His startup powers OpenAI's Voice Mode. Last month, they became a unicorn. | Russ d’Sa, Co-Founder of LiveKit

Russ d’Sa (00:00:00) :
It would be pretty cool to build a demo of a computer that felt like a human, but instead of you texting with it, you're talking to it. So I built this demo pairing kind of our technology and GPT together, tweeted it out and I'm going viral for sure. My first time ever, I'm so excited for it and then barely anybody noticed.

Pablo Srugo (00:00:19) :
No way.

Russ d’Sa (00:00:20) :
So I went to this five hour long lunch with them. They were, like, can we buy you? Can we license? Or we're going to kill you. They offered us a super low ball offer and we're like, no. And they're like, well, if you have an AI idea then, we'll 10x the offer. I'm like, I don't have a way out. We're like a live streaming company, video conferencing, I can't.

Pablo Srugo (00:00:36) :
Oh, God.

Russ d’Sa (00:00:37) :
Sorry, guys. I get an email from OpenAI, and it's like, hey, so we signed up for LiveKit Cloud in secret with a personal Gmail address three weeks ago. And we built this voice interface for ChatGPT, on top of you guys. And now we're fans, and we want to talk commercial. What was really interesting is that right when VoiceMode launched, OpenAI put this meeting on our calendar, and we go into this meeting. And they're like, so how do we scale this thing? No joke, my answer was, you're OpenAI, why are you asking me? In the early days, a Salesforce came to me, and they were like, so first question, who do you guys work with? Who do you guys power? Can you handle our scale? I'm like, oh, well, we power ChatGPT. They're like, all right, next question.

Pablo Srugo (00:01:17) :
Done.

Previous Guests (00:01:19) :
That's product market fit. Product market fit. Product market fit. I called it the product market fit question. Product market fit. Product market fit. Product market fit. Product market fit. I mean, the name of the show is Product Market Fit.

Pablo Srugo (00:01:31) :
Do you think the Product Market Fit show, has product market fit? Because if you do, then there's something you just have to do. You have to take out your phone. You have to leave the show five stars. It lets us reach more founders and it lets us get better guests, thank you. Russ, welcome to the show, man.

Russ d’Sa (00:01:48) :
Pablo, thanks so much for having me, dude.

Pablo Srugo (00:01:50) :
You've been having a wild ride, not just at your Startup LiveKit. You just raised a big Series B, $45 million last year, and then you're just mentioning you had twins. How long ago was that?

Russ d’Sa (00:02:02) :
They're actually three months old today, so they're still pretty fresh. I'm still in the newborn trenches.

Pablo Srugo (00:02:10) :
Well, congrats, man. Having one kid is a big deal. Having twins and a startup at the same time must be just absolutely insane.

Russ d’Sa (00:02:16) :
That's right, raised 45 million, raising two girls at the same time. Yeah, it is pretty wild. I feel, I've done more adulting in the last twelve months than any person ever.

Pablo Srugo (00:02:29) :
Oh, man, that's crazy. So we'll get into the story here at LiveKit, which started about five years ago. I guess, in 2021, but maybe a little bit of background. Because I know you had some other startups before that. So maybe tell us a bit about kind of your history and maybe a little more about that. Exactly the previous startup, right? Evie Labs.

Russ d’Sa (00:02:49) :
Yeah, so this is my fifth company. You know, the first one that's really kind of started to do well. But I grew up always wanting to start a company. My dad was in tech startups in the 1980s and 1990s during semiconductors and GPUs, and DSL, and all of these kind of foundational technology shifts. Mostly on the hardware side and, you know, I got out of school and I'm like, I'm going to start a company. So I went and I joined Y Combinator, the 2007 class of Y Combinator. So it was the fifth batch. Did a company then, and, you know, I was a kid right out of school, and actually most of the founders at that time were just kids out of school.

Pablo Srugo (00:03:24) :
That's the OG, yeah, YC. That's crazy.

Russ d’Sa (00:03:27) :
Yeah, it's definitely OG Batch. You don't see many pre 2010 founders still going to the YC alumni events. So it's always kind of a shock for people when I run into them there and they see my name tag, and the batch that I was in. But that's also where I met my co-founder of LiveKit. It's also where I met my co-founder. I mean, it's the same co-founder from Evie Labs. So, you know, we met way back almost twenty years ago in that YC batch doing separate companies and then I've kind of in particular always sort of done a company and then oscillated to joining a company. I feel like when you do a company, you kind of get tunnel vision. You're super focused on this thing that you're building and trying to solve this one problem, and then you don't really pay attention to what else is happening outside of that one problem you're trying to solve. And so I like to go and join a company for a while, take a bit of a breather, and get a better lay of the land, and what's changed, and what's new and what are the new opportunities coming up. So I've done that, I went back to 23andMe after my wife saw a company when 23andMe was really small, this genetics company, and then I went to Twitter. I was the seventy fifth employee at Twitter. Working on a lot of cool stuff there, left, started a company, that one failed and then I started another company, and that was Evie. And Evie was kind of in the, we pivoted many times, but it was in the content recommendation space using machine learning or deep learning at that time to recommend content. And sold it to Medium, the blogging platform.

Pablo Srugo (00:04:51) :
It was an algorithm that you would wide label to other websites that had content? Or where was it leveraged?

Russ d’Sa (00:04:56) :
Yeah, exactly. It was more, we had kind of two different products. We had an end user product that was integrated across all Android devices on Verizon phones. So we had a lot of users, and it was replacing Google Now. So when you swipe right, on iOS, that was Siri. On Android, that used to be Google Now. But for Verizon devices, we were actually that screen and we would deliver content there. You could read news articles, and we would figure out how to recommend more content to you. Kind of like a news feed, but we would also sell the recommendation API or service for other companies to use to recommend their content. Then we sold it to Medium, and LiveKit started after that. Which I'm sure we'll talk about.

Pablo Srugo (00:05:37) :
How big did Evie get? You know, revenue wise, employee wise, high level?

Russ d’Sa (00:05:41) :
Yeah, Evie got to about fourteen people, I think, at peak. When we sold the company, it was maybe eleven or something like that. So not too big, we raised about $13 million across a couple of rounds, a seed and an A. And then revenue wise, you know, revenue, we never really figured out. That's actually part of the reason that we ended up going and selling the company. And deciding to stop working on it, was we had a lot of users. We had about two, two and a half million DAUs on that product.

Pablo Srugo (00:06:13) :
Oh, wow.

