ТРАНСКРИПЦИЯ 🔗
WEBVTT
00:00:00.000 --> 00:00:08.080
If you're not doing a happy hour for the happy hour's sake, it's like, can I create something that the people that are there will actually get a benefit of.
00:00:14.960 --> 00:00:20.640
Welcome to InProd, conversations with founders and operators who are building the next wave of AI companies.
00:00:20.800 --> 00:00:25.679
We go deep on product execution, go-to-market, and the messy middle of building a business.
00:00:25.920 --> 00:00:33.039
If you're an aspiring founder, operator, or just startup curious, this is your inside look at how companies actually get built.
00:00:35.840 --> 00:00:40.960
Before we get to the episode, Laura and I are excited to announce a partnership with Overlap.
00:00:41.280 --> 00:00:50.880
Fresh out of Y Combinator's Summer Batch, Overlap is an AI-driven app that uses large language models to curate the best moments from podcast episodes.
00:00:51.119 --> 00:01:03.119
Imagine having a smart assistant who reads through every podcast transcript, finds the best parts or parts most relevant to your search, and strings them together to form a new curated stream of content.
00:01:03.359 --> 00:01:05.439
That is what Overlap does.
00:01:05.680 --> 00:01:13.680
I personally use Overlap to explore content on future podcast guests and highly recommend it as another diligence tool for investors.
00:01:13.920 --> 00:01:16.319
Check out Overlap on the App Store today.
00:01:16.640 --> 00:01:20.480
Today we are joined by Ivan Brizen, co-founder and CEO of Daytona.
00:01:20.719 --> 00:01:26.879
Daytona is building secure, stateful development environments designed for modern software teams and AI workflows.
00:01:27.040 --> 00:01:27.920
Welcome, Ivan.
00:01:28.400 --> 00:01:29.359
Thank you for having me.
00:01:29.680 --> 00:01:31.200
Excited to dive in.
00:01:31.599 --> 00:01:36.959
So you've co-founded multiple companies, Code Anywhere, the Shift Conference, and now Daytona.
00:01:37.120 --> 00:01:41.840
What lessons have you carried over from your earlier ventures that shaped how you're building Daytona?
00:01:43.439 --> 00:01:46.319
How important timing and market is.
00:01:46.879 --> 00:01:48.079
Probably the number one thing.
00:01:48.159 --> 00:01:53.519
And the reason I say that is so we created Code Anywhere in 2009, the very first browser-based ID.
00:01:53.599 --> 00:01:58.959
And so before Replit was Replit, before the older version, CodeSpace is whatever.
00:01:59.200 --> 00:02:03.599
And we did that in 2009, so super early, like way earlier than we should have had done.
00:02:03.680 --> 00:02:06.799
It was like directionally correct, but like timing was not there.
00:02:07.040 --> 00:02:20.560
The conference, we started the conference on the Mediterranean coast in Croatia, and we started it as a startup conference, and it attracted people, but did not attract sponsors because there wasn't there was no market to sponsor a startup conference.
00:02:20.719 --> 00:02:30.080
We ended up changing that into a developer conference, and in that part of the world, all of Eastern Europe has like a bunch of people that are really good engineers and people that want to hire these engineers.
00:02:30.159 --> 00:02:32.240
And so there was a huge market for that, right?
00:02:32.319 --> 00:02:41.520
And then we had expanded the conference to other geos and other themes, but always looking at like, is there a market there to support the product that we are selling?
00:02:41.599 --> 00:02:44.639
So basically, I think timing and market are the things that I learned most.
00:02:45.120 --> 00:02:45.439
Interesting.
00:02:45.520 --> 00:02:48.159
And that's very different than what you're doing now.
00:02:48.400 --> 00:02:53.360
But I'm assuming that when you became chief developer experience officer, which I didn't know was a title.
00:02:53.599 --> 00:02:54.000
Very interesting.
00:02:54.319 --> 00:02:55.280
Yeah, I made that one up.
00:02:55.520 --> 00:02:55.759
Okay.
00:02:56.080 --> 00:02:57.759
I mean, anything goes nowadays, right?
00:02:58.000 --> 00:02:59.280
Chief Vibes Officer.
00:02:59.599 --> 00:03:02.560
This was after Infobip acquired Shift, I guess.
00:03:02.719 --> 00:03:09.759
So is that kind of when you started to tap into the developer's experience and potentially thought about building a developer tool company?
00:03:10.080 --> 00:03:15.039
So Code Anywhere, I feel was like also like developer tool company before developer tool companies were cool.
00:03:15.120 --> 00:03:16.800
So like in the early 2000s.
