00;00;00;00 - 00;00;19;46
Unknown
Spin up times for Daytona is 60 ish milliseconds, and that is about half the time it takes a human to blink. And to put that in perspective, that is including network latency. So your system will say, I need a sandbox. Here's the config. Give it back. It was actually quite shocking because when he showed it to our third co-founder, he was like, dude, what the hell is this?

00;00;19;48 - 00;00;46;04
Unknown
This is like code anywhere at 2020, you know, 2016. So you started your company 12 years ahead. Our intuition was that not only coding agents will need sandboxes or runtimes or computers, but every single digitized knowledge worker will need effectively, a computer. So we decided like if this is the future, all of these app layer companies or agent app layer companies will need this new primitive.

00;00;46;09 - 00;01;05;30
Unknown
When we're thinking about like the go to market strategy, you actually have to care about the person that you are, quote unquote, selling to. The one that we kept coming back to. There's others.

00;01;05;34 - 00;01;24;33
Unknown
Hello and welcome, everyone to the fifth episode of our podcast, the Merch brought to you here from our San Francisco office. I am Code Rabbit today. I'm super excited because I'm joined by Evan De, CEO of Daytona, one of the most prominent NIO cloud companies. I would say in the age of AI just closed series A 24 million.

00;01;24;33 - 00;01;42;24
Unknown
Yeah, that goes for an introduction. Thanks so much. First time someone, addressed this is a Nio cloud. So exciting. Yeah. I mean, I guess, what would you call it? I don't know. Definitely a cloud. It is. I feel that nio clouds are more like GPU focused, which works slightly different, but I'm taking it as a compliment regardless.

00;01;42;35 - 00;02;01;09
Unknown
Thank you. Well, actually, let's let's take that tangent and give you an opportunity to to explain everyone a little bit more about what Daytona is. And, actually who you are. I love to you have such an interesting background. And, for those people who haven't heard about you and Daytona yet, what should they know about?

00;02;01;23 - 00;02;21;41
Unknown
Oh, so, I mean, let's talk about Daytona that we can't just get into backgrounds. We talk about it through it. Daytona basically is a cloud that offers composable computers for agents. And I say composable computers for a few reasons. But first off, people call them in general, the market calls them sandboxes. And when you log in to Daytona, they say sandboxes.

00;02;21;46 - 00;02;45;57
Unknown
I just feel that what we truly do is much more than a quote unquote sandbox. It literally is a composable computer. And so what I mean by composable computer, for anyone listening that has, you know, screwed together a PC in their entire life, like you have, like the chassis, you have a, you know, motherboard, you, you know, plug in a CPU, you add Ram, you had a hard disk, you might add an external GPU, depending on, you know, are you playing video games or not?

00;02;46;11 - 00;03;04;26
Unknown
And then you install an operating system on that to install a Windows or Linux, let's say for the sake of argument, you can also install a mac. You can't because of whatever. But let's say you can and so if you do that, it takes about a few hours, you know, to get that up and running. And in the world of AI agents, agents need computers for everything.

00;03;04;27 - 00;03;23;13
Unknown
And I think it's finally becoming consensus when we've seen, like, the open core, you know, craze, people buying MacBooks, not MacBooks, the Mac minis and whatnot. Because finally, understanding that agents need computers and then you have for much more things than that. But when an agent needs a computer, an agent, we also need different types of computers for different use cases.

00;03;23;18 - 00;03;43;08
Unknown
And so in that vein, Daytona can offer the computer composed two specification for that use case. So for one use case it can just be like one CPU, one gigabytes of Ram. Linux. You're done for another one. It could be like, you know, four CPUs, a bunch of Ram and have a windows on that to run a windows like, there's different combinations for different use cases like you have here.

00;03;43;17 - 00;04;05;01
Unknown
You know, a MacBook, you might have like a whole large computer to do 3D rendering. Right. And in that same vein, Daytona offers these computers aka sandboxes to agents, and we offer them really, really fast speed up times for Daytona is 60 ish milliseconds. And to put that in perspective, that is including network latency. So your system will say, I need a sandbox.

00;04;05;16 - 00;04;26;39
Unknown
Here's the config. Give it back it. The whole thing happens the the network latency to Daytona up back and getting up and running. And that is about half the time it takes a human to blink. So just to put that in perspective. So that's what we do at Daytona. Wow. Yeah. Very impactful work. I was just, as I was telling you earlier, playing around with it last night, and I was impressed.

00;04;26;43 - 00;04;46;30
Unknown
You know, of all the possibilities that you can have. Last night, I was building this app that basically, helps, put interactive simulations out. And I thought, like, how cool would it be to have, like, an interactive simulation created where you can have the students, code something and then run the code very quickly to try out if it works or not.

00;04;46;30 - 00;05;05;02
Unknown
And I use Daytona to spin up a quick sandbox in order to have that, you know, interface take place. And I was just impressed by how fast it works and how seamless the experience is. Yeah. Going a little bit in, in, in a timeline here before you did Daytona, I think there's a, there's a history primitive here.

00;05;05;02 - 00;05;22;04
Unknown
You've spent a lot of time in cloud. You've been, I think, one of the pioneers, when it comes to, like, what cloud development workflows can look like. So can you take us a little bit from where you started? Yeah. Yeah. So the the the founding team of Daytona has been together 15 years, probably a bit longer than that.

00;05;22;09 - 00;05;41;37
Unknown
And we're all in front of people. So we come from a different like viewpoint of in this space. And so we basically started first off, we had the services business of basically stacking server rooms. So like way back in the early 2000s. And I know that sounds like a long time for a lot of you, but, we were this is like 25 years ago, then.