Russ d’Sa (00:06:13) :
Right.

Pablo Srugo (00:06:14) :
Okay.

Russ d’Sa (00:06:14) :
It's daily actives on it across both our direct to user product that was in the app store and also through this integration with Verizon. And across both, yeah, total about two and a half million DAU. So a lot of users, but we just couldn't figure out how to monetize it. We decided to try to use ad networks, and cost per install was this thing that Facebook had a really strong handle on. We thought we could replicate that, get the same kind of CPI, but we were beholden to using Verizon's ad networks that they had acquired, and they're just much worse at targeting. And there's not enough contextual data, and just a news feed versus all the information Facebook has about you from a demographic perspective. And so we just couldn't extract the same level of revenue. I think we got to about a million in ARR from that, but you know.

Pablo Srugo (00:07:03) :
And was it a good exit or more of an aqua hire situation? What was kind of the dynamics of it?

Russ d’Sa (00:07:08) :
Yeah, so that was, we actually had started to try to sell the company, and we kind of lost our passion for it. It wasn't looking like we could build a strong business off of a pretty large user base and so we initially started with looking for an acqui-hire. So we hired a banker, and we thought, OK, maybe a banker will help us facilitate something better than an acqui-hire. At the same time, we'll go talk to companies.

Pablo Srugo (00:07:32) :
We can actually take some time on this tangent. Because, you know, this idea great companies are bought, not sold, blah, blah, blah and it's true. But when you have a company that, for whatever reason, either you're not passionate anymore or isn't working to the level you want it to. What you can actively do to sell your company is not at all well understood. Frankly, I don't feel like I have a playbook either. It is a bit of a dark side, so I'm really curious. Exactly, what you did and ultimately what worked.

Russ d’Sa (00:07:56) :
Yeah, I think a few interesting experiences and learnings came from that. So the first one is, you know, everyone's like, OK, you want to sell your company, you got to go get a banker. So we, when we got a banker, right.

Pablo Srugo (00:08:08) :
Classic.

Russ d’Sa (00:08:09) :
Yeah, classic and so we went and found this banker through some referrals of friends that had sold companies, and we started to work with this banker. One of the first questions they asked us was actually, how much runway do you have? How long can you stretch out your operating life? And so we had about, call it, a year, a year and a half of runway left. And, you know, we decided we weren't going to try to keep going, try to go back to the well and raise another round. We weren't really passionate anymore, we just wanted to find an exit. So, but we had the runway, and he said, OK, look, you have a year, a year and a half of runway. That's a pretty good amount of time if you want to maximize your ability to exit, or, if you want to maximize the amount you exit for, the recommendation is to try to find a very strong strategic partnership with someone first. Kind of prove out some kind of shared win that you can achieve during that partnership where they depend on you for a significant portion of whatever it is, product that you're delivering or value you're delivering and then you have some leverage in terms of the price. When it gets to the buyout table, right? When those discussions start and so we started to look for, and he used some examples from the past of companies he sold. And how those things ended up playing out. And where they got leverage points in discussions for acquisition. And so we started to look for these partnership opportunities that we could find. Of course, Verizon was one of them just, because we already had this product distributed across their devices, and there was some other drama that derailed that one. But fundamentally, we were trying to look for some kind of strategic alignment with a partner to build some new product with them and then parlay that into an acquisition that was going to be a full acquisition and favorable. But we also needed the hitch. So we didn't know if we would get that partnership in a year's time. So we started to have these acqui-hire conversations that were purely, we're going to just talk about talent, right? We're not going to sell the distribution we have. We're not going to sell the technology that we have. We're primarily just going to try to sell the team.

Pablo Srugo (00:10:12) :
And was that also through the banker? Those acqui-hire conversations?

Russ d’Sa (00:10:16) :
Those are separate.

Pablo Srugo (00:10:17) :
Got you.

Russ d’Sa (00:10:17) :
Those are me going to friends and being like, hey, I have a friend at Robinhood or hey, I have a friend at Uber, etcetera. One thing that I will say, though, about going and starting those conversations is, we did figure that even though we were looking for an acqui-hire as the context for those conversations. One thing we did realize is that we have potentially, we could get more value for the company if we can also sell them not just the talent but the technology, and then distribution is kind of like a distant third. I mean, our distribution is kind of locked into the product that we were delivering and also because half of our usage was through this Verizon channel. That's a very specific distribution channel that not the vast majority of people might want to pay for.

Pablo Srugo (00:11:02) :
By the way, just to interrupt for a second, it makes a lot of sense to go that way because you think about anything you're selling. You want to sell from a position of strength, right? So the partnership piece makes sense because you're trying to create win wins. You're not going in there and saying, listen, I'm going to fail, please buy me. Which is, you know, the worst thing you can do. You're saying, hey, we've got this thing, do you want to partner? So it's a position of strength. So that makes sense. The acqui-hire may not seem at first blush, but it is a position of strength. You're going in there saying, look, yeah, the product market fit didn't work. Yeah, we didn't get crazy traction but the people we have are top notch, stellar, like they're all going to get great jobs. It's not a problem.

Russ d’Sa (00:11:35) :
Correct.

Pablo Srugo (00:11:36) :
It's just, you know, we'd rather all keep working together as a group and maybe we can do it here. And that actually is a position of strength more so than selling technology. Because selling technology, the first question would be, well, if it's so good, why aren't people buying it? If it's so good, why don't you have crazy customers, right? Whereas you go with the acqui-hire, and if they're interested, oh, by the way, we also have this tech. And now you maybe craft a story that gets you something more.

Russ d’Sa (00:11:56) :
Exactly, yeah and so we kind of looked at it this way, where the first level of what we were going to try to sell was this really talented team. It was only thirteen of us, or eleven by the time we sold, but the team was pretty strong and so from just builders, product, technology, able to build, able to move quickly on stuff and ship things. The team was strong. That was the first kind of value proposition, right? The second thing was we had built a pretty sophisticated machine learning model for recommending content. So that was the second kind of thing that we could use to amp up how much we could sell the company for and then of course, we had a lot of users. But in a very specific distribution channel, and it would be like a call option. Maybe if the company that bought us could potentially tap into that distribution channel and so, we talked to a few companies. We talked to Coinbase, we talked to Uber, we talked to Robinhood. There were a couple other ones as well. We had been talking to this company, well they were called Pocket, but they had been acquired by Mozilla, so we were talking to them too. And that was more of the partnership kind of angle, strategic partnership angle, but we ended up getting an offer from Coinbase. We got an offer from Uber, we did not get an offer from Robinhood and what was super interesting about these offers is the Coinbase one was really interesting in particular. Because they were like, well, we can see that we might use the technology. We might want to surface crypto news, but it's kind of a stretch, and we're not really sure if we can value it, and we're not really confident how much of a premium we should pay just for this news recommendation stuff that we're not really even sure is a core part of what we do in our product. It's kind of a peripheral feature. They weren't as interested in the technology, and even Robinhood was similar when we were trying to talk about news in the market, like we can recommend financially the news and stocks and all this stuff anyway. But Coinbase was like, they interviewed the whole team, and then they were like, we actually only want five of you.