00:03:16.960 --> 00:03:21.599
And so that's where we learned about like bottom-up PLG developers using something.
00:03:21.680 --> 00:03:29.199
Although it wasn't really called that until GitHub, I think was acquired by Microsoft, like DevTools wasn't a market for VCs.
00:03:29.280 --> 00:03:30.639
Like no one knew that as a market.
00:03:30.800 --> 00:03:32.080
So we're like early on that.
00:03:32.479 --> 00:03:43.039
The conference itself, because it was a developer conference, I feel that that is where you actually did we did create communities because we did have to create a community around the conference for attendees to actually want to join the conference.
00:03:43.120 --> 00:03:54.960
So in the sense of we were selling to large companies that were developer tools like Docker or Microsoft, whatever our sponsors, but you had to also sell to the people attending, which is kind of building out communities.
00:03:55.039 --> 00:03:58.159
And so those two things aligned with InfoBIP.
00:03:58.319 --> 00:04:02.719
For those that don't know what who our InfoBIP is or what they are, they're a direct competitor to Twilio.
00:04:02.960 --> 00:04:07.919
So it's an infrastructure provider for cloud communication, so like SMS voice, email, and whatnot.
00:04:08.080 --> 00:04:13.680
And so, unlike Twilio, they were a top-down company versus like a PLG, which Twilio was.
00:04:13.840 --> 00:04:21.199
And so they had acquired the
Load full transcript
conference and brought me in to build out this entire team to basically create a PLG motion for them.
00:04:21.360 --> 00:04:28.079
And so they're had to build out the entire team and learned how that works inside of a very large organization.
00:04:28.160 --> 00:04:31.120
So they do about 2.4 billion in run rate.
00:04:31.199 --> 00:04:32.720
They're 4,000 people.
00:04:32.800 --> 00:04:36.000
So it's a much, much larger organization than ones that I'd worked for before.
00:04:36.240 --> 00:04:36.560
Yeah.
00:04:36.959 --> 00:04:37.920
That's pretty large.
00:04:38.160 --> 00:04:44.000
And then you were building in Croatia, and now you're building in Croatia and New York and San Francisco, kind of all over.
00:04:44.240 --> 00:04:47.439
How is multi-geo distributed workforce approach?
00:04:47.600 --> 00:04:51.120
How has your mindset shifted to building teams and community?
00:04:51.600 --> 00:04:56.720
What we had done, and so we did this at the beginning of Daytona V1, and we did this at InfobIP as well, which we learned.
00:04:56.879 --> 00:04:59.839
We had built developer communities all over the world.
00:04:59.920 --> 00:05:06.639
Like we've done it in Acra, Ghana, we've done it in Bangalore, India, we've done it like absolute like Berlin, Germany.
00:05:06.720 --> 00:05:07.839
We've done it all over the world.
00:05:08.000 --> 00:05:18.639
And one of our my teammates that has actually worked with me at Shift and then did that at InfobIP, but now here at Daytona, we have created these specifically developer communities all over the world.
00:05:18.720 --> 00:05:26.879
And the thing that you do learn is that they're very, very different in the sense of what makes people tick or what attracts people, right?
00:05:26.959 --> 00:05:34.639
So the same thing that you would say or try or pitch in Accra is not the same thing that flies in San Francisco.
00:05:34.800 --> 00:05:41.759
So, like underlining here the same company, same product, same type of thing, but the way you message it and the way you attract people is different.
00:05:42.000 --> 00:05:47.199
Basically, you have to figure out what gives value to these people.
00:05:47.439 --> 00:05:51.360
So, what is this community missing in that you can actually offer value to them?
00:05:51.439 --> 00:05:57.519
And if you give them value, then they will sort of like grow around you.
00:05:57.839 --> 00:06:04.560
And speaking of value, I feel like at least what I'm seeing from my side, developer environment sandboxing is becoming extremely important right now.
00:06:04.720 --> 00:06:12.639
People want to run their code in an isolated space before deploying, and that's now what you're working on, but it sounds like it's not Diktona V1.
00:06:12.720 --> 00:06:19.040
So maybe just touch on kind of how you found this and what that transition from V1 to assuming this is V2 looked like.
00:06:19.439 --> 00:06:19.839
Yeah, exactly.
00:06:19.920 --> 00:06:21.600
I I mean just use the terms v1, v2.
00:06:22.160 --> 00:06:22.560
It's easier.
00:06:22.720 --> 00:06:22.879
Yeah.
00:06:22.959 --> 00:06:31.439
So the V1 was an extension of the first product, Code Anywhere that we did, kind of, where it was automating dev environments for humans in very, very large enterprises.
