Episode 64
· 32:38
I think building stuff in the cloud and owning the infrastructure just makes so much stuff, like, so much easier. And it's frankly just, like, from, like, a business model perspective, like, so much easier. Because, like, there's such a precedent for, like, running things in the cloud and charging for it today.
Jack Bridger:Hi. You're listening to scaling dev tools. Today, we're joined by Eric Bernardson, who is the founder of Modal Labs. And I Eric can correct me on this, but I think of Modal Labs as kind of like a Firebase or Vercel of doing stuff with data and LLMs. So you can't really run models in Firebase and, I well, you can do it with Modal.
Jack Bridger:That's how I think about it. Eric, Do you wanna jump in and and give a better explanation?
Erik Bernhardsson:Yeah. Yeah. Thanks for having me, by the way. It's exciting to to to be on this podcast. Yeah.
Erik Bernhardsson:I I think I think that's roughly right. Like, we think of ourselves as, a cloud tool for or infrastructure provider for data teams and data products. We've seen a lot of, usage in within general generative AI. So we focus a lot right now on people who wanna run various types of models in the cloud and don't wanna think about infrastructure. We we also support all kinds of other stuff.
Erik Bernhardsson:So it ranges from like, you know, running stable diffusion, text to transcription, speech to text, text to speech, video processing, but also things like computational biotech. We have a few users doing that or 3 d rendering, web scraping. It's a fairly general purpose tool in that sense.
Jack Bridger:Yeah. And you've is you've just, like, kind of rattled, like, quite a lot of different use cases there. And I can see the is is it, like, the general theme is, like, if it doesn't if it's stuff that doesn't work well with, like, traditional kind of serverless stuff? Or
Erik Bernhardsson:Yeah. Exactly. I think we we we spend a lot of time on making serverless work for data, including building our entirely own proprietary infrastructure. So I think conceptually, like, we we're somewhat similar to AWS Lambda in a sense. Like, we you you describe sort of a container and, you write the code yourself, and then we have all the scaling and, you know, resource management, the provisioning, and all that stuff.
Erik Bernhardsson:But very different in the sense that we focus a lot more on compute heavy jobs, that typically require a lot of GPU or, you know, need to run for a very long time, or have some other complex requirement on the environment.
Jack Bridger:Yeah. Yeah. I I can say for myself, I have been building something that use it that does, like, image process sorry, video processing and does transcription and does, like, uses LLMs. And it's so it's it's like so you kinda think that you can just go out and use, like, Heroku, and you can just plug all these things together. But then suddenly, you've got, like, big packages.
Jack Bridger:You've got to store large files. You've got to, you know, have these models, as you said, that take a long time, and it's just
Erik Bernhardsson:CUDA driver versions.
Jack Bridger:Yes. Exactly. And then and then so I've been using AWS, and I have to learn EC 2, how to use that, which is such a pain. And then Yeah. It's expensive as well.
Jack Bridger:And if you've got GPUs, you could be looking at 300, $400 a month, which you're not even using. More. Very much. Yeah. More.
Jack Bridger:So, yeah, I very I think this is a very big I idea, and and business
Erik Bernhardsson:I agree, but I'm biased.
Jack Bridger:So Ash, could you talk a bit about, like, how dev like, developers might because I think this is gonna become more and more common for developers. And could you talk a bit about, like, the developer workflow for people working with these kind of building these kind of projects that use LLMs and models and stuff?
Erik Bernhardsson:Yeah. Totally. So I I started looking at this 3, 4 years ago. I started thinking a lot about, like, building better tools for data teams and was actually somewhat agnostic as to, like, what we would build. I I I really just wanted to focus on making data teams more productive because I noticed this is a tool stack that's missing, I think, for them.
Erik Bernhardsson:Like, they're running into a lot of, like, infrastructure problems and stuff. And and started looking at like what are the things that like data teams that make their workflows unique. And and some of the stuff for instance is like they tend to run a lot of things in production. They tend to run like more like non standard like hardware like GPUs. They they tend to have, like, very bursty workloads, like, you need to fan out and, like, you know, run things in parallel a lot and, like but also, like, there's some, like, simplifications.