Pablo Srugo (00:13:51) :
Oh yeah that's the deadly part of the Aquire.

Russ d’Sa (00:13:53) :
When I was at Twitter, actually, I used to be on the other side of the table. I was doing diligence on acquisitions we would do on the engineers on a team and this came up all the time. It's like, no, we want this person, but not this person. It's kind of a savage thing, to be honest and, when I was on the side of doing the acquiring or being part of the diligence team. I didn't feel how hard it is because I'm the one.

Pablo Srugo (00:14:17) :
Because from your perspective, you're recruiting. You're like, well, why? I mean, I take what I want, whatever. But yeah, on the other side, you got to tell half your team you're not good enough. That's tough.

Russ d’Sa (00:14:25) :
On the other side, yeah, it definitely is really tough and so for that reason, that they didn't want to buy all of us and bring the whole team over. We actually said no to that acquisition and in the end, if you look at Coinbase's stock performance, and you look at the offers that my co-founder, me, and three other people at the company got. It would have been way more of a lucrative decision, like for, you know, eight figures or a higher eight figures decision once you see it all play out. But we didn't go that route because I think ethically it just didn't feel right to us. These are people we've been working with at the company, the full team of eleven or whatever that were there. We'd been working with them for like six and a half years, with a lot of them and it just felt wrong to kind of optimize for that outcome. We really wanted to optimize for doing right by the team and I credit my co-founder, by the way, for this, really holding the line there. Because I was definitely tempted. I was like, oh man, but this Coinbase thing is really cracking, you know, maybe we can help find them roles other places. He's like, no, we're not doing it and he really held the line. And so I got to give him full credit there for that. And we're still working together. He's my co-founder now.

Pablo Srugo (00:15:39) :
That's awesome.

Russ d’Sa (00:15:40) :
And so along the way, we were going through these acqui-hire offers. We had these offers on the table, and we were trying to figure out what we wanted to do. Uber wanted to take the whole team, so that was better and I had been talking to, we had been trying to talk to Robinhood. At Robinhood, was the VP of product at the time was Josh Ellman. He was an investor at Greylock, and now he's at Apple, running Apple Intelligence. But he and I had worked together at Twitter. And, so I knew him from there, and he was kind of our in at Robinhood to talk about the acqui-hire. He was also on the board of Medium, and Ev Williams is the founder of Twitter. Who I knew from Twitter days, but also the founder of Medium and so, Josh is in the room with Ev. And he's like, yo, dude, you know what Russ is up to now? Actually, he's working on this machine learning based news recommendation thing. What came up in that board meeting that day was, we should get more potentially into the news space and work with publications. And we should have more, right now, all of our recommendations of content on Medium are purely human curated. We don't do any AI type of stuff and so, he was like, yeah, you should talk to Russ, man. He's working on all this stuff and so, Ev and I met. We caught up after probably about a decade, or maybe like eight years, since we had talked last, and just caught up, hung out, had a good time. One thing led to another, and it led to many more conversations across that time, but he ended up being interested in, of course, a team. Because we delivered the product there and leveraging us to do that for Medium. He could value the technology because this is aligned with their product roadmap and then the distribution channel. What was interesting about that is he could value that too and do a proper acquisition. Because Twitter and Facebook had this stranglehold on distribution. He was thinking about how do I, Medium is popular in its own right, but he was looking for alternative ways to amp up distribution. Because of these hub destinations for content that have manifested over the last twenty years and so he was able to do a full acquisition. Another backroom story that was surprising for me during that process that might be helpful for listeners is that, when we started to look for acquisition, and acqui-hire was kind of the default path for us. Our investors were like, oh, yeah, go do whatever you want. They kind of wrote it off. They were like, you know, it's already zero for us, so do whatever makes sense for you guys and I was like, all right, cool. I appreciate that about you investors, you know, board and then as soon as it ended up becoming a real acquisition, and they could actually get their money back. Suddenly it was like blood in the water, and they were weighing in and trying to optimize the deal terms for what gets extracted on the front end versus any kind of retention bonuses for the team, and all the people coming over. My co-founder and I kind of said, OK, here's how we're going to play this. It's like the right thing to do. First of all, they wrote us off, right? They did give us their money, so that's not lost on us. They took a bet on us and we appreciate, and value that. So we feel a sense of duty to try to get them back as much as possible. But the thing that Ev is acquiring, his acquiring is the team, he's acquiring the technology, and it's the team that built the technology that maintains, and improves the technology. So there's so much value locked in the team. The investors were trying to rip out some of that value and move the pendulum farther over to them getting a return for something they wrote off already, and it just felt wrong. The way we played this was good cop, bad cop. DZ, my co-founder, would talk to the investors and negotiate with them, and then say, you know what, Russ is just not willing to do this. You know, this term, Russ is not willing to. I can't get him to. They're like, can we talk to Russ? He's like, no, he doesn't want to talk to you guys. They kind of, you know, he torched me a little bit, but we planned that that is how it was going to work. I'm like, all right, I'll just be the bad guy here.

Pablo Srugo (00:19:47) :
You need that. I mean, the only other bad guy that sometimes solo founders have to use is just the acquirer, right? It's like, listen, the acquirer doesn't give a shit about you guys. They only want to buy us.

Russ d’Sa (00:19:54) :
Yeah.

Pablo Srugo (00:19:55) :
So transferring economics from us to you is dead weight loss to them, right? So it's like. 

Russ d’Sa (00:20:01) :
Just to be clear, it wasn't money for me or DZ. It was money for the team and I'm just like, look, this acquisition would literally not happen if the team doesn't come along, and is happy, and delivers results. All of this stuff is codified in retention as well. They have to deliver results. There are metrics we have to hit. So just kind of keep that in mind and so we just, this is how we played it. The acquisition closed, and it ended up being a full acquisition. I'm not gonna say it was a massive acquisition or anything like that, but I think it was definitely a better outcome than any of us hoped for and yeah. It was an interesting learning experience on selling a company.