00:06:31.600 --> 00:06:34.720
So our first customer was like a Fortune 500 company.
00:06:35.040 --> 00:06:46.399
The idea behind it was if you work at any of the large tech companies, so like Meta or Microsoft or whatever, as an engineer, you come in, you click a button, it's almost magical, just you get up and running.
00:06:46.560 --> 00:06:49.279
If you work at any other large company, it's like a shit show.
00:06:49.360 --> 00:06:52.399
It's like a pain in the ass to set up, a bunch of README's, whatever.
00:06:52.560 --> 00:07:04.800
Plus, if you are in a highly regulated industry, then you're probably using a remote computer, doing all that, and then streaming it through remote desktop, which is just like a completely terrible experience.
00:07:04.959 --> 00:07:12.480
And so what we did with Daytona is we made the developer experience really fast and simple and nice and a great experience, but also highly secure.
00:07:12.639 --> 00:07:21.680
So the environment would be on a remote, quote unquote, secure server, the engineer would use their local ID, it would feel very local, and everything would be really nice for them.
00:07:21.759 --> 00:07:23.920
And so that's how we started that the process.
00:07:24.160 --> 00:07:35.839
And end of last year, or I should say beginning of this year, we decided to basically repurpose our product and instead of serving humans, serving essentially VMs or sandboxes or dev environments for agents.
00:07:36.079 --> 00:07:36.480
Got it.
00:07:36.639 --> 00:07:40.079
And so something that you're focused on is stateful sandboxes.
00:07:40.160 --> 00:07:50.160
Maybe just break down why that's important and how that fits into this wave of AI-assisted coding, agent workflows, five coding, all the buzzwords in the bay right now.
00:07:50.480 --> 00:07:51.279
All the buzzwords.
00:07:51.360 --> 00:07:53.279
So our slogan is run AI code.
00:07:53.439 --> 00:07:57.199
That's where we started, but I really feel that it's much more than that.
00:07:57.360 --> 00:08:00.399
I think of what Taytona does is computers for agents.
00:08:00.639 --> 00:08:06.480
And so we build composable computers for AI agents at creation time.
00:08:06.560 --> 00:08:17.120
So whenever an agent needs a computer to do something, write code, clone a GitHub repository, run a command, computer use, even a browser.
00:08:17.600 --> 00:08:19.279
We power a lot of browser companies as well.
00:08:19.680 --> 00:08:20.959
The agent can define what it needs.
00:08:21.040 --> 00:08:25.519
So in the sense of like how many CPUs does it need, how much RAM does it need, how much disk does it need, does it need a GPU?
00:08:25.680 --> 00:08:31.759
Does it need a GPU in the sense of
not for inference, but in the sense of does it need to render things faster?
00:08:31.920 --> 00:08:34.399
Does it need to do a 3D blender type thing?
00:08:34.799 --> 00:08:36.480
And also what operating system does it have?
00:08:36.799 --> 00:08:42.159
Most right now are just Linux containers or VMs, but it can be does it need a Windows, does it need a Mac or whatnot.
00:08:42.240 --> 00:08:50.879
And so I think about it in the sense of if an agent is a digital knowledge worker, so it is equal to us sitting at a computer, right?
00:08:50.960 --> 00:08:52.000
And we have different computers.
00:08:52.080 --> 00:08:56.879
So sometimes we're working on like a MacBook Air, it's super thin, you just do email and Excel or whatever whatever.
00:08:57.039 --> 00:09:02.960
Or sometimes it's this big, huge computer that you need to do because you're doing AutoCAD or some 3D type of thing, right?
00:09:03.039 --> 00:09:07.679
And so I think that's sort of the future of uh these sort of machines and agents.
00:09:07.840 --> 00:09:25.519
Uh why we need these types of computers and super stateful is because uh right now most of the technology that people use for uh running code or similar things for agents is built on serverless technology, like legacy technology, which exists for a very good reason.
00:09:25.679 --> 00:09:32.399
It is for app deployment and scaling those things, and that's perfect, but they are usually really fast, not stateful, and not long-living.
00:09:32.480 --> 00:09:36.320
And what I mean by stateful, long-living is if you look at, I'll just take a MacBook example again.
00:09:36.480 --> 00:09:39.360
If you close the lid of your Mac and open it, it's the exact same state.
00:09:39.519 --> 00:09:40.720
Like you expect that.
00:09:40.960 --> 00:09:46.320
And if you work on your la computer or laptop, you expect it to run until you turn it off.
00:09:46.480 --> 00:09:48.879
Or hopefully your battery doesn't die, but like you want it to run.
00:09:49.120 --> 00:09:50.879
Other technologies don't have that ability.