00;05;41;37 - 00;06;01;40
Unknown
Jesus. Sorry. I don't know if I'd say that, but, like, it's been it's been a minute. It's been a minute, 22 years ago. So we were actually, you know, stacking servers, plugging in, you know, HPE, IBM servers doing virtualization, like VMware hypervisor, Hyper-V and whatnot. Cisco routers and switches, like we're doing the old, like sort of old thing.

00;06;01;44 - 00;06;21;46
Unknown
And that's how we started. After doing that for a few years, we ended up deciding to sell that services business. And the reason why is my co-founder of Daytona, was the co-founder also called Anywhere Working with me? He came up with the idea of, quote, anywhere, code anywhere, for those that do not know, is probably the very first browser based IDE or code editor in the world.

00;06;21;46 - 00;06;38;33
Unknown
So we publicly launched it. I think it was end of 2008, 2009. So it was like a very, very, very long time ago and it was ahead of its time. There was some success to that. We a few million people had signed up on that. We'd raised a bit of money. At some point in time we end up paying back our investors.

00;06;38;38 - 00;06;55;25
Unknown
It was early, but the things that we did learn is if you think about what is what are the pieces of a browser based ID, you have the ID in a browser. At that point in time there was no VS code, so we had to build it ourselves to build a whole, you know, text editor, coding experience yourself.

00;06;55;39 - 00;07;13;35
Unknown
Not trivial then to be able to run, you know, clone a repository. To run this code, you actually have to have sandboxes or dev environments or dev boxes, we call them, I think we call them dev boxes. Back then, whatever the name was, there was no Docker or doc was the beginning. So you have to like figure out the isolation layer for that.

00;07;13;49 - 00;07;34;14
Unknown
And the hardest part is there's also no Kubernetes. So you had to have the orchestration. You had to create yourself. So we had basically made most of this stuff ourselves. And even though that was really that knowledge of how to build, that is what we apply to in Daytona when we had built it. And especially this, there's a V1 of Daytona, which is very similar to code anywhere.

00;07;34;14 - 00;07;53;22
Unknown
It was an on prem automation tool for large enterprises that we did for about a year and a half. But if you think about the sandbox product, which we're talking about right now, when we started building it, we weren't thinking of it as code execution boxes, because a lot of the vendors for this type of solution today are basically very fast, ephemeral environments.

00;07;53;27 - 00;08;11;48
Unknown
But when we thought about it's like if Asians are going to do human tasks, they're going to want to run it for a long time, because imagine if your laptop dies after 15 minutes or half an hour, or you've just like managed state. Outside of that, it's like, what the hell is? When we think about it's like, okay, send agents needed to be super fast, but they also wanted to be stateful.

00;08;11;48 - 00;08;30;12
Unknown
So think of like you're closing your laptop lid, opening it. It should be the same state. And also you want it to run until you're done. Like you don't want it to die. And so there was no primitive on the cloud vendors that can do those three things at the same time, like all three. And when we started building our Daytona, we're like, oh, we can't use Kubernetes for this.

00;08;30;12 - 00;08;53;15
Unknown
We can't use Nomad, we can't use anything off the shelf. Because these, technologies were made for app deployment. They weren't made for like a runtime. And so we sort of went back to original like ideas. My CTO mostly, that we used around code anywhere and sort of reuse those principles to make that. It was actually quite shocking because when he showed it to our third co-founder, he was like, dude, what the hell is this?

00;08;53;15 - 00;09;14;15
Unknown
This is like code anywhere 2020, you know, 2016 is like, why are we going back in time? And it became it actually became true that the knowledge that we had, accrued then was very useful for today's world. Yeah. I think that's super interesting because, there was, of course, GitHub coming in with, what was the code spaces.

00;09;14;15 - 00;09;34;29
Unknown
Right. And that was 20, 20. So you started your company 12 years ahead of that. So how is that fighting the market for that? Yeah, it was really, really early. And I mean, there's other competitors and some of them exist today, which is great. So like companies that did similar to us was like Stack Blitz and Replit. There's others that don't exist anymore, but like the two of them exist.

00;09;34;29 - 00;09;53;43
Unknown
And so Stack Blitz is now both. So they sort of pivoted. So they both started a bit after us and we decided to like pull back out of the code anywhere space. At some point because like this didn't make sense. And they started a bit later. So they were able to sort of get a bit more run and they were able to, like, survive a long enough for the AI inflection.

00;09;53;47 - 00;10;11;01
Unknown
And what they did is they took their, which is quite interesting. Stack Blitz has this technology called Web Containers, which lets you run runtimes in your browser, which is really useful, at least from a cost management perspective. And they were able to utilize that and go up the stack to the app layer and create, you know, bolt replit.

00;10;11;01 - 00;10;29;45
Unknown
Similarly, although they have a similar type of runtime, I've never worked there, so I don't know the details, but similar to what they do, what it does, and they use that to create their product, which is replit. And it's great to see the two of them had sort of like went into that direction. There's new new ones like lovable from the new new ones, and they seem to be doing very well.

00;10;29;45 - 00;10;58;36
Unknown
We decided purposefully not to go up the stack and app layer. Our, our sort of vision or our intuition was that not only coding agents will need sandboxes or runtimes or computers, but every single digitized knowledge worker, which is an agent, will need effectively a computer. So we decided like if this is a future, all of these app layer companies or agent app layer companies will need this new primitive.

00;10;58;40 - 00;11;16;09
Unknown
And for us, it seemed like a non, a much bigger opportunity and not so much understood. And I think until recently it was very non consensus that this was the even needed. Right. Yeah. Yeah. And now you see it everywhere I guess Rezero also falls in that space to an extent. To an extent yeah V0 falls in the sense of like the replit.

00;11;16;09 - 00;11;32;35
Unknown
But they also have similar products, or they have a sandbox product that they've launched as well. So yeah. Interesting space. I feel like this is like one a part of your technical career, but then you also have a big part of this, like the community part, right? That also plays a big role into what you do with Daytona.

