I don't really necessarily buy learning from failure. I think you learn way more from success.
We sat down for a couple years and just had a lot of conviction. What we're building will solve a bunch of problems. Which problems? Like we don't know.
Erik Bernhardsson was Spotify's 40th employee. He built Spotify's recommendation engine as a side project and grew a 400-person engineering team at one of the fastest-scaling fintechs ever. In 2021, he started Modal with just five people and spent two years building deep AI infrastructure. Then Stable Diffusion dropped and everything changed. Revenue went from hundreds to hundreds of thousands almost overnight. And today, Modal is valued at over two and a half billion dollars and is used by companies like Mistral, Ramp, and Scale AI.
Humans are the root cause of everything. Everything bad or good that happens at a company is like downstream from hiring good people that can make those decisions and do the right things.
People have all these like random ideas of what they want, but often that's not exactly what they need. Don't build what they're asking for. The whole point of a startup is like come up with a solution that people didn't really think of and then put it in front of customers.
In this episode, you can expect to learn about how to build a deep AI-native company, how to grow a company from 10 to 10,000 people without it destroying itself, and how to decide what problem you should solve, and so much more.
I love like taking a problem that people think is really hard. I kind of scratched the surface and I realized like, yeah, it's kind of hard, but not quite as hard as people think. To me, that's really exciting.
What's a moat of a startup? There's no moat. It's just...
I'm super curious. You studied physics, right? That doesn't seem like the natural starting point for the company you're building now. Or can you walk me through why you decided to do that?
Honestly, I grew up coding. I was coding my whole life since I was eight or something like that. And I did a lot of programming competitions. So I kind of felt like, and this is maybe me being like arrogant or something, but like when I started university, I was like, I kind of already know how to code. So I felt like maybe I just start something else. And I mean, I was also into math and physics too. I did a lot of math and physics competitions too, but I was better at coding. And yeah, I actually don't regret it at all that I studied physics. It gives you a very wide, big sort of toolbox that you know a lot of different tools that helps you understand the world and think in, you know, gives you a lot of mental models that are very valuable for a lot of different things.
And would you do the same choice now with AI and software engineering?
I mean, probably. I think even more so, right? Because I feel like computer science is even less, in a way, important, right? Like a lot of what you study with computer science is like build a compiler, build a file system. Like I don't know if you... I think it's helpful to understand some of those stuff. Like I didn't take it. You definitely don't need to study things like object-oriented programming. But like all the stuff you learn in physics, like control theory, you know, feedback loops, like you learn like statistics, like stuff like that. I mean, especially in Sweden, like physics is very math-heavy. Probably all over the world, actually. But I think all of those things you learn are quite valuable in themselves.
How does that translate to the... because you're in a really interesting position. You've got a pretty technically difficult product. But you're also, as the founder, in a position where you have to manage people, you have to hire people, you have to manage internal stakeholders.
Yeah, those are systems, you know.
Okay. Interesting. What sort of frameworks do you use then? Like frequently.
I do feel like there's sometimes a little bit like over, you know, mathematics toolbox and they're trying to apply it to everything. Like obviously, like you know, something as complex as humans, like you can't reduce like math. But like I don't know, like I think a lot in terms of like prioritization, like how should we prioritize things. Like I have like kind of like frameworky type thing in my head. I think in terms of like how do you grow teams, like how do you organize stuff, like work in progress, like how do you minimize that? How do you get stuff done quickly? What are the feedback loops? Like how do you learn fast from customers? Like all these things. Like I wouldn't say it's like super mathematical, but like I still have like a sort of a quantitative approach to it, and I think it's actually quite helpful.
Right, but you're really looking at, again, what you said, the systems, inputs, outputs.
Yeah, exactly. And that's like some of the things you study as physics. Like control is like a good thing. Sometimes it's like a black box and you're just trying to understand like why does this end up having like weird oscillations and like you kind of figure out, okay, we need to dampen it. Like you know, there's all this interesting stuff in physics I think helps you a little bit with like these complex systems.
Right. Is it because it's mainly error detection then, creating environments, simulating, running experiments to figure out, you know, or just theorizing what actually happened given these changes?