Pablo Srugo (00:20:39) :
Yeah, that's a huge learning and you know, what comes out of that to me is, especially for the acqui-hires, just the importance of relationships. It's so cliché, but at the end of the day, you look at all your meetings. You knew someone somewhere that then started that thing up and if you don't have that. You have to find a way to build that. Otherwise, this option is frankly not available to you. It's not like you can email Corp Dev at Uber and be like, hey, we're ten engineers, do you guys want to buy us? It's not going to happen.

Russ d’Sa (00:21:09) :
In all situations where a company is sold, not bought. It's almost purely relationship driven, in my opinion. You got to have a connection to someone, and someone's going to kind of help do you a favor. Are you calling a favor? I remember this story that Ron Conway told me about. I won't name who the companies were, but Ron Conway told me that there was at the time a fairly well known company that was on the ropes and almost out of money. And, he literally called another company and said, hey, remember that thing I did for you? You got to return the favor. You got to buy these guys and they went and they did it. They bought the company for Ron Conway.

Pablo Srugo (00:21:51) :
Crazy.

Russ d’Sa (00:21:51) :
Because he made the, strong armed them into doing it. It's like, it is pretty wild, how those connections matter in the kind of the downside case. You know, the upside case when a company is bought, like then it's like kind of all gravy.

Pablo Srugo (00:22:08) :
So post acquisition you stayed at Medium for like about two years and then you start LiveKit in early 2021. 

Russ d’Sa (00:22:12) :
Yeah. 

Pablo Srugo (00:22:13) :
What was the original idea for LiveKit and why did you go out, and want to do another one?

Russ d’Sa (00:22:19) :
Yeah, so the LiveKit started. It's the first thing I've ever done that sort of happened organically without intentionally trying to start a company. First time ever, the previous four, I've always been trying to start a company. Trying to be like my dad, and they never worked out. But LiveKit in particular, so we had sold to Medium, and it was six months before the pandemic. I think our acquisition was July 4, 2019, and celebrate America, celebrate, you know, selling the company. And then pandemic hits, suddenly everyone's at home and can only connect with each other over the internet. And you're using Zoom, Discord, and Google Meet. The reason you're using those apps is because they were built for video and audio streaming at scale. That's the reason you use those apps. In particular, the rest of the internet, though, wasn't actually built for that purpose. Whenever you type in HTTP into your browser address bar, that stands for hypertext transfer protocol. It's a protocol for transferring text over a network. But pandemic hits, you can't go anywhere. You need to use your webcam and your microphone to talk to other people. Well, that's voice and that's video. And you actually need a different protocol to do that. One that had been operating in obscurity until the pandemic hit. That protocol is called WebRTC and if you were to guess, what is the protocol that Google Meet and Zoom and Discord use? Well, it's WebRTC under the hood. That protocol is quite difficult to scale. You need a lot of infrastructure. You have to build around it to scale and so Zoom had built all that infrastructure over a ten year period, and Discord had been working on their stack for ten years, and Google Meet similar. If you were a developer trying to integrate real time audio and video streaming into your app, I was working on an app at that time as a side project while I was at Medium. I was working on a Clubhouse for companies to use during the pandemic, drop-in audio conversations for companies. I needed audio and video streaming. So I started to look around for infrastructure to help me do it and it turns out that all this stuff that Zoom does under the hood is locked within the Zoom app at the application layer. There's really nothing at the API layer that made it possible or easy for a developer to do and so I started to work on my own stack. And DZ, my old co-founder from Evie, better engineer than I am, I said, hey, I need help on this stack. Can you help me work on it? So we started to work on it together. My Clubhouse for Companies idea ended up not having legs. I tried it with a few design partners. They weren't really using it that much and found out there were different jobs to be done than what I had originally imagined. I thought I could just pattern match from Clubhouse to an enterprise version of Clubhouse and it would just work. And it turns out it didn't work but some friends said to me, hey, the stack that you had built to power your Clubhouse for Companies, can I use it too? So we made a GitHub repo kind of casually. We started a Slack instance, invited them in so they could give us some feedback on it, how to make it better. If they had any bugs that they ran into, we could fix it for them and then slowly, over like a month, two, three months. All these people I'd never met before were getting invited to the Slack channel, and I'm like, okay, it's kind of weird. And they're starting to say, hey, do you have a React Native SDK? Hey, do you have a, you know, can you fix this bug? We're like, oh, okay, it's kind of odd. This is starting to take on a life of its own. So we said, well, you know what, we don't have an idea yet for another company. We're at Medium, kind of like working here, trying to make things work, and treat Ev well for acquiring the company and ship some products. So in our nights and weekends, we would just work on this audio-video infrastructure stack.

Pablo Srugo (00:25:54) :
And this was what? Mid 2021 at this point?

Russ d’Sa (00:25:57) :
This was actually right at the end. This part where it's starting to snowball a little bit is at the end of 2020, and so we still haven't left Medium yet. Then three to four months later, it feels like it hit this inflection point. Where every day there were new people coming into this Slack channel, and we're like, all right, well, this is kind of becoming a real open source project. Developers are struggling with how to build this stuff for themselves. All this infrastructure is a pretty hard problem. So why don't we just leave and focus on working on this open source thing until the next startup idea comes to us? We figured it would be a crypto wallet or something. We were feeling maybe bitter about passing on Coinbase when the IPO happened. But yeah, we ended up leaving Medium and starting to work on LiveKit proper as an open source project we launched summer 2021, and it just exploded. The day we put out the first release in open source, it wasn't a company. It was just me, DZ and our, kind of, who is now our first employee at the time. Just a contributor that we knew from Evie, actually one of our old Evie employees. It was just three of us and this open source project, and then a public company comes in and tries to acquire us immediately. It wasn't even a company. I'm like, what are you guys acquiring? You buying the repo or? And then, so in the first two weeks it was trending across all of GitHub, top ten repo across every language. Spotify, Oracle, Reddit, and X, these large companies were starting to deploy it and start the repo, and we're like, what's going on here? So we started to talk to them. We had meetings with them and we're like, why are you using it? That was our primary question. There's commercial things out there like Twilio and stuff, super expensive stuff. But why LiveKit? You know what they said was, hey, we can see your code, and you guys designed a really great system, the server and client and the interplay between the two. I mean, it's a really great API design. We can deploy it and control our own destiny to a degree.