00;11;32;40 - 00;11;57;38
Unknown
And I'll get into more of that as well. Can you take us to, like, the times of shift to. Oh, yeah, we we jumped off, after there was a point in time when we ran, conferences. And so some of my team, which is here as well, in the room we used to run these very large, conferences across Europe, Eastern Europe specifically, there were about, at that height, about 4500 people per event.

00;11;57;43 - 00;12;17;02
Unknown
And so we have this history of doing, conferences as well as a business. So this was a vendor neutral, like developer conference. You guys can think of it, the ones that are just like the AI engineer conference. But for pre AI, right. It's just for developers. And so we had speakers there from like founders of Netlify or Postman or whatnot at these conferences.

00;12;17;07 - 00;12;37;05
Unknown
And they were actually really successful businesses. They're not, you know, venture backed businesses. They're bootstrapped, very scalable. And we learned a lot in a decade or so running these businesses. And after that business had sold, I said, I'm never going to do conferences again. There is like, that is such a stressful job. Oh, God, it's so exhausting. It's exhausting.

00;12;37;05 - 00;12;56;21
Unknown
And so it is crazy because unlike tech, there's no patches. There's no, you know, there's no there's no latency. Like everything happens that moment like is the speaker there? Does the stage work? Does power work? Does do you have an audience? So I had we had one conference when the stage died. Stage literally died when black. The stage died.

00;12;56;28 - 00;13;16;08
Unknown
We had this amazing speaker, by the way, just getting to history. He was going through his talk and hitting his clicker as if it was working like he was so good. He just finished it. And when his talk finished, the engineering team or the tech team was able to solve it. But it was like it was just like, you know, ovation, applause.

00;13;16;21 - 00;13;36;47
Unknown
Like he didn't flinch. It was amazing. Anyway, I said, we're never gonna do that again. And we decided to do Daytona, especially the V2 that the sandbox we're doing right now. We had a very, detailed go to market ground game, which does a lot of events, like you guys have partnered with us with a bunch of these, like hackathons and meetups and drink ups and all these things, because for us, it's like really easy.

00;13;36;54 - 00;14;04;30
Unknown
We've done these huge conferences, like doing a, you know, 20 person, 200 person meetup is like, is there another conference coming up anytime soon? Yeah. Unfortunately, yes. Yes. So March 9th, we rented out the Chase center, so very small that Chase centers are where the Warriors play. It's on the court. The conference is called compute. The direction of the conference is to bring speakers together that are basically building out the new compute paradigms and infrastructures for this new genetic world.

00;14;04;31 - 00;14;28;11
Unknown
There were so many buzzwords in that. I apologize for listening, but I truly believe, and that's how we're building out Daytona, that the, the, the infrastructure companies that exist today are not solving for what agents will need. And the they're not thinking about the agent as the consumer correctly, I believe. And so we've brought together just a bunch of these great, founders that are solving and are working on this.

00;14;28;11 - 00;14;47;37
Unknown
And so that is the theme behind the conference. And so yeah, it's called compute website compute Daytona audio. And I'd love for people to join on that one. Yeah I think it's going to be exciting. I'll be there. So come and see me. I think circling back to shift real quick. So then shift got acquired by info bit, right?

00;14;47;37 - 00;15;13;19
Unknown
Yeah. Info info. Most people probably don't know. It's a direct competitor to Twilio. Probably one of the biggest. They do. I think it's a bit over 2 billion a year, in run rate. So a fairly large company, privately held, and they're neat. They, unlike Twilio, were always a top down, wholesale resale company. It's a very sales driven organization.

00;15;13;24 - 00;15;53;07
Unknown
And if you look at the competitive landscape of, you know, Twilio and Cinch and these other companies that are sort of in that that bracket, about 50% of their revenue originates bottom up, which is what you and I do. And, you know, rabbit and Daytona, which is like getting that developer mindshare and working upwards. So the reason they had acquired the shift conference was that to be it their reinvent or their signal or their company is what we do and that be the starting point of their new teams we had built out, you know, the Devrel team, the content team, the startup team and everything regarding that to sort of be a more head on

00;15;53;12 - 00;16;16;15
Unknown
equal or tech Twilio dominance in that market. I say, and then did a I mean, did it work out fine for a bit? Yeah. I mean, they're still a growing company. They're doing really well. I was there for shy of three years before moving on to, to start Daytona. So the conference still exists. It's grown since then and it's now on three continents, so it's working out fairly well, I would say.

00;16;16;29 - 00;16;31;01
Unknown
All right. That's good. I think give them some of the marketing techniques that I think you definitely stood out with from your experience. Feels like the for example, the shirts that you gave out ran a code, I think. You see, actually, I should have put one on here. You see people running around all in San Francisco with those.

00;16;31;01 - 00;16;50;28
Unknown
That was definitely a big one. Yeah, the shirts were an interesting one. So when we were doing and these are some things that we had learned, there's more coming. There's more things that we haven't done yet that are coming, that are hopefully useful when we're thinking about like the go to market strategy. But we thought about and why San Francisco specifically, you'll see more of these shirts in San Francisco than anywhere else.

00;16;50;28 - 00;17;07;56
Unknown
I think we've given out 6 or 7000 of them in San Francisco. Yeah. So that's a lot. That's very saturated. Yeah. And the thing with swag, which is quite interesting, and this is how we think about a lot of things I tweeted about this earlier today is that you actually have to care about the person that you are, quote unquote, selling to.

00;17;08;01 - 00;17;23;56
Unknown
So when you think of swag, we did a bunch of research before and like, what is the most interesting piece of swag that people actually like? Tech swag. The one that we kept coming back to, there's others, but the one kept coming back to was New Relic shirt. Yeah. Which is data nerd. So it's a data nerd on the front.