00:09:50.960 --> 00:09:52.960
They will like kill your machine at a certain point of time.
00:09:53.120 --> 00:09:54.240
They don't save the state.
00:09:54.320 --> 00:10:04.879
And so for us, thinking about what the future of agents looks like and what they're doing, it seems that they should have equal rights to a computer that works in the same fashion.
00:10:05.120 --> 00:10:05.360
Definitely.
00:10:05.440 --> 00:10:08.399
I mean, I think we're moving to a world where we'll be managers of agents.
00:10:08.559 --> 00:10:08.879
Exactly.
00:10:09.039 --> 00:10:10.480
And yeah, they will be our employees.
00:10:10.559 --> 00:10:14.000
And so and imagine what you can do without a computer, not very much today.
00:10:14.159 --> 00:10:14.480
So, yeah.
00:10:14.720 --> 00:10:15.120
Exactly.
00:10:15.440 --> 00:10:23.919
How it looks today in practice, maybe just give a few examples of use cases and how you kind of balance this because I know you're bottoms up, but you're also selling to enterprises.
00:10:24.000 --> 00:10:29.919
So, how you balance nailing the developer experience while also meeting the enterprise's security needs.
00:10:30.159 --> 00:10:33.519
Yeah, so we're completely focused on bottom-up and PLG.
00:10:33.600 --> 00:10:34.960
That's like where our focus is.
00:10:35.200 --> 00:10:37.519
The reason we do that is a few.
00:10:37.759 --> 00:10:43.759
One is the cutting-edge agents are all bottom-up, so they're the new startups that you talk to and whatnot.
00:10:43.840 --> 00:10:50.320
And so we want to be really close to them when they say, Oh, our agent needs the ability to do X or needs this feature or needs this tool or whatever.
00:10:50.480 --> 00:10:52.080
Can we build that really fast for them?
00:10:52.399 --> 00:10:54.000
And so that's like the number one.
00:10:54.240 --> 00:11:00.799
The other is just getting or creating the uh idea of what this net new primitive is.
00:11:00.879 --> 00:11:05.679
And I say net new primitive because something that's long running, fast, and stateful does not exist on any cloud.
00:11:05.759 --> 00:11:07.519
Like those three things together does not exist.
00:11:07.919 --> 00:11:23.840
And so we're trying to market and shape what that looks like and getting that speed of adoption of these companies because enterprises do take a longer time, so the the smaller companies like basically sign up today and they're deploying already tomorrow.
00:11:24.159 --> 00:11:39.120
Now, the the cool thing is that agents are so new that everyone's at the same level of sophistication, more or less, like enterprises and individuals, so that enterprises now actually come in PLG, and then you know, procurement comes into us.
00:11:39.200 --> 00:11:40.879
It's like, hey, our team is using this thing.
00:11:41.039 --> 00:11:41.360
Yeah.
00:11:41.600 --> 00:11:46.480
Now show me all the compliance, security, and you have to run through all these things.
00:11:46.559 --> 00:11:55.200
So, yeah, at the same time, so the way we focus and build a company and the go-to market is all PLG, but we also are very attuned that we have to be compliant.
00:11:55.279 --> 00:11:58.480
So we're like HIPAA and SOC, and we have all these security features.
00:11:58.559 --> 00:12:01.519
We have pen testing every month and all these things that we have to do for the enterprise.
00:12:01.840 --> 00:12:05.759
Are there a few quintessential use cases that you're seeing from enterprises?
00:12:06.080 --> 00:12:12.559
Or have you been maybe more on the you know surprised side of the spectrum when you see how they're running the sandboxes?
00:12:12.879 --> 00:12:18.639
So the first set of enterprises have been basically building out their versions of everything that we see.
00:12:18.720 --> 00:12:26.240
So, like their version of ChatGPT, their version of Cognition's Devin, they're like their version of that, and that was pretty you know, expected.
00:12:26.639
--> 00:12:30.960
But what's interesting, the newest use case that we have, like a few, is like AI scientists.
00:12:31.840 --> 00:12:46.960
So think of like you know, chemical industry, farm industry, those types of they have agents that are actually like quote unquote scientists trying to build or like we don't see what they do is essentially, but from the explanation of what they've explained to us, they're essentially AI scientists.
00:12:47.039 --> 00:12:52.480
And the funny thing about this is that because of what they're trying to compute, the actual sandboxes are huge.
00:12:52.720 --> 00:12:59.360
So the the average sandbox size on Daytona is like uh two CPUs and four gigs of RAM, like like pretty small.