Pablo Srugo (00:28:01) :
We have tens of thousands of people who have followed the show. Are you one of those people? You want to be part of the group. You want to be part of those tens of thousands of followers. So hit the follow button. And what were some of the use cases that X or Spotify were using it for?

Russ d’Sa (00:28:15) :
So for Spotify, what they were exploring, and they still use this today. But what they were exploring building kind of like live concerts using LiveKit. So, being able to have an artist go up there and do kind of half talk show, half live concert, or play some pre-release stuff that they're working on for their closest fans. That was the Spotify use case. Oracle’s was crazy. That was like a Cybertruck police drone. They use us today too for a bunch of different initiatives, but that was the first one they had an idea for. I don't even know if the Cybertruck was out at that time, but they were trying to build these police drones using Tesla vehicles.

Pablo Srugo (00:28:54) :
Crazy.

Russ d’Sa (00:28:55) :
It's pretty wild. They do a lot of stuff with the government and military as well. X was looking at Twitter spaces, and then there was Reddit, who was trying to use us for Reddit Talk. Reddit uses us for some other stuff today, not Reddit Talk. Reddit Talk ended up getting shuttered and shut down post, kind of, coming out of the pandemic a couple of years later. But Reddit Talk was what they were exploring to use us for initially. So we were talking to them, and when I say these use cases like Reddit Talk, Spotify stuff, Twitter Spaces, these are all really massive scale use cases. The communities on these applications are huge and so what they said to us commensurately was like, hey, none of the providers, even on the commercial side, can do massive scale. None of them can do the lowest latency possible through an edge network. All of them are single server systems where they put the server somewhere in the world and everyone has to connect, regardless of how far away they are, to that server over the public internet. Which is slower than the private fiber connectivity between all of the different data center providers. Nobody has an edge network. Nobody can do massive scale and just horizontally scale up servers infinitely and nobody can deliver five nines of reliability. So if a server falls over, or if an entire region of a cloud provider goes down, or even if the cloud provider itself across every region goes down. How do you make sure that this is almost a public utility where you just can fail over and still provide high quality of service via another data center or another region, or another cloud provider. And so, they said, look, you have this single server system with SDKs and open source, but we don't really want to have to figure out how to solve the reliability, latency, and scale of a global network. If you guys figure out a way to take your server and turn it into a multi-server system where you can put them all around the world and they kind of intercommunicate and form this single gigantic mesh network of WebRTC servers. Almost one big distributed WebRTC server, we would happily pay you for that as a commercial product. So that's when we raised a round, our first seed round from Redpoint.

Pablo Srugo (00:31:05) :
How much did you raise in that round?

Russ d’Sa (00:31:07) :
We raised $7 million for that round.

Pablo Srugo (00:31:09) :
And this was still a lot of COVID era use cases?

Russ d’Sa (00:31:13) :
Yeah, COVID area use cases like, you know, another example here. Within the first three weeks, there was a company that started to go and deploy us in all of the dispatch centers around the U.S.,
for 911 emergency.

Pablo Srugo (00:31:23) :
Okay.

Russ d’Sa (00:31:24) :
So that company was called Prepared, they actually sold to Axon and now we operate the same thing but via Axon now. Where LiveKit is run in these dispatch centers and so when you call 911 for about thirty five or so percent of the calls. That's actually going through LiveKit infrastructure and the dispatch agent can talk to you. And this is a human here, when I say agent. Human dispatch agent can talk to you. You can stream video to it. You stream your audio, you stream GPS data, all this situational data that helps the agent understand better about what the situation is that's going on and how to best help you. Now what's happening is they're actually integrating a LiveKit agent into that workflow where the agent can do things like figure out, oh, okay, fire needs to, firefighters need to get looped in. And they can dispatch out to firefighters, and bring them in automatically. Does stuff like gun detection and license plate reading, and information gets forwarded to FBI field offices. And so there's all kinds of cool stuff that that agent is doing together as a copilot with the human dispatch agent in this 911 call. Anyway, but that whole effort started very quickly after we launched the first open source release. So the use cases here were the 911 is kind of like public, you know, public service or law enforcement, kind of government related use case. But yeah, it was like live streaming and video conferencing was really where the concentration was around what we were providing with our initial kind of release of LiveKit. I should say video conferencing for the open source release. Because video conferences didn't need to have like ten thousand, one hundred thousand people in them. But then when we launched LiveKit Cloud, we raised that round and we started to work on this thing that these large companies asked us for, this commercial global network for streaming media. That's when we launched that thing at the end of 2022, is when live streaming suddenly using WebRTC became a thing, became viable. All the stuff that powers Twitch and YouTube Live and things prior to something like LiveKit Cloud existing is using a different protocol, a different technology. A legacy technology that actually isn't even truly live streamed.

Pablo Srugo (00:33:35) :
Are those big companies like Twitch, YouTube, are they relying on your product, or do they have their own stacks? So I guess, because my big question is, a lot of these use cases. Like, they kind of, who I would think would use this would be the new apps that are now powering new use cases around live streaming and whatever, and they don't want to build their own from scratch. Versus a really big, like Twitch and YouTube, that have been there for a long time. But a lot of that wave, like 2020, 2021, early 2022, was really hot and then by the time that you've got this thing. You know, the COVID era stuff, falls, remote is like falling. Is that affecting you guys? How is that playing out inside LiveKit?

Russ d’Sa (00:34:09) :
We hadn't felt the effects, like the negative effects of it. Because I think two things were happening. One, well, I'll explain kind of why we hadn't really felt the negative effects. Because something shifted fundamentally at the end of 2022, and really the way when it shifted was at mid 2023. Which I'll get to, like the entire company charter changed. But it changed in such a way that it layered in this new phase of growth as maybe some of the COVID era stuff was tapering off. So we didn't feel the taper off because we were kind of all in on this new kind of wave of growth and a new use case. I'm not trying to be purposely vague. We'll talk about it in a sec, but, you're right that, net new live streaming kind of companies in certain niches were starting to use us. So, Whatnot for live shopping is an example of this and there's still a top twenty app in the App Store and Pump.fun. We power Pump.fun, which is, that's a whole, another candle. But yeah, that's kind of a net new live streaming use case, right? That is kind of cropped up and really hit scale. And so end of 2022, we launch LiveKit Cloud. We're powering now all these live streaming kind of use cases in these niche categories and video conferencing use cases. And then ChatGPT comes out and GPT-2 sucked so badly.