Yeah, I think that's a good point. Like the sort of hypothesis, you know, formulate a hypothesis, test a hypothesis. I think another thing that's like kind of valuable in physics is also that like to some extent like you kind of realize like when you study physics like everything is just the model. Like there's no like reality in a way like doesn't even really exist.
Because like first you learn like Newtonian physics and then you're like, and then you study like relativity theory and you're like actually like all these things like they don't matter at like high speeds. And then you learn like quantum mechanics and you're like actually they don't matter at like low scale either. So like one thing I kind of appreciate with physics is like it teaches this sort of idea that like you can have like very valuable like abstractions and approximations, but like you know for different parts of the reality you still have to apply like other models to have more accurate predictions. Right. I don't know, I think it's like stuff like that.
Yeah, it depends. Like I'm guessing this is maybe a horrible comparison, like the Paul Graham founder mode essay that went really viral on Brian Chesky sort of zooming in really to the finite details of the business and being able to zoom out. I guess maybe with those two theories, you know, with your understanding of physics, maybe that's something that you're able to do really easily as well.
Well, here's a mathematical theory. How about that? It's like it's a lot about random sampling. Like how do you keep an organization of thousands of people accountable?
Like there is no way, like you can't do that, you know, with every single thing. But what you can do is like you can actually pick kind of like random things and look at them and see if they're good or not. And actually like from a game theoretical point of view, like I think that's as good, like it gets you pretty close to be able to like hold everything accountable to the same bar.
So what does that look like? Does that look like you going into a random team meeting every week or every other week or every month? You just... what does that look like systematically?
Yeah, sometimes I just like look at like random code reviews and just like take a look, you know, and like drop some feedback. And by the way, like I'm like a super curious person and often like the feedback I drop is like, this is great. So like I don't think it's like necessarily like there's like random sampling. It's not necessarily like I drop in and like, this sucks, make it better, change everything. Like no, it's more about like just like making people feel like they're seen and making people feel like they're actually recognized for the valuable work. So it cuts both ways. Like sometimes I drop in to say actually like I don't know about this, like maybe we should do it the other way. But I think it's also about like, you know, recognizing people's good work. So it goes both ways.
It's kind of like you're keeping a pulse on the company by checking in and doing this random sampling.
Yeah. And I'm kind of a naturally very curious person. So like I always go around like ask people a million questions. I think a flaw in this is like if you have like a biased random sampling, if people know you're always going to go and ask questions about certain projects. So you have to make sure as a founder it's like genuine, like true randomness, which I like... it's harder because like I sometimes bias like towards certain things that I know better.
I'm curious on that aspect. Like I, you know, I'm a big fan of the Founders podcast with David Senra, and I think it's so interesting to see like there's so many different ways to win, but there's only like a couple of ways for individual founders to win. I'm curious as you've gone, you know, with building the company and even before then, you know, being CTO and running different sized teams, have you noticed that you specifically need certain structures? Or because you're so curious, maybe you know you need to put systems in place to allow that to happen to succeed as a founder? It's a big, very wide-ranging question, but I'm kind of interested, like how do you match your own personality to the success of the company and the way you operate?
First of all, I don't think there's a way to win. I think there's like 2,000 ways to not.
Okay. Okay. Charlie Munger, show me where I'm going to die.
Yeah, exactly. Or like I was actually thinking about Dostoevsky or whatever it is. Or Tolstoy, sorry. It's like, you know, every happy family is like happy in the same way. Every unhappy family, they're all like unhappy in different ways. Maybe it's the other way around. To me, like a lot of what makes you successful is like avoiding like 5,000 mistakes. Or by the way, like not even just avoiding, it's like at least like, you know, when you make those mistakes, like recognizing them pretty quickly. So I'm always like nervous about like silver bullets, but like I don't know, like a lot of it comes down to like hire smart people, give them a lot of freedom, like you know, find things in the market that there's demand for. Listen to your customers, but don't listen too much because sometimes they give you like weird advice. Iterate very quickly, you know, get stuff in front of customers like very fast or get stuff out in production very fast, you know, build a culture where like people can like, you know, iterate and learn very quickly. There's a lot of those sort of basics I think is like quite important, and even there I think a lot of people screw it up.