Erik Bernhardsson:Like, they almost always tend to use Python. So we decided we're just gonna go all in on Python. So we decided to build this as, like, an end to end Python SDK. So so unlike other frameworks, like, model is not like an API. It's it's it's like it's really just you're running code.
Erik Bernhardsson:And so the way you interact with model is, like, you describe your compute environment in code and it's commingled with the, with the with the app code itself. So, basically, you can, like, define, like, here's my machine learning model. I wanna, like, run this, you know, diffuser model or whatever, and here's the image definition all in code. And that way, we we have a way, a command line interface where you could just like deploy that into the cloud or just like run an ephemeral app. And and the nice thing about model is we spend a lot of time on the container infrastructures and, basically built on a file system with a bunch of other stuff.
Erik Bernhardsson:So in in event and what that enables us to do is we can take code on your computer and launch that in the cloud and execute it in basically a second, in many cases. Which to me is like why like what I wanted to build in the first place. Like I wanted to build something that feels exciting to use. Like feels like I'm almost running things locally. Right?
Erik Bernhardsson:Like that has the sort of same feedback loop. Because I think that's the key to developer productivity is to have that, like, super fast feedback loop. And that's the thing you get with model and like many other, infrastructure providers. Like you don't have to, like, build a container, push the container to cloud, like run the you know, Like, you just, like, run the code almost as if you're running it locally, but it executes in the cloud. And it feels like it's executing locally still because it's so fast.
Jack Bridger:Yeah. Yeah. That's huge. It's like just building containers takes ages with these sorts of projects, right, alone.
Erik Bernhardsson:Yeah. Totally. And this is kind of annoying. Like, I I mean, Docker, I think, conceptually is a very cool tool. I always struggle because I find it from my line interface super confusing.
Erik Bernhardsson:With modal, you don't need to use Docker. Like, we don't use Docker. In fact, like, we don't, you know, you don't need to install anything to use modal. You just, you know, do like well, you have to install modal, which is pip install modal, and then you can immediately start running jobs in the cloud.
Jack Bridger:Yeah. Your, getting started guide is is very fast. I went through that. It's, like, super quick. Yeah.
Jack Bridger:Yeah. Yeah. And could you tell us about, like, what kind of people are, like, adopting Modal, like, and why?
Erik Bernhardsson:Yeah. That it's been kind of a journey in itself. Like, we we we actually it's funny because, like, when I started building this, and and I think still to some extent, my vision was always, I wanna build a general purpose tool. Like, I wanna build something that can run almost anything. Like, almost like a Kubernetes for data team, but but in the cloud.
Erik Bernhardsson:So so for a long time, I I I tried to sell this to, like, a lot of enterprise customers and and, you know, it's it's kind of a hard sales process because you're you're trying to convince them to, like, switch something out. And then about a year ago, we started seeing a lot of traction. Suddenly, there was a lot of, like, new cool gen AI models coming out. And and a bunch of those people, you know, a bunch of people were playing around with those models and trying to deploy them in model. Suddenly realized, like, wow, there's a lot of people out there, like, we wanna run this stuff on model.
Erik Bernhardsson:Maybe we should focus on this use case. And and so that's been, like, the vast majority of our growth in the last year or so has really been, like, enabling that new cohort of in many cases, these these are, like, traditional software engineers now coming into the data world, or or people with, like, other different backgrounds. And a lot of those people, they they don't it there's kinda starting from clean slate. Like, they don't have any infrastructure, and they don't wanna deal with all the traditional, you know, setting up Kubernetes and AWS and all that stuff. Like, they just wanna run something and and make it work.
Erik Bernhardsson:And so realized this is a great opportunity for us to to, to offer serve a a service that's much better than anyone else. So so so a lot of the the usage we've seen so far tends to be, you know, a lot of these stable diffusion, Dreamboost, ControlNet, some of these, like, you know, newer, language models. Like, a lot of people doing fine tuning in model, like running on fine tuning. We see a lot of audio actually. Speech to text to text to speech.