Pablo Srugo (00:35:28) :
Yes.

Russ d’Sa (00:35:29) :
And then all of a sudden, you use this app that wraps GPT-3.5 and you're like, holy crap. This is a crazy jump, right? And I'm like, wow, I'm texting with a human, with my friend but it's a computer. What if I took LiveKit stuff, WebRTC technology, and I pair it with ChatGPT. What if I could build a demo? It would be pretty cool to build a demo of a computer that felt like a human, but instead of you texting with it, you're talking to it.

Pablo Srugo (00:35:56) :
I see.

Russ d’Sa (00:35:56) :
So I built this demo pairing our technology and GPT together, tweeted it out. And I'm like, I'm going viral for sure. My first time ever. I'm so excited for it and then barely anybody noticed.

Pablo Srugo (00:36:10) :
No way.

Russ d’Sa (00:36:12) :
First demo of this too, where you were like Samantha from her.

Pablo Srugo (00:36:16) :
Like the first voice AI demo, before the voice mode.

Russ d’Sa (00:36:19) :
Voice AI demo, latency was like super, not super low.

Pablo Srugo (00:36:24) :
But for then, yeah. It's all about expectations now, the expectations are so high. But back then, you know, just about anything would have been impressive.

Russ d’Sa (00:36:31) :
Yeah, yeah, like, you know, Alexa, Siri, would take ten seconds to respond to you and it took like three seconds. And it felt like a way more conversational, and it was definitely state of the art for that time. But yeah, nobody really paid attention to it. It did, I think, it got like ninety likes, far from viral. But then what happened was, I said to my co-founder after I totally died. I'm like, man, maybe I should just try to figure out, through the YC network. Maybe I should try to figure out how to get to Sam and just be like, yo, let's build this in ChatGPT. But then I forgot about it, and I'm still trying to run this live streaming video conferencing business.

Pablo Srugo (00:37:03) :
How big was that business at that point? Were you doing millions already? ARR?

Russ d’Sa (00:37:06) :
Not that much. We had done about I think at that time August 2023, is when we hit a million in run rate. 

Pablo Srugo (00:37:14) :
OK.

Russ d’Sa (00:37:15) :
And so I would say at like, we were maybe sitting around $300k or something like that.

Pablo Srugo (00:37:20) :
Okay you had something but it wasn't like something to protect. Where it was like you got this big business you got to worry about.

Russ d’Sa (00:37:24) :
No definitely not and then what happened was on that day I think it was like August 4th 2023, I got pinged by the, well I won't say it. I won't say who it is but I got pinged by a hyperscaler and they were like, well we want to talk about collaboration, you know, that kind of.

Pablo Srugo (00:37:41) :
Yes.

Russ d’Sa (00:37:42) :
Vague thing. So I went to this five hour long lunch with them and they were like, can we buy you? Can we license? Or we're going to kill you. They offered us a super lowball offer and we're like, no. And they're like, well, if you have an AI idea. Somehow tie it to AI, then we'll 10x the offer. I'm like, I don't have a way out. We're a live streaming company, video conferencing. I can't.

Pablo Srugo (00:38:02) :
Oh, God.

Russ d’Sa (00:38:03) :
Sorry, guys.

Pablo Srugo (00:38:04) :
Oh, so that hyperscaler came to you for the live streaming stuff. Asking if you have it, OK.

Russ d’Sa (00:38:09) :
Yeah.

Pablo Srugo (00:38:09) :
Wild, all right.

Russ d’Sa (00:38:10) :
Because they want to be cool, too and then I'm driving home after that long lunch. Thinking the hyperscaler is going to kill us and then I get an email from OpenAI. And it's like, hey, so we signed up for LiveKit Cloud in secret with a personal Gmail address three weeks ago, and we built this voice interface.

Pablo Srugo (00:38:29) :
Holy, shit.

Russ d’Sa (00:38:31) :
Like on top of you guys, and now we're like fans, and we want to talk commercial. And so that's when we started to talk to them. That was kind of the start of all of this thing where Ben Thompson calls OpenAI the accidental consumer company.

Pablo Srugo (00:38:46) :
Yes.

Russ d’Sa (00:38:47) :
Because they never meant to create ChatGPT as this massive model.

Pablo Srugo (00:06:48) :
Correct.

Russ d’Sa (00:38:50) :
And OpenAI goes and turns LiveKit into the Accidental AI company because, you know, a lot of people at that time were starting to do this thing. Where it's like they figure out what their AI angle is and then they pitch it to VCs and like.

Pablo Srugo (00:39:01) :
Yeah, it was like nine months after the ChatGPT moment. That's right, yes.

Russ d’Sa (00:39:04) :
We were not trying to do that. We were like, this is a live streaming video conferencing company.

Pablo Srugo (00:39:09) :
And we got our lane, which is actually funny. Given that you guys weren't at many millions in ARR that you were thinking that way versus like trying to.

Russ d’Sa (00:39:16) :
Yeah.

Pablo Srugo (00:39:16) :
Throw AI on top of whatever.

Russ d’Sa (00:39:18) :
Totally and then when the OpenAI thing happened, right? Suddenly it was like, oh, wait, this company could be much, not bigger, not in terms of dollar value. Just like the impact that we could have on the world could be much bigger. Like, we should change lanes, you know, kind of what we're going after. We could build probably a healthy business on it, but here we have an opportunity to build kind of a generational infrastructure company. At the time when OpenAI first started to work with us, we were network infrastructure, and it was like, okay, OpenAI is building the brain, they're building a human like computer, that's AGI, right? That brain needs to be able to get signals that go into the brain and then signals that come out of the brain, right? Like, when the computer speaks, you got to be able to carry that signal back to wherever it needs to go. When the AI needs to see and hear, you've got to carry that signal from the source to the model and so we can build the nervous system for AI here, right? This network infrastructure, the backbone for multimodal AI, and then there was this other realization that happened. So we were super excited about the backbone part. That was awesome and we were like, all right, we're going to pivot the whole company to this and all this stuff. And the unique thing here was that WebRTC, that protocol for streaming audio and video, it was designed for client devices. It was not designed to run on a server program. So what that means is it's designed for you and me to connect to one another, peer to peer and stream audio, and video directly to one another from our devices. Which have cameras and microphones built into them. It's what we're doing now. You know, there is a server in the middle, but fundamentally what we're doing is client to client. But what OpenAI was doing with ChatGPT voice mode is they had a server program. Think of it like, you know, you have your Next.js backend. It's a new type of backend, an agent, a voice agent that is consuming audio from the user streaming in from their phone, acting like it's another phone. It's streaming into that computer program. The computer program is then going and orchestrating these models, converting it into text, the speech, and then running it through an LLM. And then taking the tokens, and converting that back into speech. Those text tokens, and then streaming it back out to the user. And so, there's some orchestration you got to do there. There's interruption handling and understanding when the user is done speaking, so the AI can start to process what the user said. There's orchestration, there's the core WebRTC streaming over the network you got to handle. Anyway, so we built a bunch of stuff for OpenAI to be able to do this and then we said, you know what, people are probably going to want to do the same thing. Lots of developers are going to want to build all these voice AI, it wasn't called that at the time, applications.