Just because they're basic doesn't mean people follow them. Yeah. I think I heard you say even on the customer thing, like listen to their pain points but don't necessarily listen to their solutions. I thought that was interesting.
I don't know. There's like this MBA way of building companies that I think is like very bad, which is like you go out and interview like 500 people and you listen to what they want and then you build that company. And that's like very rarely something like people want.
You know, take their pain seriously, but don't build what they're asking for. Like that's something I've been kind of adamant about is like people have all these random ideas of like what they want, but often that's not exactly what they need. And that's of course like that's the job of a startup is to like come up with a product. Like if the problem was like easy to identify, like there would already be a solution. So it's like sort of almost like by construction the whole point of a startup is like come up with a solution that people didn't really think of and then put it in front of customers.
Yeah. How did you go about that? We're going to skip forward. We're going to go back to KTH and Spotify as well. But how did you go about that in the early days? Because I've heard you say like this was a product you wished you had. I imagine, and this might be naive, it's such a hard product to iterate on quickly. Tell me if that's wrong to get that customer feedback. Did you just have such conviction yourself in the problem? Or what were those early days like of starting Modal?
Like I think there's like the sort of stereotypical YC company is like you apply for something and then by the time you're in YC you already pivoted five times and by the time you take money you pivoted another nine times and it's like you try things, like kind of throw spaghetti at the wall. And it's like, no disrespect, like there's an enormous amount of very successful companies that come out of YC that are very, you know, successful. And but like cliche of just like iterating and trying different things. I think when you're building infrastructure, like infrastructure is hard to build. The most extreme example would be like building a database. If you build a database, you kind of need to sit down for two years and just like make the database work. You know, if it doesn't work, like what are you even trying? You can't refute, you know, a hypothesis fully unless you actually like in a way like build the MVP. And so I think with infrastructure, like you know, you need to sit down for two years and just like build it. The nice thing is like when you emerge out of the cave, hopefully if the thing works, which is a big if, it may still not work, but if it works, like you have a little bit of a moat because you know you invested a lot of technology into this. So yeah, so like I had the confidence of, you know, what we're working on. This is also during the ZERP era. So it's like kind of exuberant in many ways.
Although you guys didn't raise a crazy round like...
No, we didn't. But like I don't know, it was like kind of happy days. Like the stock market was like going bananas during the pandemic, you know. So I think that helped me a little bit with confidence. But like yeah, so like we sat down for like a couple years and just had a lot of conviction in like what we're building will solve a bunch of problems. Which problems? Like we don't know. But like we felt like there's something here, there's clearly something here, like this will be valuable somehow. And then it turned out like it was, you know, maybe that was lucky. I don't know. But you know, I think it's just a different way of building a product.
Yeah. Well, to your point, you probably have to do that with hardware because there is just such upfront investment.
Yeah. I mean, we're not hardware, but like infrastructure, right? Like anything that's like, you know, harder, right? Like that's not like... I mean, consumer companies I think is a little bit different in many cases, although Spotify is actually quite similar to Modal, like in the sense Spotify took several years.
Is that just because the way... I mean, because early days Spotify, like you had to do the peer-to-peer sharing mixed with like there were some innovations there. Is that or...
There's a lot of that too, yeah.
Yeah, okay. All right, well let's go back then. The Biography Podcast is presented by Rho, the all-in-one banking platform for startups. Thousands of startups like Perplexity, Product Hunt, and more use Rho. You get everything you need to manage your startup's cash: fast banking setup, cards with up to 2% cash back, and yield that turns company cash into extra runway. All super important in the early days of launching. But the thing founders really love about Rho is their team. They are obsessed with helping founders disrupt the status quo and will go to the end of the earth to help them do so. And exclusively for Biography Podcast listeners and viewers, you'll get a $1,500 statement credit plus a ton of exclusive perks when you manage your company's cash with Rho. To learn more, visit rho.co/biography. Terms and conditions apply. Rho is a fintech, not a bank. Checking and card services are provided by Webster Bank, member FDIC. See rewards terms for details.
Because you're a graduate in Sweden. Why join Spotify? Because I think you've said you joined as the 40th employee. Was it clear then already that this would... you know, I don't know, like you probably had a ton of opportunity. Why join that company? Obviously a great bet.