Erik Bernhardsson:We see a lot of people using it actually for, like, music. It's been kind of an interesting use case, like, a lot of, like, music generations. There's some Gen AI algorithms for for making music. It's kinda cool. And then more recently, we've seen a lot of video processing too, which is kind of exciting.
Erik Bernhardsson:I think there's a lot of, like, very interesting stuff where you kinda combine, like, sort of traditional, like, computer vision with, like, some of the new, like, Gen AI stuff, and there's some really cool applications that are running a model.
Jack Bridger:Yeah. That's that's really, really cool. And how would you say these people are, like, discovering model?
Erik Bernhardsson:It's all inbound. Like, it I mean, like, I'm not maybe trivializing a little bit. Like, basically, like, you know, I I I think so far we spent almost nothing on, like, sales and marketing. But that being said, like, I always had a presence on Twitter for instance. I always been blogging.
Erik Bernhardsson:So I think a lot of people have been discovering it through that. Like, you know, I have a little bit of an audience that I've been trying to, like, leverage. So a lot of people, the typical, you know, they they come inbound and, you know, like but they probably heard us. Maybe on a podcast like this or or or saw my tweets or whatever. So I I don't know.
Erik Bernhardsson:I I I think of myself as, like, half, you know, social media intern, head of social media. Like, I spend a lot of time just, like, talking about model and, running blog posts or whatever. And and that so I I think that's, like, been the channel so far that's been driving mostly growth.
Jack Bridger:Yeah. That makes sense. Do when they come in, is it like does that tend to be, like, a frustration that they, like, kind of made them come and try out model? Because it can be scary sometimes to, like, change your workflow in a sense. Not scary, but, like, there's a resistance to try something new sometimes unless there's, like, a real, like, pain.
Erik Bernhardsson:Yeah. I think I I mean, in terms of, like, user adoption and, like, onboarding flow and, like, how people get started, like, I I don't know if, like, people come from, like, a position of, like, hating previous tools or just being curious about model in general. But it, like, I think, for me, like, thinking a lot about, like, bottoms up adoption felt more natural to me. And and so I think model because I'm just, like, kinda hard to explain what model does, I think, to a lot of people. So I I always, like, feel like the best way to to to sell the value of model is to get people to try it.
Erik Bernhardsson:So a lot of, like, how I think about sort of acquisition and adoption is, like, let's get people to the website, and then let's get them to try model in, like, less than a few minutes. Right? So we spend a lot of time, like, optimizing for that flow. Like, how do you get how do you get people to install the app and then run some code in the cloud in in in just a couple of minutes. Like, unlike a lot of traditional infrastructure setups, like, there's actually a pretty complex configuration process in many traditional tools.
Erik Bernhardsson:But with modal, it's like, you just install the Python library. That's that's it. And that's very intentional because I think so I think there's there's a lot of, you know, like, value and, like, feeling that magic. Like, I think the product, like, sells ideally sells itself better. And especially since we're focusing so much on developer experience, I wanna, like, make clear to users in a few minutes, like, we were thinking about it differently.
Erik Bernhardsson:We were, like like, you're running things in the cloud, like, within a few minutes. Like, that is part of, like, why the product is so good. Like, you can actually do that without any hassle. And and also why I think, you know, in general, like, as I look at look at growth going forward, like, Roe is gonna focus a lot on self-service and sort of bottoms up adoption. We've done, like, nothing top down so far.
Erik Bernhardsson:We'll obviously do that in the long run too, but but bottoms up is where it's at right now. And that's why also I think a lot of our use to skew, I would say, like, early stage, a lot of indie hackers, series c startups, and series a startups. That tends to be, like, the sort of earlier part of the company journey. Starting to get some enterprise customers too, but it was sort of that's more in the future.
Jack Bridger:Yeah. Yeah. No. So it it it makes sense to me why you're getting that kind of user, totally. It's like because they're otherwise forced to go and use, like, enterprise tools that just don't make sense for that.
