22

        Yes

    21. JFJared Friedman38:22

        ... um, and finds an audience and, and life, uh, livelihood that way.

    22. MUMukund38:27

        Yeah, I mean, we are excited about so many ideas coming to life. Like we really want to like reduce this gap between idea and reality and, and, you know, truly enable people, uh, to express themselves and, and, and really, really like have this Cambrian explosion of ideas like which is great for YC.

    23. JFJared Friedman38:41

        I would argue it doesn't have to be actually. Like the whole like I think it's just really interesting the whole like explosion of being able to start businesses that aren't like venture funded, that aren't trying to raise lots of capital.

    24. MUMukund38:50

        That's right.

    25. JFJared Friedman38:50

        That it's just like one person like following their passions and like having control over their life. I think it's like it's really, um, uplifting message.

    26. MUMukund38:59

        Right. And I think we're just in the early innings of this right now. Like I think, I think this, this explosion is gonna grow and, and, and we'll see larger and larger, you know, projects being built on, uh, Emergent. Yes.

    27. JFJared Friedman39:07

        Yeah. Okay. Well, that's all we have time for today. Uh, Mukund and Madhav, thank you so much for joining us.

    28. MUMukund39:13

        Thank you.

    29. JFJared Friedman39:13

        This was a really fascinating conversation, and congratulations on all the growth, and we're excited to see where things go from here.

    30. MUMukund39:18

        Thank you. Thank you so much for having us.

Episode duration: 39:32