Yeah, I know, it worked out. So maybe let's kind of rewind. The year is what, like 2008 or something like that. So I did a lot of programming competitions and then when I joined KTH there was a bunch of people that were a little bit older than me that were also doing a lot of programming competitions. So I got to know them, they're really... And at the same time, I was also working on with a friend this like book recommendation system. Like we built this website for book recommendations to Sweden. It's kind of a side project, but arguably it was like my first startup. Like we never raised money or anything, but like we generated some... Yeah. We had like a few thousand dollars in revenue every month, which is like, you know, it's fine, you know, when you're like 20-something still in school.
Anyway, so I got in touch with the Spotify people who all these like programming competition crew, a lot of them ended up joining Spotify. And then they were like, well, Erik is like building recommendations, like maybe we should hire him too. So I ended up meeting the then CTO for coffee and he was like, yeah, like maybe you should join Spotify and build recommendations because we're going to need that. And I was like, hell yeah, that sounds awesome. Can I do my master's thesis at Spotify? Because I was looking for a master's thesis topic. So I joined Spotify as like kind of an intern. It was like kind of unpaid, I think, from what I remember. Actually, maybe I was paid, but it wasn't like a full-time job exactly because I was still in school. But I spent six months like basically with Spotify building a music recommendation system. And then they hired me early 2009 to do music recommendation.
Immediately basically turned out like they had too many other things I needed to do. So like I got pulled in. They're basically like, okay, stop working with the music recommendation system for now. We need someone to do business intelligence. So then I ended up doing like reporting, business intelligence for a couple years, which actually was like super valuable later in my life because I learned a lot about like how to think about a business, like how to make... I was making a lot of board decks, my first job, like pulling data, like understanding customer retention, customer behavior, like things like that. So I ended up doing that for a couple years while actually working on the music recommendation system in the background, complete like skunk works. Like I didn't tell people. I just kept working on it. So I was like, obviously this will be valuable one day, but no one told me to work on it. So that was my first few years at Spotify.
I think it's insanely underrated how hard it was for you guys to crack the algorithms and like how early you guys were doing machine learning even at Spotify. What was unexpected? Because it sounds like a downgrade, you know, doing these reports and working on less. What were some unexpected lessons from that? Like is it just like you said, just seeing how the business operated and getting a more holistic view of that?
You know, the chief product officer, the CEO, like they would come to me and like I was this like random new grad and they're like, you got to get me this data like now, like we have a board meeting tomorrow. So like I don't know, I was kind of... I like that sort of excitement and the sort of influence that I had. And it was like fine, but like, you know, from a theoretical perspective, it's probably not that interesting. It was like not like hard math or anything, but like I don't know, like I gained like a lot of appreciation for like how to use, how to pull data that sort of helps you answer questions about a business, which I find like super fascinating. You know, like I'm happy also not doing advanced math. Like if at the end of the day, I think one thing I learned a lot about Spotify is like at the end of the day, like I'm just excited about building a great product, you know. And Spotify at that time, Spotify is weird because there's always product-market fit. It was very unclear if there was a business model. Business model fit was unclear, like will we get label contracts and all these things. But it was very clear when we put it in front of customers, like people loved the product. And I think that was like really, it was just amazing to work at that company at that time because all my friends were using it, like all my family, you know, like I don't know, it was fun to have impact at such a level with, you know, and build a consumer product at that scale. So that was like ultimately also really exciting.
Yeah, 100%. Yeah, it's a great... I mean, it's one of the best European companies, one of the few actual European tech companies. And you're right, like the more you play, the more you pay principle and all these things that you had to do a lot of business model innovation as well. I'm curious then, like what made you want to... at the end of your tenure at Spotify, you were there for a while, like you saw the company I think grow probably from 40 to like several hundred plus, you know, probably a thousand plus people. That's insane.
That's insane. Walk me through then, you know, the end of your tenure there. Like what made you think, okay, now it's time to move on and...
So I ended up spending a couple years like doing this reporting stuff and then I actually quit Spotify because I wanted to... because I got offered a job at a hedge fund in New York and I had a summer internship in New York. I really liked the... I was really excited about maybe like go back and do some like hardcore math, make a bunch of money. So I moved to New York and this is early 2011 and took a job here. The job was fine. I didn't love it. Like basically like I just like missed Spotify a lot. And so just like five or six months in, Spotify opened an office in New York and they basically reached out. They're like, we need people here in New York. Erik, you're in New York. Do you want to come back? And I was like, hell yeah.