00:12:59.840 --> 00:13:07.039
Even you have quite a like just one VCPU, one gig of RAM, but these are like 128 CPUs or 512 gigs of RAM.
00:13:07.120 --> 00:13:14.000
Like these are these are essentially a whole physical machine that would just wrap into a sandbox, which kind of seems odd.
00:13:14.080 --> 00:13:25.519
Like you can essentially take the machine and do yourself, but the tooling and that that layer of abstraction that we've created makes it so much easier for the agent to do its job that they use us essentially to wrap a full machine.
00:13:26.000 --> 00:13:29.919
And then how are they measuring the success once it is embedded?
00:13:30.320 --> 00:13:33.039
Metrics, ROI, the success of Daytona.
00:13:33.279 --> 00:13:33.519
Yeah.
00:13:33.759 --> 00:13:36.960
The metrics are if you don't have it, it doesn't work.
00:13:37.039 --> 00:13:38.480
There's no other way to do it right now.
00:13:38.639 --> 00:13:46.960
And so what I want to touch on is a lot of companies, especially larger companies that have come to us, have already tried or already have built out their own versions internally.
00:13:47.120 --> 00:13:52.720
So they've built something on Kubernetes or whatever it may be, either completely on-prem or on AWS or whatnot.
00:13:52.879 --> 00:13:56.799
And because that was not built for this, they essentially have scaling issues.
00:13:56.879 --> 00:14:00.320
At one point in time, it becomes so large that it was not meant for that.
00:14:00.480 --> 00:14:03.200
And then you start looking for a provider that can offer that to them.
00:14:03.360 --> 00:14:08.320
And so now we spin up on our own infra like 1.2 million of these sandboxes every single day.
00:14:08.399 --> 00:14:14.639
And so, to your question of your ROI, it's like it was a pain in the ass, and our engineers had to, or DevOps people had to manage it.
00:14:14.879 --> 00:14:16.080
Now it's just like done.
00:14:16.240 --> 00:14:18.080
So that's basically the ROI.
00:14:18.480 --> 00:14:20.960
Now, a quick word from our sponsor.
00:14:21.279 --> 00:14:27.600
Laura and I are thrilled to announce a partnership with Grata, an intuitive deal sourcing and diligence platform.
00:14:27.840 --> 00:14:31.919
Keep missing out on private market deals, ditch the Rolodex and get Grata.
00:14:32.399 --> 00:14:40.559
Grata's AI-powered platform provides accurate data, deep insights, and time-saving workflows that help get more deals done.
00:14:40.799 --> 00:14:46.000
And it's trusted by over 500 firms in 3,000 deal makers, including myself.
00:14:46.320 --> 00:14:55.919
I've personally used Grata in both my deal sourcing and diligence workflows and would recommend their tool to any investor across private equity, growth equity, corpdev, or investment banking.
00:14:56.159 --> 00:14:58.960
We've negotiated a special offer for our audience.
00:14:59.200 --> 00:15:08.480
Mention partner path to the Grata team during sign-up, and you'll get free executive seats for your partners and MDs included in your subscription.
00:15:08.799 --> 00:15:10.320
Now back to the episode.
00:15:10.720 --> 00:15:16.879
Do you think there'll be a future where all AI agents will need their own dev environment?
00:15:17.200 --> 00:15:21.200
I think there's the future that every single agent will need multiple of them for every single task.
00:15:21.279 --> 00:15:24.480
And why I say multiple is because an agent gets a task.
00:15:24.559 --> 00:15:28.080
And so we can talk about an agent, let's just the one everyone knows like ChatGPT is an agent, right?
00:15:28.159 --> 00:15:33.279
And so you ask ChatGPT, literally, when you hit agent mode, it does spin up one of these computers to do that.
00:15:33.360 --> 00:15:37.440
They're not our customer, to be very clear, but like it's similar to what Aiken to what we do.
00:15:37.679 --> 00:15:42.399
And so that is for every single chat with ChatGPT, it spins up one.
00:15:42.559 --> 00:15:58.159
As we continue in the future and agents start being able to utilize more, there's also a feature called forking where the agent can like take the state of the machine, and there's like one or two, three options where it can go to solve, and then it can like fork the machine like three times or five times or ten times and then go through that.
00:15:58.399 --> 00:16:05.600
So as agents become better at utilizing that, I feel that there'll be more than one per actual task.
00:16:05.759 --> 00:16:13.600
So if we think about how many agents will exist in the world, like humans to the power of n, and then for every single one of those tasks, it will spin up one or multiple machines.
00:16:13.679 --> 00:16:15.360
And that number starts getting really stupid.
00:16:15.519 --> 00:16:17.519
Like it's a very stupid number that we start talking about.