00;17;23;56 - 00;17;40;55
Unknown
Right. And the reason why that one you can still find on eBay and people are selling it like for like 50 bucks or hundred bucks, right? So it's not like throwing in the trash can be largely selling it. They want it. The reason they wanted it is because that shirt was about the the wearer of the shirt. It was not about New Relic.

00;17;40;57 - 00;18;03;40
Unknown
It's about the person wearing it saying, I am that. Yeah, data nerd. Like, I am proud to be that thing right? And that's why it was like super successful. And so when we were thinking about our sort of slogan and we wanted to do when you think about what Daytona did originally, or sorry for the sandbox, the first was like run AI generated code securely at scale is sort of like a longer definition.

00;18;03;40 - 00;18;21;00
Unknown
We do more than that. But at that point in time, and so we dumbed it down or numbed it down to just like run AI code. And the reason it worked really well is that people interpreted in different ways, like people that train models like run AI code, execute. I don't run AI code. People that, you know, use cursor run AI code.

00;18;21;05 - 00;18;39;47
Unknown
And then everyone felt that they were very affiliate to that message and it didn't say Daytona, it said it on the back somewhere. But it's like running AI code. We got so much positive messages from people like people got jobs because of it. They saw someone put put in our slack. I was wearing it on the Bart and got interviewed by CEO and got a job because no way.

00;18;39;49 - 00;18;55;40
Unknown
Yeah, sure. Sorry. That's right. And then have people like, oh, can you please ship me shirts had like four of them. And so people actually want the shirt. And if the shirt or the swag was just at Daytona, people were like, okay, I'll wear it when I sleep or go to the gym. Yeah. I'll be like, wear it with pride because it talks about them.

00;18;55;40 - 00;19;10;19
Unknown
It doesn't talk about us. Right. And that's how I think about these things. Yeah. One of my old, colleagues at a previous company, he picked one up and he was wearing it with pride every day. So it's funny, you see people and I say it's an insane thing. Yeah, yeah, yeah. In Silicon Valley that works very well.

00;19;10;24 - 00;19;32;18
Unknown
Let's move a little bit into the, the product as well. Again, can you tell us a bit more about, like what what kind of problem specifically does Daytona solve, especially in the AI world that products like code up, code space, a GitHub code space, and then dev pod and Gitpod and so on don't really address I can talk about use cases and get why they're they're very why we solve those.

00;19;32;32 - 00;19;51;35
Unknown
We break down our use cases into three way we look at it. And so the first use case is the code and command execution, which is the canonical example is like a replit or a lovable or a mom or a bolt, like none of those are our customers, but in the sense of like an agent actually needs to clone a GitHub repository, you know, run the code and things like that.

00;19;51;40 - 00;20;14;17
Unknown
But outside of that, you have things that are code and execution that are not coding agents. So if you think of like we have quite a few AI scientists, so think like drug discovery, type things. And so that will run code. You can think of more engineering companies think like architectural type, agents that have to like, understand layouts of, you know, floor plans or whatever.

00;20;14;17 - 00;20;32;07
Unknown
And then they will run code for that as well. So what can you do as a infrastructure provider for that? Right. Yeah. We will run through all of those what we do. But just to go through them, the second one is the way a human would interact with the machine. And so computer use and browser use are an example.

00;20;32;07 - 00;20;57;38
Unknown
So does your agent need to interact with a, you know, windows desktop. Does it need to interact with the browser. Like for example, let's say you're a QA agent for Booking.com. Like does the website work as you want to work? Or do you need to have your agent work in specifically Microsoft Excel or Microsoft Excel with some certain plug in, for example, because it's only in Microsoft Excel, there's no other way or agent to get to that other than spinning up windows machine to do that.

00;20;57;41 - 00;21;19;11
Unknown
Right. And so can you do that as well. And then the third, which is quite new, but it's growing insanely fast for us is RL environment infrastructure, not to be confused with setting up the RL itself, but once you have the RL environment you have to run training runs against them. Yeah. And so can you spin up, you know, 100,050 thousand 500,000 concurrently and keep that running to get there.

00;21;19;11 - 00;21;37;28
Unknown
And so those are the basic three. Use cases that people come to us for Daytona. Yeah. And when you think about these, the interfaces with which an agent can interact with like a code spaces is very, very different. The speed is very different of the time it takes to speed up a code spaces like three minutes, like, right.

00;21;37;28 - 00;22;06;14
Unknown
And we spin it up in 60 milliseconds. Yeah, that's orders of magnitude difference. And also the concurrency. Can I spin up 100,050 thousand? Whatever. You can't get that. At that scale, you can get that on a lambda, but you don't have the state, and you don't have it to be able to run forever. And the other thing that we think about on Daytona, specifically, I'm sorry, from jumping, is when we think about the sandbox, as we call it, as a primitive, we think about it through two axes, and one is the isolation layer and the other is the tooling layer.

00;22;06;23 - 00;22;23;07
Unknown
And so the isolation layer is like, can it spin up very fast? Can you run it at scale? Can it expand? Can you add a CPU. Can you add different operating systems like it's the computer? Yeah, part of it, the other part is the tooling part. And so when you buy a laptop you get tools in there even if you don't think about it.

00;22;23;07 - 00;22;39;11
Unknown
So you have like a file explorer in a terminal like these different tools. And so Daytona also comes prepackaged with a bunch of tools specifically for the agent. Now they can be things that make the agent more productive. Think like we have a headless terminal. So you can think of it like a terminal, but it's an actual visual terminal.

00;22;39;11 - 00;22;56;18
Unknown
It's API endpoints. And you just send commands and you get all the context. But also the things that are limiting is like, oh, can you block, you know, can you disable what your agent can access to the internet. So we have like a built in firewall. Yeah. And so we're adding also a secrets manager. Sorry in there as well to make sure that your agent can't actually access your credentials.