Pablo Srugo (00:41:52) :
Insane timing. It's crazy timing, yeah.

Russ d’Sa (00:41:54) :
Super lucky and I feel it's like unheard of luck, you know?

Pablo Srugo (00:41:59) :
Well, because you can be too early, too late, you know, whatever and this is just as that wave is about to come. Because the first company is like, OpenAI is going to do it and then. You know, others are definitely about to follow, like in spades.

Russ d’Sa (00:42:12) :
You know, I think to win the game, you got to be in the game and I think I was just hanging around the rim for a long time. And so it finally happened, like the stars aligned. But anyway, we turned it into a framework, kind of like our version of Next.js. We put it out there, and we're like, all right, developers, voice mode launched. Here's how you can build the same thing. You take our framework, you build the thing, and then you can run the thing on our network. But what was really interesting is that right when voice mode launched. Right before it launched, we hadn't put out the framework yet, but we were about to put it out. OpenAI put this meeting on our calendar, and we go into this meeting, and they're like, so how do we scale this thing? How do we deploy and scale these voice agents? I'm like, no joke, my answer was, you're OpenAI. Why are you asking me? And you guys are the best. And they're like, well, we just want to kind of understand best practices because it's a bit different than what we are used to doing. ChatGPT is a web app, fundamentally, or it's built on top of HTTP, and this is built on top of WebRTC. It's a different protocol and one thing to know about HTTP is that HTTP is a stateless protocol. What that means is that you don't have any context building up across requests. Every request is independent, but with a voice conversation. We're having a conversation that has been building up over an hour and so, every time I say something and then you process what I said and then say something back. We're not ending the call and then going back into a call when I say the next thing. It's like this is a stateful interaction where you're building up context over time, and you're connected for an hour or thirty minutes.

Pablo Srugo (00:43:47) :
And in ChatGPT they're faking that by just. Every time you send a message they're resending the whole message, the whole chat.

Russ d’Sa (00:43:52) :
Exactly, they're faking it. It doesn't matter per text because you're used to a text conversation. Text conversations are not real time.

Pablo Srugo (00:44:01) :
You got the time to read it and then time to write it. And yeah, exactly.

Russ d’Sa (00:44:04) :
And it's like you can wait five seconds before the AI responds. That's kind of baked into the user experience but for a voice conversation. If you're trying to build an AI that talks to you just like a human. Imagine if I say something and then it takes ten seconds for you to respond, or you say something back and then I say something else, and it takes like ten seconds for you to hear it. That just breaks your expectation of how the experience is supposed to be. But here's what's crazy, is this thing that we did with OpenAI for ChatGPT. It's like this iceberg where stateless to stateful, suddenly all the infrastructure to support an application that looks like that is completely different and it's not just the framework. It's not just the protocol for transferring over the network. We're talking about how you deploy an application like that. Because instead of having a bunch of web servers and you just go request, request, request, request round robin across these web servers. Because every request roughly takes the same amount of time, so you can just round robin across them for load balancing. Suddenly a request comes in that's a session that might take thirty seconds, and then another request comes in and that's a session that might take an hour. The way I think about it is like multiplayer video games. When you play Fortnite, one Fortnite game might take five minutes and you'll win. And then somebody else is in a Fortnite session that might take thirty minutes for somebody to win. And so, load balancing a multiplayer video game is very different from load balancing a web application. And so we told OpenAI, here's how we have done it in the past and how we recommend doing it. Then we started to work on a deployment stack. So if you think of the agents framework as our Next.js, now we're working on the Vercel piece. You can build your agent, you can deploy your agent to LiveKit Cloud. We will spin it up all around the world. We will manage where that agent is and where the user is, and how they're connected, and all the latency between them. And then when they're done talking, spin them back down and put them back in the pool to wait for the next session. or if that agent gets disconnected somehow. Transferring all the context and everything, resuming somewhere else with failover. So we manage the whole process for you.

Pablo Srugo (00:46:08) :
And do you compete with like a Vapi, for example? Or do they sit on top of you?

Russ d’Sa (00:46:12) :
So Vapi and Bland and Retail and some of these. Most of them actually are built on top of LiveKit and so they use our infrastructure for all of the transport. Then they have their own infrastructure for orchestration, and they do their own deployment thing, and all of that. But I think the key differentiator there. They are using our underlying transport, but the key differentiation is that they kind of come in at a layer above, at the abstraction level. So their lower code and where they have an advantage, or where they deliver a better experience to some cohort of people, is that the time to market is very quick. You click some buttons, put in a system prompt, click go, and you're off to the races. LiveKit is more of a high code solution where you need finer grain control over how this agent is going to act. I think of it personally as Webflow versus Next.js and Vercel. Solid businesses, and they address a different segment of customer. But my personal opinion is that eventually you need to go to code. Eventually, if you have ten people that all come into a low code solution and, through clicking some buttons and getting a system prompt together, have a viable product. Eventually those folks are all going to compete with each other. The barrier to entry for competitors is lower. You have to differentiate, you have to have more control to differentiate, right? And so more control than a UI can give you and so that's kind of where I draw the distinction between some of the low code platforms versus what LiveKit offers.

Pablo Srugo (00:47:43) :
And then, yeah, with ChatGPT. Did you end up getting a deal done and you ended up powering their voice mode or how did that kind of all end up?

Russ d’Sa (00:47:49) :
Yeah, yeah, we've been powering it for like three years.

Pablo Srugo (00:47:51) :
Wild.

Russ d’Sa (00:47:52) :
Yeah, the scale has been insane there.

Pablo Srugo (00:47:55) :
It's got to be crazy.

Russ d’Sa (00:47:58) :
And I have stories about that too. Them flipping the switch on stuff, and full credit to the team. Our stuff just scaled. It scaled to the moon. We didn't have downtime, no outages, we didn't struggle at all.