00:16:17.759 --> 00:16:25.200
And so as that number gets more and more stupid, how do you balance the cost, the latency, and everything that your customers care about?
00:16:25.519 --> 00:16:32.480
So we've built our own complete like orchestrator and tech stack, everything all the way down to bare metal, which enables us to do a lot of things really fast.
00:16:32.559 --> 00:16:35.120
We spin these things up in with network latency.
00:16:35.279 --> 00:16:42.559
So
agent request create sandbox, the creation of sandbox and get a reply that's created is in sub 90 milliseconds.
00:16:42.639 --> 00:16:45.679
And so you blink or I blink at 110 milliseconds.
00:16:45.759 --> 00:16:46.879
So like that's how fast it is.
00:16:46.960 --> 00:16:48.720
It's so that part is super fast.
00:16:48.960 --> 00:16:52.879
The scale that we have right now, again, we spin up millions a day.
00:16:53.120 --> 00:16:58.080
We're also overutilized, like we use about 15% of our capacity every single day.
00:16:58.159 --> 00:16:58.639
That's purposely.
00:16:58.879 --> 00:17:01.360
Want to make sure that we can have these big spikes, which we can talk about.
00:17:01.440 --> 00:17:03.120
There's some very big spikes sometimes.
00:17:03.360 --> 00:17:22.559
But we had a deployment every single day, like literally every day we did deployments, and we've kind of slowed that down a bit just because the risk of what might happen because of how many companies now are reliant on datoness infrastructure, that part has slightly slowed down, not in the development process, but at least in the release process.
00:17:22.720 --> 00:17:29.039
So we just want to make sure that everything's extra surely tested and make sure that no issues happen when you deploy a new version.
00:17:29.279 --> 00:17:41.039
I did read about that, the fast iteration cycles within on your blog on the website, and it seems like you've cultivated this 996 or generally gritty work ethic and culture.
00:17:41.200 --> 00:17:49.200
Is that something that you set from day one with the internal processes as the Yeah, so we I don't know, it's 996 feels like almost too rigid.
00:17:49.279 --> 00:17:50.319
It's like it's not 996.
00:17:50.480 --> 00:17:51.920
I was thinking about this when we're driving here.
00:17:52.079 --> 00:17:57.119
It's like we're not 996, we're like always on, but there's like a there's a vibe to that.
00:17:57.279 --> 00:18:05.279
Whereas like people obviously during work hours, Monday to Friday, like everyone is completely always on, but even at 2 a.m.
00:18:05.440 --> 00:18:08.640
on a Sunday, there's always someone on, depending on how it is.
00:18:08.720 --> 00:18:16.240
And this is there's no schedules, there's no timelines, there's no, oh, you work from two, like it's just the vibe that we're working at.
00:18:16.480 --> 00:18:19.759
And one thing that I believe is like that is the culture.
00:18:19.839 --> 00:18:21.039
Like I work all the time.
00:18:21.119 --> 00:18:24.400
Like, my everyone in my employees are like, Do you ever sleep?
00:18:24.480 --> 00:18:25.599
Like, that is always a question.
00:18:25.759 --> 00:18:26.559
Like, I always reflect.
00:18:26.880 --> 00:18:27.839
Like with the baby now.
00:18:28.000 --> 00:18:28.319
Yeah, yeah.
00:18:28.480 --> 00:18:32.160
I'm like, I just woke up with the baby and never applying to Slack while holding the baby.
00:18:32.240 --> 00:18:32.799
Yeah, exactly.
00:18:32.960 --> 00:18:46.559
And that I
think sort of feeds into the culture where everyone is just more attentive, but also, and I've had startups that were not successful, when things are going in the right direction, everyone I think just feels that energy, and everyone's more energized to actually wanting to get that done.
00:18:46.720 --> 00:18:49.519
So I think it's not necessarily that we push to that.
00:18:49.599 --> 00:19:00.160
We as the founders are both like that, but also that whole pull from customers is there, and it's just like I feel like people are energized to do that.
00:19:00.240 --> 00:19:02.079
So a customer pings you at like 3 a.m.
00:19:02.240 --> 00:19:10.799
for something, and if you're awake, you actually reply to it because you actually want to help that person because they're better off and we're better off all better off together.
00:19:11.039 --> 00:19:13.039
How do you screen for that in the interview process?
00:19:13.200 --> 00:19:15.440
Or are these all folks that you've worked with before?
00:19:15.680 --> 00:19:16.960
So you just know it's enabled.
00:19:17.359 --> 00:19:17.839
A lot of them were.