00;22;56;23 - 00;23;14;07
Unknown
And so those are the things from a tooling perspective. They're not in any of the the technologies that exist today as well. Right. So walk me through what what actually happens when I execute. For example, like yesterday I was executing Daytona dot create for a bracket. What what happens there where there's a few things and how like how Daytona actually works.

00;23;14;07 - 00;23;31;59
Unknown
And so one to start you have to define a snapshot and do a snapshot. You can use the default one, but you can define your own. And to just explain how that works is if you define in our snapshot system generally works, if you define your own new new snapshot and by the way, a snapshot, you can connect your, you know, Docker Hub or your artifactory in your company.

00;23;32;01 - 00;24;00;37
Unknown
Just pick off of that. What happens? Daytona will take that and it will propagate it across our cloud. So basically when you think of this is also quite unique for the Daytona. The sandbox itself uses CPU, Ram and disk of the underlying node, so it doesn't use external network drives. It uses the actual drive. Why that's important. One is that the state of the file system of the machine, you as a developer don't have to think.

00;24;00;39 - 00;24;15;15
Unknown
You don't have to think about, oh, is the file that I created now going to be available the next time I turn it on? Is the dependency that I install is going to be available. It's all there is. Literally the whole machine that's like one part of it. The other part is that when it turns on, it turns it up from a snapshot.

00;24;15;15 - 00;24;33;40
Unknown
Right? It's not for it's not an empty machine. And the way that works is as soon as you define a snapshot, we propagated across all the nodes, there's an I'll go to that. But you can imagine it goes over all the nodes. And when you think of a like a server that exists, your snapshot is literally on the SSD disk of a server, right.

00;24;33;41 - 00;24;51;54
Unknown
And then when you hit create it literally that the agent or Mars system will take you to the machine that has your snapshot, and then it'll basically just spin it up from there. And that's why it's super fast. There's no network transfer of any sort. And to be able to do that, it's also very complicated because you can't have everyone snapshot on everyone's on every server.

00;24;51;54 - 00;25;05;50
Unknown
That's not possible. So then we have like this algorithm that figures out which ones are used more, which is use less. We also have like a diff layering of the snapshot as well to compress them. There's a bunch of things that are actually done in the background, so that when you say they are going to create, it's about 60 milliseconds, right?

00;25;05;50 - 00;25;23;31
Unknown
So there's a lot of things happening in the background there. I can only imagine. I think there's another thing that you briefly touched on, as well, which that's interesting kind of dates back, I guess a little bit earlier to the days where you're defining the mission of the company, which was this primitive of like having standardized development environments versus kind of like these cloud development environments.

00;25;23;31 - 00;25;41;57
Unknown
So, what's the difference there and how, I mean, we've changed our mission since that. That was like the first company then, which was like we were trying to create a standard around dev environments for humans. Yeah. As it turns out, now we're creating that for agents. And basically the mission of the company now is basically to enable an agent to do the work.

00;25;41;57 - 00;26;00;27
Unknown
Like can we enable agents to do work is essentially the mission of the company. And so how we do that is basically when we think of our product features and our future product roadmap, which we already have are like product two and product three. It is all about thinking it through the quote unquote eyes of the agent. Right.

00;26;00;42 - 00;26;18;24
Unknown
What does an agent need to get his task, its task done? And I think that's quite different from most vendors where most people are of thinking about, I am a human and I need this to get done versus when I tell an agent what to do, how does it go off and do that, and can I make that?

00;26;18;24 - 00;26;37;44
Unknown
Can we, as Daytona make that easier for an agent? So our mission is to make the agent's life easier, essentially. Right. I feel like that's almost everyone's mission now who works in software. Well. So the thing is that what you're trying to do, the most people's abilities in software is to create an agent that makes a human's life easier, right?

00;26;37;53 - 00;26;55;02
Unknown
That's what ML software does. The distinction is we don't care about that. I mean, someone else cares about that. Yeah, we care about when you tell your agent what to do, you want that the result to be really good. But what I want to make sure is that when the agent goes off to do its task, that their life is better.

00;26;55;02 - 00;27;13;18
Unknown
So I don't care. We don't care. This seems like a very bad soundbite. Don't pull this out. But what we think about is once you've given the agent the prompt, can we make it easier, faster, and cheaper and said in the sense of tokens for your agent to get the successful result? Yeah. And more secure as well. Oh, secure.

00;27;13;18 - 00;27;32;53
Unknown
Okay. Absolutely. Yeah. I think that's the other primitive. Let's talk a little bit more about the new architecture that you just launched. I think that one of V2 introduced this drop as runner architecture that we talked about. Why was that such a pivotal change? And what does that unlock for you? I mean, there's a few things that we've, unlocked in in the way that we worked on.

00;27;32;53 - 00;27;47;40
Unknown
I think we touched a bit about it in the way we distribute these things across our runners. Or basically, you can think of them as bare metal machines. We call them runners. You can plug in with a drone. You can now plug in your own runners as well. To be very clear. Also Daytona has like a five ways to consume it.

00;27;47;40 - 00;28;03;21
Unknown
Like there's our cloud, multi-tenant cloud. There's a single tenant cloud. You can bring your own runners, you can run it on prem, you can run it on cloud. So there's a bunch of ways to do that as well. But another thing to touch on, which I believe is there is also like the declarative builder, which is which I file under tools for agents.

00;28;03;26 - 00;28;22;15
Unknown
And so when you think about the snapshot or the default image that you use to create a sandbox, you basically define it as a human before time. And that's great. But what happens if your agent now says, okay, I now want to install the I'm missing this package. It's fine post create. The agent can just like install the package.