Pablo Srugo (00:48:09) :
Question for you, when you have a customer like OpenAI, is everything else irrelevant? Do you know what I mean? It's such a big behemoth. Do you have other customers that sum up to some percentage of that together, or is it just like.

Russ d’Sa (00:48:22) :
Yeah, well, I would say a couple of things. One, in terms of customer satisfaction or happiness. They kind of overshadow a lot of the customers in that you care a lot about them having a good experience, for sure. Because they're a really strong engineering team and they're operating at a really large scale. Them being happy with the product you provide. That satisfaction makes its way out to other customers who are interested in your stack and are like, oh, well, if you're good enough for OpenAI, then you're good enough for me. And that actually has been, with large enterprises, we've had conversations where the first thing is like, okay, so who do you guys power? This was in the early days. I think a lot of people know we power this now, but in the early days. Salesforce came to me and they were like, so first question, who do you guys work with? Who do you guys power? Can you handle our scale? I'm like, oh, well, we power ChatGPT. We're like, all right, next question.

Pablo Srugo (00:49:15) :
Done, yeah and that probably takes care of a lot of questions. Like, okay.

Russ d’Sa (00:49:18) :
To that degree, OpenAI definitely matters a lot, but I think the product that OpenAI is building with voice mode is a consumer product and it's also a peripheral feature of ChatGPT. ChatGPT opens up to text. It's, like, only a segment of, and it's not a small segment, but only a segment of ChatGPT users use voice mode on a regular basis. And so, but if you think about this, like in aggregate. Where today, here and now, is the large majority of voice traffic? Well, it's in Telephony, right? There are a trillion phone calls happening around the world every single day and billions of those calls have some kind of B2B workflow or process that the caller, and the person answering the phone are trying to navigate together. It's a human these days, but the human is the person answering the phone and so what we've seen is this flood of people that are trying to take AI, build these agents that actually answer phones instead of humans answering them. Maybe for a front line, maybe to facilitate the entire workflow within a business process. But if you take all of those different pockets of use cases together, it's several, several times bigger than ChatGPT voice mode. So yeah, OpenAI is a big customer and has a lot of traffic. But at the same time, the pie is a lot bigger than that in terms of all the voice workload that is available or has the opportunity to put AI into the middle of that.

Pablo Srugo (00:50:58) :
Perfect. Well, listen, let me stop there. I'll ask a few kind of quick fire questions. When did you say you hit a million ARR? When was that?

Russ d’Sa (00:51:04) :
That was August of 2023, is when we crossed our first million. Yeah.

Pablo Srugo (00:51:09) :
And how quickly did growth go after? You don't have to tell me where you're at today, but just like, you know, the year after that. What did that look like?

Russ d’Sa (00:51:14) :
Yeah, we started at zero and then we 3x'ed and then 5x'ed and then yeah. 3x'ed again and yeah, so it's been good.

Pablo Srugo (00:51:24) :
Epic. When was the moment when you felt like you'd found true product market fit?

Russ d’Sa (00:51:28) :
Oh my gosh, this is like a crazy question because up until very recently, I had not felt like we had product market fit, and everyone's yelling at me, like, dude, what's wrong with you? This is the thing, I've had literally this conversation with people in the past. Which is I should have realized that when OpenAI started to use us, we had product market fit, at least for this AI agents, voice AI use case. But I think, as a founder and as an engineer. And also as someone who's come from the consumer product world in my past. That's where I spent most of my time, in my twenties. You're also taught at Y Combinator that, you know, to a degree. That product market fit is this elusive thing, and it's very hard to know when you have it and when you don't. And sometimes there are red herrings that can kind of give you the indication that you have it when you really don't. And so I kind of always assumed, I took more of a pessimistic view in that we don't have product market fit yet, we don't have product market fit yet. And maybe I sort of drew my comparison points incorrectly, where I'm like looking at something like Cursor.

Pablo Srugo (00:52:31) :
That's PMF, yeah. Tough bar.

Russ d’Sa (00:52:35) :
Product market fit is a spectrum. It's not a binary thing and we do actually have very strong product market fit. I would say that the moment we had it was with OpenAI, becoming our design partner for VoiceAI. But the moment I realized it wasn't until maybe four months ago, as crazy as that sounds.

Pablo Srugo (00:52:50) :
And then last question, what would be your number one piece of advice for an early stage founder that's still looking for product market fit?

Russ d’Sa (00:52:56) :
I would say that it kind of ties into something I said earlier. You know, you got to be in it to win it or you got to be in the game to win the game. You have to be around the rim and the extra step there is which rim you choose. If rims are markets, I think choosing the right market is super important, right? I think if you haven't found product market fit, you've built a product and there's a market you're in, right? And if you're still searching for it, well, the first step, in my opinion, is look at the market that you're in and try to assess the potential of that market. Let me just use voice AI as an example. Voice interfaces to computers, not that I'm trying to invite more competition but in earnest, trying to do right by founders, is like Voice AI. Voice interfaces to computers is very nascent. We're still in this text mode. Most of the internet is still built around text and that is going to shift in a huge way over the next five to ten years as the models become smarter and more human-like. You interact with them in more human like ways. Like, voice, right? Is the primary way we interact with each other. So it's going to be that way with a computer in five to ten years. But, you know, there's a lot of opportunity. It's a big market, and it's nascent. And so put yourself in a space where the market is really big, where there's a ton of opportunity, and then be nimble with the product. Be willing to try something, get rid of it if it's not the right thing. Try something else that is kind of an educated guess as to what's going to resonate, and then get rid of it if it's not the right thing. And sort of iteratively move your way through the market as the maze until you find the thing that really hits. That would be my advice for trying to find product market fit, is picking the right market is actually much more important than building the right product in that market on your first try. You can iteratively build the right product, but it’s very hard to iterate your way to a market. You have to kind of pick that right.

Pablo Srugo (00:54:48) :
Perfect. Well, Russ, thanks so much for sharing your story, man. It's been awesome.

Russ d’Sa (00:54:51) :
Thanks, Pablo, I really appreciate it. Thanks for having me on.

Pablo Srugo (00:54:54) :
Wow, what an episode. You're probably in awe. You're in absolute shock. You're like, that helped me so much. So guess what? Now it's your turn to help someone else. Share the episode in the WhatsApp group you have with founders. Share it on that Slack channel. Send it to your founder friends and help them out. Trust me, they will love you for it.