Erik Bernhardsson:Yeah. Totally. And and the one of the things I realized is, like, those are very different parts of the segment. Like, if you look at, like, startups, like, they often, like, start out on one side of the that spectrum. Right?
Erik Bernhardsson:Like, you know, like, we're focusing very much on bottoms up, focusing very much on, like, early stage customers. Many other end you know, tools focus on the other side of spectrum. Eventually, we'll probably, like, you know, try to conquer the whole spectrum. But right now, like, right you know, our acquisition is very focused on on this side of the early stage.
Jack Bridger:Yeah. And, so it sounds like you've kind of found a really nice, place, and, it totally makes sense. What was it like when you, when you got started? And maybe this also we didn't really introduce you, but, for those listening, Eric is a very, has had a very successful career, was, like, one of the early engineers at Spotify. I think was, like, the person that discovered that the freemium model actually works at Spotify.
Erik Bernhardsson:That's somewhat exaggerated. But I I did a lot of the early cohort analysis, and, it was a very exciting sort of, like, for me to understand, like, start up math and, like, you know, like, really understand what makes a product like Spotify work on a economic, you know, and financial side. Yeah. That that was sort of formative experience in my career.
Jack Bridger:Yeah. You're very humble as well. But so and then Eric was also a CTO with I think you had, like, 300 engineers in your team for better.com. Yeah.
Erik Bernhardsson:Something like that. 300, 400. Yeah.
Jack Bridger:So how was it going from, like, being the CTO and having 300 reports to just, you know, going out and starting to you know, the early days before model was a thing. Like, how did you go from there to to model?
Erik Bernhardsson:It's I mean, it's like I don't know. Like, I I feel very good about, like, managing people, but also, like, having done that at large scale, like, I I managed, like, 3, 400 people. Like, the day I quit, I was just, like, I swear to God for, like, a year. And I kinda ended up it was, like, so nice to just, like, you know, sit and just, like, crank out code again. Like, I I felt young.
Erik Bernhardsson:Like, I felt like, you know, like, reborn. Now I'm managing a few people again. There were 14 people. And and I have no qualms about, like, growing that team and, like, you know, starting more, like, doing more, like, management again. But, for for for at least for me, there was something very nice about going from, like, and and I've done that pendulum swing a few times in my career.
Erik Bernhardsson:Like, I'm actually pretty happy, like, you know, going in both directions. But, at the time I started model, like, I went in, like, full on, like, hacker mode for, like, 2 years. I'm, like, finally coming out of that cave.
Jack Bridger:Okay. And so how did you how did you come out of the cave? Was it, did you try lots of other ideas before you landed on model?
Erik Bernhardsson:Yeah. I I had a few other ideas. Like, I I thought a lot about you said I was, like, brainstorming about some prototypes, and I didn't really, like, feel like it really exciting. And then it and then, like, I started thinking about the data stack in general. And, originally, I was actually thinking a lot about workflow scheduling because I I have a background.
Erik Bernhardsson:I built a workflow scheduler back in, like, 2011, 2012, Cluigi that a bunch of people use. So it's always been a space I've been, like, thinking about. Not a space I always think about is vector databases. I built a vector database in 2012. Now it's, like, kinda funny.
Erik Bernhardsson:It's, like, coming back. But yeah I built a workflow scheduler. I I I thought a lot about like you know maybe I should like turn that into product and then it's like actually like workflow scheduling is kind of annoying to it's like sort of hard to work with and like you really wanna own the compute layer in order to like for for a lot of the data stack. Like like a lot of the fundamental problems with the data stack, I think really and what I realized is like really stems from like on the compute layer like being like hard to work with. So I stopped working on the workflow scheduler, still have the code super actually.
Erik Bernhardsson:It's kind of fun project. But yeah, I started working on what's now model, like late 2020, early 2021. And pretty quickly realized like this is what I wanted to do. That being said, I think a lot of the vision is very intact, but like, they're like how I think about like use cases has completely shifted. Right?