00;28;22;15 - 00;28;36;53
Unknown
That's fine or two packages, but if it has to do it for every single next run and it might just do like 100,000 runs, that's going to be kind of a pain in the ass. Like it just time doesn't make sense. Like if or if it's more like it's to spin up the sandbox in 60 milliseconds, then it's going to wait, install it every time.

00;28;36;53 - 00;28;54;01
Unknown
That makes no sense. Or it's gonna have to, like, wait for a human to come back and, like, create a new Docker image and upload and whatnot. And so what we have is this thing which is declarative builder. You can think of it like a CI runner, basically where the you can or the agent can define declaratively. I need to take this old image and add these dependencies.

00;28;54;01 - 00;29;13;35
Unknown
Add these in vars, add these, you know, things here. The system will build that on the fly. Deploy the sandboxes across the nodes and the agent can instantaneously spin them up again. And so that's been very very very useful especially for people doing things such as RL. It seems to be quite useful for them, especially switching gears again a little bit.

00;29;13;40 - 00;29;33;32
Unknown
Daytona's also open source. So I think that's especially he had code, but we, we, we take, open source very serious. And we always find it's so great to, to see other companies who are supporting open source, either by making their software open source or helping the community. Can you walk us a little bit through the business model, how that works, and why did you pick that choice?

00;29;33;32 - 00;29;53;35
Unknown
Yeah, to be honest, we did it much better in Daytona. We won. There was no time to do it individually. Daytona redo and Daytona, we won was essentially an open core, but you didn't see things outside of the core. So it was basically a very clean, Apache licensed open source project, the original one. And so anyone could take it.

00;29;53;35 - 00;30;08;54
Unknown
Anyone can do it, anyone could fork it, you could do whatever you want. There was no real no limitations with Apache license, right. As a contributor, there's like overhead there, which you ideally don't want to have there. And that I think that was really great. And we got a lot of love from the community to do that. And that was our old product.

00;30;09;04 - 00;30;27;52
Unknown
When we did their new product, what happened was it was a very hard shift. So basically New Year's Eve last year, so 13 months ago we decided, oh, we're pivoting. We can we're not going to service our credit customers anymore. We're going to help them go to competitors, but we have to move on. And so but we have to launch it really fast.

00;30;27;52 - 00;30;44;46
Unknown
And so we launched in three months, after that. And so it's like, okay, we have to move really, really fast. And so there was no the time pressure to deploy that was very strong. And so we're like, okay, how do we solve this problem now? And so for the Jona now we decided to we do have no separations of repository.

00;30;44;46 - 00;31;07;10
Unknown
It's all in one repository. The whole thing is open source. It's under an AGL license. What does that mean for the audience? For the audience, it means that the other hyper basically you can't make a competitor, you can't make a cloud product, but you can run it for free in your org. So what that means is even if you're I'm going to make this up, Goldman Sachs, you can take this on and run it for free, no questions asked.

00;31;07;15 - 00;31;29;01
Unknown
If you're Amazon you can't compete. You can't like spin it up and run against this. And so that is the difference there. So there are some limitations. But for the vast majority of users it doesn't affect them at all. Like they can literally use run in production right. Yeah. Internally. So that is the slight diff. And we put everything in there because it's a GPL like license the way we could only do it.

00;31;29;01 - 00;31;46;46
Unknown
There was no time for separation. So you have everything there. So anyone using Daytona can now go to the open source repository. If they find a bug in using it, they can literally create a PR right or at issue or whatnot. The other interesting side effect of having the whole repository open source, which I did not comprehend, is that you?

00;31;46;52 - 00;32;04;57
Unknown
It helps your model integrate a much better because you can give it the whole repository. Right. And I've had people reach out to me, oh, it's just like I just give it the whole repository and I can just go off and do because it understands the entire product itself. So we are again, a GPL three, repository, but everything is in there.

00;32;04;57 - 00;32;32;51
Unknown
So you can take a look at everything we do. It's absolutely in there. That's public. Of course that's not. Yeah. Yeah. I mean, you just had 50,000 stars, I think. Congratulations again. Not terrible not to. Oh, it's. That's amazing. I'm very happy to see that. I think it really pays tribute to all the success that you've been doing in community building and like the, the culture that you also build around the company for other people who are doing open source, which is obviously it gives you a lot of roadblocks in terms of scalability and so on.

00;32;33;02 - 00;32;48;07
Unknown
How do you build a business around that? I mean, it's really hard. Like they just GPL was about the business. So we are like, it's not part of a foundation. It is the way we think of it. One, it gives people, I think, peace of mind from a security perspective. It's like, here's the source code, take a look at it.

00;32;48;07 - 00;33;05;42
Unknown
You know what we're doing, which is great too. It helps people also understand like not just agents but humans can understand. And that's like easier for them to like literally raise tickets. It's like, oh, this is like we also have people that like, have, contributed bug fixes, like they saw a bug in the dashboard and just like, you know, yeah, here's a fix.

00;33;05;42 - 00;33;26;14
Unknown
I'm like, awesome. Thank you guys. Yeah. On the other hand, it is a vendor project. It is our project. And so we the way we build and decide on features is from what the market asks for us. And so there obviously there's trade offs to that. Luckily the speed that we execute at is very high. The velocity of features is very high.

00;33;26;14 - 00;33;56;39
Unknown
And I assume we haven't gotten any pushback from the community because it's just like, oh, here's another feature that we essentially need. There may be a point in time where, like the community has different thoughts, and I assume it's at some point something else might show up in the market, I don't know. But from what we see here and do from a business perspective, we have to be able to serve our customers, which means that we have to like lead the, the, the, the open source project in the direction that the company sees fit.