No. It was because like I realized like finance is not for me. Like I just like lacked, you know, the sort of meaningfulness of building a consumer product. So I came back to Spotify late 2011 and then at that time they were like, yeah, Erik, we got to do music recommendations. Bring back your system. And I was like, great, like I have this system I've been working on, you know, so let's plug it in. And it actually worked pretty well. So I ended up spending another three, four years almost in New York at Spotify focused mainly on music recommendation systems. We built a whole team of about 20, 25 people.
In the end, there was like Spotify got very big, got very messy. Like I'm kind of a person I kind of wanted to like build like cool systems and optimize stuff, but Spotify made this acquisition and it was kind of just a lot of stakeholders, a lot... and in the end I was kind of burnt out. So I left Spotify in early 2015 and joined this fintech startup which was 10 people back then called Better as a CTO.
You mentioned Better. That sounds like a pretty intense experience as well.
Yeah. Yeah, it was a weird experience.
Extremely educational. Better was like a weird company. I mean, it was like fintech and I think like me and the CEO got along really well. Like, but you know, he's nuts, like in a good way. Like he wants to change the world, very smart, super mathematical. I mean, in my five years there, we grew the company from 10 people to 10,000 people. I grew the tech team to 400 people. I mean, in the last year I left there, the company made, you know, quarter billion dollar in profit, which was like the fifth year of the company. So some stuff was going really, really well. However, a lot of that growth ended up what we realized was like fueled by low interest rates. And so when interest rates went up at the end of the pandemic, or sorry, during the pandemic, the company kind of imploded. And at that time it was like we were supposed to go public and they had to have like big layoffs. I mean, there's like a famous clip that leaked and, you know, suddenly the company went from like glory to like it was like suddenly this like disgraceful company. I at that time had already left.
But I mean, I don't know, it was a crazy ride, like going from 10 people to 10,000 people in like five years.
So I owe a lot to at least like learning about how to build organizations, how to hire, how to structure things, how to not get ahead of yourself in terms of, you know, revenue and growth. Because we did overhire a little bit. The company eventually ended up going public two years later in 2023, but like at a small fraction of its original value. So it was a wild ride.
What did you take away from that? Because I imagine, you know, you're then two years building this infrastructure, this new company. I feel you said you overhired, right? Like that seems like it'd be really hard to judge in the moment. I'm curious what models you do have because you guys are growing really fast right now as well. And I know you've talked about like these AI companies like initially weren't maybe the exact target company or customer, but they've just such a perfect customer for the solution you guys had. I think it was Midjourney that was sort of an inflection point for you guys.
No, Midjourney, Stable Diffusion. Stable Diffusion as a model when it came out being...
Yeah. So maybe just to, you know, keep it in order. So I left Better. This was during a pandemic. I kind of at that time I also just had like two kids. I was again like a little bit burnt out. Like basically took like half a year off. And started like hacking on various ideas. And Modal is like actually my third idea. Had a few different ideas before, but in general like I just kept coming back to like it's hard to build AI, machine learning products, data products. And so I started hacking on something and that I got like super excited about. And by the way, like maybe this also like why I ended up with Modal. So the story is roughly...
When I was at Spotify, I open sourced a couple of things. One thing was like a workflow scheduler called Luigi, which was like moderately successful at that time. Another thing was a vector database actually that Twitter and a bunch of Instacart and a bunch of other ones ended up using called Annoy. But anyway, so I came back to sort of Luigi in 2021. I'm like, I should have maybe started a company around that. And then like I started playing around with the idea, but I realized it's kind of hard to do a workflow company because it's hard to make money unless you own the execution layer below. And I was like, why don't I just focus on that instead actually. So I kind of paused this all work on workflow scheduler and then I was like, why don't I just build an execution layer for like arbitrary workloads targeted towards like data, AI, machine learning. And that was like my third idea, like kind of like startup idea. I wasn't really sure if I wanted to start a company, but I was just like kind of hacking on stuff and suddenly I was like obsessed with it. Like I was spending like all my time like just writing code and like building stuff, which was kind of fun because like I went from like managing 400 people back to just like writing code myself all the time.