Erik Bernhardsson:Like I I started building something that is, you know, I thought it was like this is gonna be a replacement for Kubernetes. And then like I mentioned like we we realized like, you know, year and a half in, like we barely had any users. I had a lot of confidence in the platform, being useful for something, but, it wasn't really clear, like, what's the killer app here? And then all this, like, gen AI stuff started happening. And so that that was obviously very kind of, you know, serendipitous discovery that, you know, was great for us.
Erik Bernhardsson:It would have been like, I think we would have found something else eventually. But I wasn't very happy we have that sort of wedge like that. You know, now that's like one place in the market where we have product market fit. Now the next challenge is like expanding that. So I think I think now, you know, now that, like, growth is, like, starting to work, and I feel like the, like, core technology is, like, kind of on we understand a little bit better.
Erik Bernhardsson:Kinda going back to your question now, it's like for me, like, okay, let's come out of the cave. Let's, like, try to focus more like user acquisition. Because I I I feel like basically the last two and a half years, I've been, like, deep in, like, you know, the rabbit hole, like, building infrastructure. Yeah. But yeah.
Erik Bernhardsson:Now I'm ready to talk to customers.
Jack Bridger:And that yeah. And, actually, like, that's kind of a question that people are like, you're kind of going off to people working with LLMs, I guess, for customer acquisition and, like, how where do where do people building with LLMs, like, hang out and stuff like that?
Erik Bernhardsson:Yeah. That's a good question. I I would actually say, like, it's been less LLMs than I would say, like, other gen AI. Like, LLMs are very API driven, so a lot of people building on a LLMs. Like, the money tends to like, all the money tends to go to, like, OpenAI, in a sense.
Erik Bernhardsson:Like, it's, like, less common that people run their own LMs. We we're starting to see a lot of use people doing more, like, fine tuning and especially there's some, like, interesting tools from, like, structured outputs. But to answer your questions, like, where do they hang out? Like, I I don't know. Okay.
Erik Bernhardsson:I mean, that's Twitter or at, like, you know, like I I think I'm in New York. So, like, I think San Francisco is probably a little bit more like a physical, like, scene. Like, I I get the feeling, like, people hang out actually in person and go events and, like, talk all the time. There is a little bit of that in New York as well, but I think it's less pronounced. Like, I think I don't know.
Erik Bernhardsson:Like, the the the the positive side of me says is because, like, people are so busy building. They don't have time to, like, go to these events. But I I don't know. Like, I I'd imagine it'd be a very different sort of, you know, experience being in San Francisco, but New York is also a good place to be. And we're doing reasonably well just, like, you know, finding people on Twitter or, like, we have a Slack channel that people join and sometimes, like, we start talking to people that way.
Erik Bernhardsson:I I think that's been the general experience, like, you know, finding users.
Jack Bridger:Yeah. Yeah. That makes sense. And, I know you've talked about a lot about, like, developer experience and stuff. And have you had, like, any kind of process or, like, things that you would recommend that other people do, to kind of ensure that it's good experience for developers?
Erik Bernhardsson:For their own like, you're talking about, like, dev tools, founders?
Jack Bridger:Exactly. Yeah.
Erik Bernhardsson:I I mean, I I think, like, a couple of, like, things I feel very strongly about, but it I also, like, feel like it's, like, every product is, like, you know, everything is, like, idiosyncratic, like, not everything translates. But, like like, just to tell you, like, some of the things that I felt pretty strongly about is, like, one of the things is, like, I was always very adamant that I wanna run everything in our cloud and, like, just build, like, a multi tenant system and not compromise on that. Like, who knows? Like, maybe if someone in, you know, in the future wants to pay us a lot of money to run it on prem, like, we'll maybe have that conversation. But I think building stuff in the cloud and, and owning the infrastructure just makes so much stuff, like, so much easier.
Erik Bernhardsson:Like, you have full control of the infrastructure, you can upgrade the code whenever you want. It's frankly just like from, like, a business model perspective, like, so much easier. Because, like, there's such a precedent for, like, running things in the cloud and charging for it today that, you know, people sort of understand. Like, okay. I'm paying for your you to run this code in the cloud or or running this product.