00;33;56;41 - 00;34;25;34
Unknown
When you talk about that, some people pick up bugs, like how many contributors do you have right now and how many people, are there maintainers outside of Daytona, how the companies are? So for now, there's about 200 ish contributors like that, and there's not core maintainers, in Daytona itself. That being said, we're working on a new project which is not part of Core Daytona, and we're putting together a team of like core maintainers, very open source, very community driven.

00;34;25;39 - 00;34;40;45
Unknown
And I believe it's something that will solve the problem, but it will also solve our competitors problem. But it will be open source, and that will be a very much, you know, more a community led, because it's for the benefit community, but that will be in a separate repository 100% and people can take it from there. So yeah.

00;34;40;59 - 00;34;55;46
Unknown
Okay. What would you recommend to somebody who wants to start a business out of their open source project? If it's an open source project that they're turning into a business, or they're have a business that is an open source project. And so one of our investors actually told me when we're going to do open source, it's like, you're an idiot.

00;34;55;51 - 00;35;18;19
Unknown
That was that was a comment. It is like, and the way he said it was to have a successful open source project and a successful business, lightning has to strike twice. Yeah. And it's really hard. Like it is really hard to have the two of them. And so we're lucky that the AI wave is happening. And so just everything is interesting to everyone.

00;35;18;19 - 00;35;38;31
Unknown
So the open source project as it is, is interesting. And the product is as is not to downplay the community, not to downplay the product itself, but you're riding this wave which is making it just easier directionally to do these things. And so you do an open source project. What we've seen successfully is one and do it successfully as a business is a complete other.

00;35;38;43 - 00;35;51;33
Unknown
And the thing that I think that a lot of founders especially want to spend a lot of time because to be honest, we didn't know a lot about open source or in the sense that we consumed open source. We're like, oh, this license, I can use this, I can't, but we weren't part of the community building open source project.

00;35;51;33 - 00;36;10;22
Unknown
So I spent a lot of time just meeting founders in the open source community and hanging out and reading and talking to lawyers and trying to figure this stuff out. It took a while, and what I understood is that most people actually don't understand, like the vast majority, people don't understand what they are signing off to and what they can or cannot do with that.

00;36;10;22 - 00;36;25;47
Unknown
And just knowing that will teach you a lot of how to decide how to build a business. And I'm not trying to be ambiguous about the answer is I truly feel the more founders I've talked to, the more actually don't understand. And they're like, oh, I picked this license because everyone is. I'm like, how are we gonna make money?

00;36;25;48 - 00;36;46;15
Unknown
Oh, I'm gonna make a hosted version of this. Okay. Yeah. Me? Yeah, maybe. Right. It depends on the product. The customer, the the scope, the whatever. And so you have to think about those things really well on just being a successful. I'll take a step back. Just being on a successful open source project, there's like a few baseline things.

00;36;46;27 - 00;37;16;06
Unknown
Make sure your repository just has like, you know, the baseline a good read me, you know, a contributors file the, the the licensing file and all those things. You really take your time to do that and that will go a long way for that. That's obviously just to start. The second thing is the community. You should definitely be part of the community that you're building upon because you there's like a bunch of open there's most of the projects on GitHub, you know, they have sub hundred stars.

00;37;16;08 - 00;37;31;36
Unknown
Like there's not that many. Yeah. Start like they can be their own thing. Let's just take a proxy for for some sort of success. Like you really want to be part of that community solving that problem and an active. And then when you show your solution, it will either get picked up or not because it was picked up.

00;37;31;36 - 00;37;50;34
Unknown
You know, you're in a right direction. Keep going. If it's not, then probably probably not. So and then basically after that you have to have like great guardrails into what you will accept, what you will not accept, what are like what are the steps to contribute to this thing and what we do. And so there's a lot in that on its own to run.

00;37;50;34 - 00;38;13;18
Unknown
And then the second part of that is like, okay, that is great. I want to sell this now to XYZ. Is it like cloud, you know, sign up with your credit card or is it like an enterprise on prem deployment? Are you selling services? Are you selling like there's a bunch of things to think about there, and there's no one easy soundbite for me to say how to solve this, but, a lot of things to think about.

00;38;13;31 - 00;38;30;14
Unknown
I think there was already some, some good bits in there. So one of them was, you know, know your community and be part of it, take good care of your license and, yeah, truly think about the business model. I think that's, to take away, I mean, people a lot of people don't like, don't want to think about money and business model, and that's fair.

00;38;30;19 - 00;38;46;17
Unknown
But like, in the end of the day, even if you're just going to be a project that works like off services, like you have to eat, like you literally have to eat, and if you can't eat, you can't support the, the project itself. So you want to think of that to some extent, so that that project can survive, if nothing else, right?

00;38;46;17 - 00;39;04;19
Unknown
Yeah. Great. Well, thank you so much for sharing a lot of the insights. I like to always close off, the podcast with a couple of rapid fire questions. So, we, let's start with one. What's one open source myth that you would like to kill? One open source myth that I would like to kill. Wow. This rapid fire.

00;39;04;26 - 00;39;23;55
Unknown
Yeah. Not all open source is actually open source. And people should be very wary about that. Just having repos, just having your code out there is not actually open source. And the community will not like that if you call yourself open source. So it might not be a myth, but it's something that people don't. But the average person in the space, I don't think, internalize quite well.

00;39;24;00 - 00;39;39;49
Unknown
Okay. So what what is then the difference between having my wrapper open and truly being open source? So like, you can have your repo or have proprietary code and be completely open and you get a lot of benefits for that. So you can like from the security perspective from like, your agent having context from getting PR is all that's there.

00;39;40;00 - 00;39;57;19
Unknown
But it's not open source. It is like it's source available or whatever the word is, but it's not actually open source. And so if you want a community, you definitely want it to be like, I think GPL is even like pushing it. I made it. So you want to be MIT or Apache if you truly want to have like a an actual awesome community around it.