And so started hacking on it in 2021. Actually joined as a co-founder. We formally incorporated in June 2021, raised a little bit of money early 2022. And again, this was during the ZERP era. Like it was like kind of take some money. I was like, sure, great. But you know, we didn't grow like super fast because we felt like there was like so much like foundational technology we had to build. We pretty quickly built our own file system, built our own container runtime, like all these like deep like technology. Then in like late 2022, this like almost like two years in at this point, we barely barely had like any revenue. Like we had like almost like no customers. Like the customers we had, we had like basically begged to like use Modal and they were like friends.
How much had you had burned up until that point?
We never burned that much capital. We were like kind of small.
All right. I mean, also initially...
Also initially like a self-funded Modal. Like I was like, I invested my own money into Modal. And so like initially I was like very like lean, like this is my money, like, you know, then we took external capital at some point. So we were only like four or five people even like two years in, maybe six people. But yeah, anyway, Stable Diffusion came out that year and around that time, ChatGPT and Stable Diffusion actually came around, came both came out like I think October, November, something like that. September maybe 2022, I believe. And yeah, a lot of people just started coming to us and they were like, yeah, actually, can we run this on Modal? And it turned out we're like quite well positioned to support a lot of those workloads. So that's when we went from like monthly revenue went from like 100 to 1,000 to like 10,000 in like a few months to like 100,000 almost, not quite, but like something like that. Like it went like nuts. Like over like it was like end of 2022, early 2023 and we felt like we might have something here, but I was still like kind of nervous about it, but it felt like we had something.
Why were you nervous? Was it because you just come from this experience of growing really fast and or...
I think so. And there was like one big customer that came in like early 2023. I was so paranoid. Like I remember like when they put all their like load on Modal, like I set my like alarm at like 3:00 a.m. in the middle of the night because I just wanted to go up and make sure like everything was like running well. I didn't need the alarm because I woke up anyway. And like I just like woke up and like checked my computer like, oh, okay, things are still running, you know, and then I woke up like 5 a.m. like, okay, things are still running. Like I just, you know, and so that was like, I remember that. And but I was still nervous because it was just like one customer. So then like I think like March or April there was like another big customer that came in and then like May there was like maybe a third customer. Suddenly we had like a little bit more like revenue diversification. Like we had like three kind of big customers. I mean big, but you know, for us they were big, right? We crossed like a million annualized around that time. And at that time I felt like, okay, maybe we actually have product-market fit.
Walk me through that moment where you're like, okay, maybe we have product-market fit now. What did you think were the most important next steps? Was it literally just building out the capacity to serve more of the customers? Was it building out a go-to-market team? Or what did you... I imagine there must have been like a hundred things you could have done at that point.
I came from a consumer background, so I came from Spotify and then Better, which were both like just like mega consumer focused. So like I don't think I thought about like PLG per se, but like basically that was all I knew. Like I didn't even think about enterprise sales. I had zero background in enterprise sales, which is like maybe a weakness. Like I had to learn a lot about it. And so initially like a lot of how Modal, how we built Modal was like we just, we had like an amazing onboarding experience. We made it super easy to get started, write your first code, you know, like almost like to the point where like it's like frivolous. Like you sign up for Modal, there's like confetti fly or whatever. There's like small things like that. Like I spent a lot of time like making sure that like that experience was like magic because I always felt like the sort of the setup experience is like, that's like the first time you see the product. You want that especially to be super good. So a lot of our growth in the early days, it was purely inbound. Like I, you know, we had the marketing channels was like me posting stuff on Twitter, like, you know, because I had a decent following. I'd been blogging for a long time. And so that was like our distribution channel was like me talking about building Modal and like the stuff we're launching.
I think in terms of like what we were focused on in the technology side, like we felt like we had product-market fit in one area, which is like serving these like Stable Diffusion like models. We weren't necessarily sure how big that market was, but we're like, okay, let's like at least like try to go out and get more customers. And also initially I would say at that time there was a lot of like kind of bootstrapped companies and nothing wrong about that, but like I think a challenge was like we would see some of those companies kind of go mega viral and then like next month just die.