Erik Bernhardsson:So I I think that's, like, one thing that I've personally, like, you know, always been arguing for. Like, you're always gonna lose some customers that for security compliance reasons, like, can't run the thing in the cloud. I feel like we have the, you know, time on our side. Like, I think every year, like, people get warming more and more warming up to the idea of, like, putting their stuff in someone else's environment. You know, starting with the cloud in 2008, like, you know, but now also, like, with products like Snowflake or whatever.
Erik Bernhardsson:Some other things that I I spend a lot of time thinking about is, like, just to, like and we talked about it, but, like, the onboarding flow. Right? Like, I think to me, like onboarding flow, a lot of like onboarding, like user acquisition, like the the account creation, feels like kind of jankily put together. To me, that's like the core product is still like how do you get started. Right?
Erik Bernhardsson:Like the core experience of the product is, you know, how easy is it to get set it up and and launch this. Similarly, I would say about docs. Like I feel like a lot of products, like completely mess up docs. Like they have this like amazing product, but then you go to the docs, like, you know, like, one of the things I often do when I look up other developer tools is, like, like, I wanna understand it, so I go to docs. And then there's this wall of text of, like, key concepts.
Erik Bernhardsson:We have a sprocket and a pudonga and a kokimoki, and like, what like, what does it all do? Like, how do I, like, combine it? Like, just show me some code. Like, how do I, like, build, you know? So that that's something where I feel like like, just, you know, putting together a bunch of, like, hand like, nice sort of, like, examples of, like, how do I get started?
Erik Bernhardsson:Like, I wanna take this quote, and so when I run it, it feels like such like a last mile thing that that I I noticed a lot of dev tools, that product, you know, could really do much better.
Jack Bridger:Yeah. And I'm sure, like, everyone agrees with you in a sense that, like, these things, like, if you said to any dev tools founder, like, do you care about docs? They're like, yeah. Okay. Docs are important.
Jack Bridger:But then why do you think, like, some of them, when you go there and it's like, as you said, that there's just like there there's a wall of text. They're telling you loads of, like, made up words that mean something in the in that tool. Why do you think that's, like, so common?
Erik Bernhardsson:I I feel like sometimes, like I don't know if this is necessarily true, but, like, one model of this that I have is that I think people approach, like, docs as, like, they approach code, which is, like, when you, like, write code, you, like, you take your thing, you break it down into, like, you know, like things that a computer understands. So that's why like you end up with this like key concepts. You have to write this YAML. You have to do this. Like but like you're not writing for a computer.
Erik Bernhardsson:You're not telling a computer what to do. Right? And and when when you're like explaining, you know, when you're talking to a human, it's a very different thing. Like you need to like, try to think, okay, what is the mental model that this human has of your product? Like how do I convey the right way to think about the product to them?
Erik Bernhardsson:Thing to do. I think a lot about this, like, with, like, designing APIs and designing SDKs as well. It's a very hard thing because you're, you know, like, writing code for computers is kinda easy, but now you're actually writing code for a human, you know. And so you need to think about like what their mental model is of your product, and like how they would wanna write the code. And so, you know, so so and I don't know, like it's a little bit of a different mindset that I think is not as common in software engineering.
Erik Bernhardsson:I I don't know. That that that's, like, one of my fears for, like, why some of these things are very hard.
Jack Bridger:Yeah. I think that's a really good point. I I I agree. It's easy to just break it down rather than putting yourself in the mindset of others. And I guess, like, watching people sometimes, I don't know, like, see
Erik Bernhardsson:I think it's really hard. It's like I don't know. Like, I imagine, like, making music is very hard because, like, you keep hearing your own music, but then, like, you have to picture what is, you know, what does it sound like to a person who's hearing it for the first time? Like, I don't know. I it's like to think about that as, like, kind of weird analogy.
Erik Bernhardsson:Like, I don't know. It must be very hard.