00:19:18.000 --> 00:19:22.079
So we're now 18 people of the 18 people, 14 we've worked with before.
00:19:22.319 --> 00:19:32.160
So one thing that I've learned as a founder of multiple companies is that you do have like a network of people that you have like a high throughput of understanding with.
00:19:32.319 --> 00:19:38.160
And so if we work together, let's say we've worked for 10 years, then it's super easy for us to work together again because we already know.
00:19:38.319 --> 00:19:45.440
Like there's no BS in conversations, like you can tell me that you know this is dumb, or I can tell you no, we have to do this thing or whatever.
00:19:45.599 --> 00:19:50.400
But the conversations are really high, high fidelity, high throughput, and and you know how they work.
00:19:50.559 --> 00:19:57.119
And so when we started Daytona V1, we didn't do that, but in V2, we're like, oh shit, we know all these great people, and just like hire them back.
00:19:57.599 --> 00:20:00.400
Now you can't scale that forever, obviously.
00:20:00.559 --> 00:20:06.640
But hopefully, now we've added a few people that we haven't worked with before, but all of them have worked with someone in the team before.
00:20:06.720 --> 00:20:09.839
So now we have that like second order of connections to that.
00:20:10.079 --> 00:20:24.880
And so our hope is as we try to screen for these things, I think that's very hard to screen for, but as we try to screen for this these things, I hope that the company or team is big enough that it will self-select out if someone is not sort of in that mindset.
00:20:25.279 --> 00:20:27.279
What AI tools have you leveraged internally?
00:20:27.359 --> 00:20:35.279
To whether it's like speed up hiring or speed up productivity or just general throughput to help your team perform better besides cursor.
00:20:35.839 --> 00:20:36.880
Besides cursor, yeah.
00:20:37.039 --> 00:20:38.000
I was gonna say cursor.
00:20:38.079 --> 00:20:43.839
I mean everyone, depending on what part of the org they are, they use different tools to speed things up.
00:20:44.079 --> 00:20:52.720
We also have our own version, our CTO has his own version of a conductor where it's like cloud codes and multiple data owners doing a lot of things at the same time.
00:20:52.880 --> 00:20:57.599
And so different tools for different people, people on the go-to market side use other tools.
00:20:57.680 --> 00:21:01.519
Like we have adios, like the AI CRM type thing that we use as well.
00:21:01.680 --> 00:21:04.160
So different tools for different use cases.
00:21:04.640 --> 00:21:05.039
Interesting.
00:21:05.200 --> 00:21:11.759
And then from a user point of view, are there any patterns or pitfalls that you typically see folks getting into when they onboard a tona?
00:21:12.160 --> 00:21:19.039
And how have you gotten ahead of those just given the pattern recognition since you've been building this for what over a year now?
00:21:19.440 --> 00:21:19.920
Six months.
00:21:20.160 --> 00:21:20.720
Six months?
00:21:20.880 --> 00:21:21.359
Six months.
00:21:21.680 --> 00:21:22.480
Feels like longer.
00:21:22.720 --> 00:21:24.000
No, we started no longer.
00:21:24.160 --> 00:21:24.880
I'm sorry, I'm sorry.
00:21:25.039 --> 00:21:26.799
We started it February 1st.
00:21:27.200 --> 00:21:30.640
Yeah, I mean you've you've done a lot within the period.
00:21:31.039 --> 00:21:36.880
So there are a lot of things on our side that are not at the level that we would like it to be.
00:21:37.200 --> 00:21:46.960
And so being just some of the UI or UX that should inherently answer questions, docs are okay, but could be much, much better.
00:21:47.119 --> 00:21:55.680
And so a way to sort of like tackle that, we do invite every customer that we see potential.
00:21:55.759 --> 00:21:59.680
We have sort of like an automation flagging mechanism when people sign up.
00:22:00.240 --> 00:22:01.759
We do invite them all into Slack.
00:22:01.839 --> 00:22:06.400
Like we have like 350 channels of Slack now, of customers in Slack.
00:22:06.720 --> 00:22:16.640
And through Slack, we talk to and mitigate these things because our customers were like, hey, any questions you have in the docs, not in the docs, not sure, just ask us.
00:22:16.720 --> 00:22:18.400
We don't ping anyone, just they ask us.
00:22:18.559 --> 00:22:21.039
And then they have this like free flowing conversation with us.
00:22:21.279 --> 00:22:27.440
Every customer that's entered Slack, I don't think anyone has ever not successfully gone to production.
00:22:27.599 --> 00:22:28.880
Like, not a single one.
00:22:29.119 --> 00:22:38.960
Because in that way, one, they have the feeling of like someone is hearing them, like support, but also any questions, any feature requests, anything is done really, really fast.