We had a little bit of...
Yeah. So like we were indexed to all these companies that would like, you know, kind of just like boom and bust. So I think we started at that point to focus a little bit more on saying like we want like venture-funded startups as customers. And part of that reasoning was just like those customers are probably building something slightly more durable and they have a little bit higher margins and so they're better customers. So I think we launched, that was like maybe our first like growth hack is I think like towards the mid to end of 2023 we launched like a startup program, like basically like the idea like we can give a bunch of credits to people. It's very crude. It was basically like, are you a startup? Okay, here's like $10,000 in Modal credits. But that was like I think our first sort of like growth thing that started working.
And then I think we hired our first salesperson in 2024. But even after that, like I still, we still couldn't really figure out like how do we get in front of customers because we sort of captured like a smaller, narrower market. We were competing with neoclouds. We were, you know, we were not like super indexed towards like LLMs and open-source LLMs were not like a super big thing at that time. So I think like getting to enterprise sales, it wasn't until like I would say like last year, 2025, where kind of figured out like we're like enterprise sales.
That sounds insane given the adoption and growth you guys have had. I mean, pro...
I mean, still like almost all our growth is inbound.
Yeah. I mean, like, you know, it's a good problem to have, but like our sales team is like so overwhelmed with all the inbound stuff, like just like, you know, getting that done.
That's great. Is that because of that consumer mindset? We think you built such a wonderful experience and the product is...
I think so.
I think so. And it's something I took away from Spotify in a way. Like the Spotify had this like magic onboarding experience where like, you know, you try it out the first time, you double click, you know, a song and it just like works and it's like in the cloud. And like I, there's like something similar about Modal, like you run some code and it's in the cloud, like just works. It's magic.
Founder of AssemblyAI, and I was blown away by how thoughtful they are about building quality, scalable products. I'm using AssemblyAI myself, including to transcribe this podcast. Their speech-to-text models are just insanely accurate. Their models go way beyond transcription. I mean, you can identify speakers, highlight key moments, and extract actionable insights from every conversation you have. Now, the cool thing about it is that it's developer-friendly and it's built to scale. So it's flexible, has pay-as-you-go pricing, and I've seen it firsthand. AssemblyAI even transcribed my previous podcasts flawlessly with one click. Now, Dylan wanted my listeners to get some perks, so they've doubled the free credits, giving you $100 to try their models. It's a great way to test out the models for transcription, speaker diarization, and more. So stop what you're doing and go to assemblyai.com/biography to see this episode's podcast transcribed and to claim your free API credits.
And so I think nailing that and then like the other part of it is like just like talking about the product, having a distribution. And I think in the early days I call it like founder-led marketing. Like I think that was just like a distribution channel that like for me felt very natural because I always been a big blogger and tweeter. So I think those two things kind of put together for us was like kind of led to pretty strong like PLG product growth in the early days.
Is there anything you would do differently? Looking back, more emphasis perhaps on your own brand and, you know, getting customers through your tweets or content, or hiring a salesperson earlier?
I think we should have invested in sales earlier. I think I held back on sales, looking, waiting until we had like signal that it was working. I think we should have just assumed that eventually it's going to work. So let's just like hire a bunch of people. We waited, like we went like going from one salesperson to like our second salesperson, that took like a year, maybe more, something like that. So I think in hindsight I wish I'd hired maybe like two to three salespeople like from day one, even though I was like, I don't know if this is going to work. Like I feel like I should have just done it. Like I don't know. It's like one of those things where like in the hindsight I wasn't quite as aggressive as I should have been.
Yeah. What sort of signal could you have used at that time that would have made it obvious that you should have taken that risk or that bet to hire more salespeople?
I don't think it was actually a signal that we should have looked at. I think it should just be like assuming things work out, what are we going to need in a year and just like hire those people. I think a lot of startups don't assume, I don't know, maybe that's wrong, like think that far ahead like things are going to work out.
Yeah. I mean, if anything, I think they assume everything's going to work and they overhire. They hire like, because I think that's like a, you know, you can go in the other, that's the error of the commission, right? Like, you know, some companies they're like, yeah, we're going to need a chief of staff, we're going to need a COO, we're going to need a CFO. I always think is it ridiculous when you like see these like, you know, small tiny...