Jack Bridger:True. True. True. And that's a good segue because, you you're bringing up music and that's the Spotify connection. And I know that you were doing
Erik Bernhardsson:That's right.
Jack Bridger:A ton of, a ton of, like, data stuff throughout your whole career. So to kind of flight step down a level, like, how are you using data, at model? And, like, are there any things that you would you found to be helpful?
Erik Bernhardsson:It's funny. We actually don't use it that much. Like, I I I don't know. Like, I to me, it's like data is always a tool. Right?
Erik Bernhardsson:And, like, I I'm like wouldn't like, you know, if you talk about, like, who who cares about data, like, who wants to build products of data in a way? I'm probably, like, the top percent. Like, I'm obsessed with, like, using data when I can, but I'm also like I you know, to me, it's just a tool that like, in some cases, it's very hard to get the right data. Like, we're early on. Like, we don't necessarily have, you know, 100,000,000, you know, users, whatever, using us.
Erik Bernhardsson:Like like, the sort of you know, we have a very small end of customers that, you know, use us. So we we use data a lot, like, internally for, like, understanding, like, operational stuff, like understanding latency, understanding utilization, like, stuff like that. But but it's been hard so far to look at, like, the onboarding funnel and understand the conversion rates. And is it going up? Is it going down?
Erik Bernhardsson:Because frankly, n is just not big enough, which means, like, I think you have to be more quantitative qual qualitative. You have to kind of just, like, look at it. It's like, you know, it's like design. Right? Like design when when you Google, like, you can, you know, run an AB test and try, like, different, you know, 55 different shades of blue.
Erik Bernhardsson:But if if you're not Google and don't have a 100,000,000 users, you're just gonna have to use your own, like, preference. Right? Like, this blue looks better than this blue.
Jack Bridger:Yeah. Yeah. Okay. That's great. That's great to hear.
Jack Bridger:And I think probably, like, some people will seek find some reassurance there from you. But when you say, like, looking at qualitative data, like, how how has that been for you? Like, to how are you looking?
Erik Bernhardsson:A lot of it is, like, talking to customers, frankly. Right? And and and that's, you know, so so we can't necessarily be data driven, but we can talk to a lot of customers and, like, we can get sort of qualitative data from customers. And I think that's very important to me because you have to have a feedback loop. But one of the things I care you know, I I thought was very important for Modal is that even though it took us a year and a year or maybe a year and a half to start having, like, substantial amount of, like, real users, like, using those in production, we always try to put this in front of users.
Erik Bernhardsson:Initially, that was just, like, friends and, like, people I knew and, like, you know, I I would demo this and, like, you know, hear people's reaction. And I think it's very important to have those types of feedback loops. And and one of the things is, like, people will generally, like, give you feedback, but it's often, like, kind of you have to, like, go a level be beyond what they're saying. So it is trickier with, like, this this, you know, human feedback. But, like, often, you'll hear feedback like, oh, it would be really cool to be able to do x, but then they have to kinda think, k.
Erik Bernhardsson:Well, why they're trying to do x? And then maybe there's something else that would achieve the same purpose that's actually much simpler. So I I I try to do that a lot. Like, I try to talk to a lot of customers. I try to listen to, like, what features they wanna have from model.
Erik Bernhardsson:And then I kind of ignore that and try to understand, like, what it actually wants and then build that.
Jack Bridger:Yeah. That makes sense. That's I think I I think that's is it like the classic example of, like, people complaining about, like, the how long the elevator takes and then someone just put, like, a mirror. And then Yeah.
Erik Bernhardsson:Yeah. Yeah. Something like that.
Jack Bridger:Right? Like They're bored, basically.
Erik Bernhardsson:Yeah. I think that's right.
Jack Bridger:Okay. Super interesting.
Erik Bernhardsson:Should put a mirror in the in the model experience so you can see yourself.
Jack Bridger:Thought that. Like yeah. If you just felt like a webcam, people wouldn't
Erik Bernhardsson:Exactly. Yeah. Except
Jack Bridger:for the fact they're gonna have to, like, agree to display their webcam. But yeah.