00;39;57;27 - 00;40;18;44
Unknown
Okay, okay, I see, switching gears again, what's your favorite programing language praise of programmers? I mean, I second most of them, so I'll just I'll just pick TypeScript. I'll just stick with TypeScript. TypeScript. Okay. Yeah. Favorite coding ID? Oh, so, like, I actually don't use them at all anymore. At all. So it is all cloud code, everything.

00;40;18;44 - 00;40;32;27
Unknown
It's just like almost everything. Cloud code. I like look at it. I was like, oh yeah, let me see. That's it. So like not I cursor is still there but not turn it on anymore. Fun fact that colleague was using VS code the other day and like that felt. So I'm like what is this? What do you do?

00;40;32;29 - 00;40;54;07
Unknown
What are you doing? Oh great. That's great. Actually, now you remind me. I wanted to ask you. So we talked about this earlier. You famously mentioned at some point that it is a little bit of a bottleneck. So close. Yeah. Can you explain that? So, I mean, I was thinking, you know, the thing that we do at Daytona really, really well, and what we think about is like, we have a lot of tools headless.

00;40;54;07 - 00;41;09;39
Unknown
So just like API endpoints for the agent. So even if you think of like the way, like we have obviously for file system operations, you have forget you everybody have LSP have all these things. But like I say terminal is used quite a bit. And so it's essentially what you, you have like an API endpoint to like open a terminal.

00;41;09;39 - 00;41;36;27
Unknown
Yeah. And then you fire off commands through like through our SDK or API and it has context. It is essentially like having an open terminal window. And you can open up multiple of these. And so the agent does have to think about anything around the actual tools and or terminal. It's just like an API endpoint. So it uses much less context, much less need for the agent, and can just like fire off and do the tasks without thinking about the tasks that it's doing.

00;41;36;32 - 00;42;04;30
Unknown
And so CLI tools are great and like cloud code, you have them running in Daytona and all these things. But what I feel from a token consumption speed of execution is if you the more things you can remove for the agent to work on or think, the higher the success rate. That's right. I mean, I had, interestingly enough, last week the creator of, fast MXGp, one of the very popular open source frameworks for, for agents to use tools in there.

00;42;04;41 - 00;42;21;03
Unknown
What's your opinion then on, on the Mic protocol? Because that really just adds an extra layer to the API, does add an extra layer. And then you have I don't know if you saw the article that the team at Cloudflare created, which essentially says, oh, spin up another sandbox to use apps, which is it works very good for your business.

00;42;21;03 - 00;42;39;29
Unknown
So to see the article. So basically the article says and the example there is so zip a lot of context. Right. And so let's take an example of an MTP with for slack. And you can say oh search you know for you or search for the, the devrel team in code rabbit. And it will actually like go search all the people in slack.

00;42;39;29 - 00;43;11;22
Unknown
Right. And it'll, it'll, it'll fill up the context with the entire list of people to get to your name. Right. And so it fills up the context really fast. And so the way I think it's called, it's called code MCP. I think the article, I have to check, I forgot and so it's basically, oh, spin up a sandbox, then run code against the MCP to get your name from it so that that whole list doesn't fill up the context window works better great for our business, but if you think about it, it is an API that's wrapped in MCP that then has a sandbox on top of it.

00;43;11;22 - 00;43;30;01
Unknown
So it's like I feel that there's a lot of layers there that probably overhead or a lot of overhead that shouldn't be there. So I think there's there's work to be done still on on the on the protocol. Yeah. Next. Next file. Fire. Fire. Question. I'm I'm getting out of hand here. What is, what do you think is one of the most overhyped dev tools right now?

00;43;30;15 - 00;43;51;52
Unknown
Overhyped dev tools. Oh, I open CLA for sure. Yeah, I think that's super. Oh, right. I think they're great. Don't get me wrong, I don't think. Yeah, not bad at all. Not at all. I mean, you can spin it up on your Mac. You even got this little quick start guide. Yeah. So I just think that I think the thing in the market right now is that if something is great, it gets like super hyped out.

00;43;51;54 - 00;44;10;35
Unknown
And so I think they're I don't think that they don't deserve or he doesn't deserve the hype very much. If you're watching this great work. Yeah. I just think that the the way the market works right now is like, if something's great, then everyone gears to worried and then tomorrow something else happens and just like they move on.

00;44;10;40 - 00;44;32;20
Unknown
And so I just think that the level of hype right now is just like in it's way more than I think it deserves. Not that it doesn't deserve hype. Yeah. No. Great. Great answer. Next one Mac, Linux or Windows? Just like I'm just a mac guy. I'm such a simple I spend almost a lot of my life like servicing, debugging windows and knew everything about it.

00;44;32;20 - 00;44;52;42
Unknown
Like on my Mac, I know nothing, it just works. I love it. We talked about this already. I like this question. Favorite license? MIT Apache, or so I again like, MIT super. You know, you can do whatever you want. I feel that Apache is probably the best in between that. And so, like, if you're doing anything, it's very, you know, permissive and you can do everything, but it does give you recognition that you built it.

00;44;52;42 - 00;45;12;32
Unknown
And I think that's the fair that the one would I would always sort of choosing to go for you. Okay. Well, we'll end on that note. Thank you so much for taking the time, Evan. It's been great to have you on the show. Congratulations again to you and your entire team who has worked so hard over the last months that I've known you on getting the next series on making your customers happy.

00;45;12;32 - 00;45;26;19
Unknown
And, to everyone who's listening to this, check out Daytona. It's super easy to set up. Make sure you. Yeah, agents have the best tools at hand to run their code in the sandbox and see you at the next episode. Thank you. Thanks so much. Thanks for having me. Appreciate.