Companies that barely have revenue and they have like a CFO or something doing... [laughter] What are you managing? There's no finance. You're like, you're not making money, or like a CRO with no revenue. Like, what are you optimizing?
So I think, you know, I sometimes like I struggle with that, but I think I should have in hindsight been a little bit more towards that.
Because I was like too lean. I was like too like I held back a little bit.
How are you investing today then with with the same sort of idea where you're, you know, maybe being a bit more aggressive with how things are going in the future?
I'm trying to be a little bit better in terms of just like thinking what are we going to need assuming things work out. Like, you know, we're probably going to need a CFO. Let's start looking for one already. We're probably going to need, you know, HR, legal. Like we we haven't had a lot of these functions. We we don't have a product team still. I mean, the work structure at Modal is very basic. Like there's a lot of engineers and then there's like sales. That's it. I mean, there's like some smaller stuff. There's a little bit of marketing and design and stuff like that, but like in a way it's like kind of simple and clean. And it's like nice because like it means like things I don't have to think about five things. But but like yeah, I mean over time we're gonna need a lot more. We we don't have an HR team. We don't have a product team. We don't have legal or finance. So so those are four functions like we should probably hire.
That's insane. That's crazy. That I mean I guess it goes to show like if you just build things people want and try and get it in front of them. That's really all that matters. How you're thinking about, you know, these new functions and adding that to the team without because complexity is going to go up. Um, but I I want to I'm sure you're being careful about making sure that doesn't come at the cost at an extreme end of productivity and making sure the organization moves fast. I'm curious how you're thinking about doing that and especially because you've already been part of organizations that have grown really fast. I'm sure you've seen like warning signs or things that didn't work or did work. Um, how are you thinking about making sure the company stays efficient when it grows?
You don't have to necessarily think too much about it. I think up to like 30, 40, 50, 60 people, uh people are like pretty good at like kind of understanding at a high level what the company needs to do. I think as a culture we've always been pretty good at like letting people just make their own decisions and, you know, like what do you think is best for the organization? Just do it. Like I I want to have this like mindset where people just come in every morning and there's like what can I do for the company and just like do that, right, a very high degree of self-organization.
I think what ends up happening when you're like 40, 50 people roughly is like maybe a little bit earlier actually, there's like too much stuff going on. So you need to like assign a little bit more like sort of split things up and assign like sort of ownership over things. That's when it gets like kind of complex because, you know, and there's no good way to do this, right? Like there's so many different ways you can distribute teams. My general feeling is like first of all you should try to collocate teams as much as possible like geographically, right? Like we have multiple offices so that that's like one heuristic. I think the other thing is like as much as possible you want teams that own like business outcomes. Like basically like there are like high-level like business goals that they can own. Like ideally they should have like a P&L, like a profit and loss associated with each team.
And everyone in the company?
I mean like no, I'm just saying like you you're optimizing for a lot of different things. You're not going to get all of it, but like it it's nice. I'm saying it on a team level. I I don't think on individual level.
That that's maybe my Scandinavian thing, but like I I [laughter] don't I don't think individuals should have, you know, too much, you know, like it should be everything should be on a team level.
However, that's also not realistic. Some people are going to have to own more like platforming functions, right? So so like I I think eventually like what you end up having happening is like basically end up having two types of teams. Some are like business verticals and they own like a feature area or like a product area and ideally you have like very clear metrics or KPIs that say like these are the things we care the most about. Or you have like more of a platform team and they kind of serve like everyone and they build just like internal tools for everyone else to use. But but that that that transformation is is tricky and and I think it happens around especially as you go from like 50 people to like maybe 150, like you kind of have to like constantly like sort of restructure your teams to like figure that out. So that's a little bit of a thing I think right now where we're thinking a lot about like how do I set the organization up for success in that sense.
Are you breaking down the entire structure each time and reassigning people?
No, that's impossible. You have to start with the existing teams. You can't move around people like all the time. Yeah. I think you have to like kind of organically like, you know, take a team, split it, and then like take two teams, two teams, merge it. I mean, you're definitely going to split more than you merge,
But but I think that's like or or like reassign projects from like one team to another or like move people between teams like,