Erik Bernhardsson:Yeah. Because, like, here here's, like, an example where, like, I think sort of illustrates the the the the the point I'm making, which is it's specific about waiting is that I think traditionally, like, you talk to a lot of people, they wanna have ability to run infrastructure locally.
Jack Bridger:Yeah.
Erik Bernhardsson:But then, like, why do they need that? Like and that's, like, what I realized, like, at some point, I had this revelation that, like, actually, the reason why they wanna have an ability to run things locally is because it's fucking slow to run things in the cloud. But what if we fix that instead? What if we actually make it fast? And so that's why I got obsessed with this, like, okay, what if we actually make, like, cloud running things in the cloud take a second?
Erik Bernhardsson:Now actually you don't need to have a local environment replicating infrastructure environment. It seems to get so much easier because now you have just one environment. It's running the same thing. That's an example of like how I think sometimes people like kind of vocalize like their problems. Like, they did, you know, come up with solutions that are actually not the right solutions.
Erik Bernhardsson:You kinda have to look beyond.
Jack Bridger:Yeah. Actually, like, rogue question. When I did it, yeah, it was, like, super fast. And is there any, like, kind of do you ever have any issues when people just don't realize what you've done because it was so fast? Like
Erik Bernhardsson:Yeah. I don't know.
Jack Bridger:I don't know. I mean,
Erik Bernhardsson:I yeah. No. I I don't I think I think people realize it's still running the cloud. Right?
Jack Bridger:Yeah.
Erik Bernhardsson:Yeah.
Jack Bridger:Yeah. Okay. If you had any advice for any founders listening, what what would you share?
Erik Bernhardsson:To me, the best advice or, like, the the like, what I always, like, think like, again, like, I think, like, advice is, like, very, like, has to be, like, reinterpreted, you know, within the context of every company, because every company is different. But like, personally, I always feel like the more I spend time on the core product and just like build like good stuff, you know, the easier things get. Like I like I I'm you know, I I feel like the like advice today is like you need to be out there like selling and you'd be out there like doing marketing. Like I actually didn't do that for like two and a half years. Like now I'm starting to do it.
Erik Bernhardsson:Like, I just focus, like, all of my energy on, like, let's just build a good product. And and so, you know, but but validating that with real users, like, what good it is, like, you have to validate that. And and, you know, so far, up until a few months ago, we were to 12 people out of, you know, out of which 100% were software engineers, where we were 100% software engineers very recently. So so that's like one thing where, like, I feel like I've gone a little bit against the grain, like, just focusing on just, like, only our software engineers. Everyone just builds the product, spend all the time just, you know, focusing on like, building the absolute best product, and and almost nothing like trying to market it.
Erik Bernhardsson:That's maybe like one of the things where I don't know. Maybe maybe I'll have to, like, you know, maybe I'll regret this in in the future, but, like, so far that's been, going pretty well for us.
Jack Bridger:Yeah. That's awesome. Actually, I think we've had some similar advice. It's not common, but some people have said this as well. So definitely seems to be a winning formula.
Jack Bridger:Yep. I guess there's a few. Yeah. That's awesome. So if people want to learn more about, Modal, or if there's anything you wanna share, Eric, let us know.
Erik Bernhardsson:I mean, I I I think the best thing is to just try it out. Right? Like, go to modal.com, m o d a l. Play around with it. If you like it, that's awesome.
Erik Bernhardsson:If you don't like it, I was hooked to understand why. Like, I'm always, like, curious to hear, like, feedback or whatever. But, yeah. Like, I I think the best way to, like, learn more about model is to to to to to try it out or or read our docs. Model.com/docs.
Jack Bridger:Awesome. And very nice domain. So, yeah.
Erik Bernhardsson:Thank you.
Jack Bridger:Thanks everyone for listening, and we'll be back again soon.
Erik Bernhardsson:Awesome. Thanks for hosting me.
Jack Bridger:Thanks, Eric.
Listen to Scaling DevTools using one of many popular podcasting apps or directories.