00:22:39.039 --> 00:22:44.240
And so what I'm trying to now achieve is like that
we don't have to be talking to them through Slack.
00:22:44.319 --> 00:22:52.319
Not that we'll stop, but can we automatically mitigate these things through like workflows and sign up and a better C line, a better documentation, whatnot?
00:22:52.640 --> 00:22:59.519
When you look at the number as an 18-person team managing 350 Slack threads, yeah, my engineering changes my engineering team is like, what the fuck?
00:22:59.599 --> 00:23:00.240
Like, what are we doing?
00:23:00.319 --> 00:23:05.039
Like how Yeah, I mean, naturally, as an investor, my question is how do you operationalize that?
00:23:05.279 --> 00:23:08.960
Do you have any tips for other people who are trying to manage hundreds of accounts?
00:23:09.279 --> 00:23:14.720
So the good thing is like there's so like just a shout out, I don't know uh who invested about Pylon we use as a support ticketing system.
00:23:14.880 --> 00:23:15.119
Yeah.
00:23:15.279 --> 00:23:18.319
And so they they are a layer on top of Slack, and so that helps out a lot.
00:23:18.720 --> 00:23:20.799
So it turns Slacks into tickets, right?
00:23:20.880 --> 00:23:31.680
So it's it's much, much easier to see visibility, but also definitely growing out, adding more people right now to like the GTM team that can sort of like filter errors before engineering gets in there.
00:23:31.839 --> 00:23:41.039
The other part why we did put engineering at the beginning, and this is part of like our thesis, is because we were we started building in February, so it's like very, very early, and you're bringing on customers.
00:23:41.279 --> 00:23:44.160
A lot of the support requests were actual problems.
00:23:44.240 --> 00:23:45.920
Like there were like bugs, there were issues.
00:23:46.000 --> 00:23:48.640
And so no one can solve that outside of the actual engineering team.
00:23:48.720 --> 00:23:58.319
And so it's great that the engineering team is like very front-facing, and then they'll solve that faster, better, get better understanding of what the actual problem is because there's no message lost in the ether, right?
00:23:58.400 --> 00:24:00.319
You're directly talking to the person that has the issue.
00:24:00.559 --> 00:24:06.000
Now, as we become more stable, and it's like a very you can see the patterns in in issues, what they're happening.
00:24:06.160 --> 00:24:10.720
And so a lot of these things now can be solved without the engineering team or structured in a way that the engineering team can do that.
00:24:10.799 --> 00:24:15.279
And so now we're sort of like separating the engineers from from the Slack channel slightly.
00:24:15.519 --> 00:24:28.319
Yeah, and you mentioned a good market team, and I know that you're primarily bottoms up today, and you probably have some really interesting lessons, positive and less positive, about your experience building developer tools.
00:24:28.799 --> 00:24:34.400
Is there any like feedback or pieces of advice for folks who are starting out to get the right reach?
00:24:34.799 --> 00:24:44.559
So reach uh a lot of people ask me this, and I feel that like creating the PLG developer community thing is very different for different types of people.
00:24:44.799 --> 00:24:47.039
So I suck at writing.
00:24:47.279 --> 00:24:48.240
Like I'm terrible.
00:24:48.319 --> 00:24:49.599
Like I was speaking right now.
00:24:49.759 --> 00:24:50.319
Yeah, exactly.
00:24:50.559 --> 00:24:56.319
So writing, like it's a really it's really hard when I do write an article, it takes a long time for me to do, so it's not efficient.
00:24:56.480 --> 00:25:04.559
And so let's say Swix, for example, was like he created the whole his whole persona on his like DX Tips blog and the book he wrote and things like that.
00:25:04.640 --> 00:25:05.920
So he's great at writing, right?
00:25:06.160 --> 00:25:23.200
And so I to always tell founders like do what's like most natural to you, and one that actually comes off natural and it's not like bot or it's not fake or whatever, because we have done in the company conferences, we've done in-person things, and our whole GTM is in person.
00:25:23.279 --> 00:25:32.480
Like we are now adding other things, but the whole GTM was in person just because that's the most natural version of us that you can get, which means it's the easiest on us to do.
00:25:32.559 --> 00:25:39.359
Not that it's easy to do at an event every two nights, like it's it's a job, as you know, doing events, but it's easier for us to do and it comes off more natural.
00:25:39.519 --> 00:25:47.200
I think people are attracted to that because you're in your natural sort of state and then conversing and creating things in your natural state.
00:25:47.440 --> 00:25:50.799
Is there anything you tried early on that didn't work?
00:25:51.279 --> 00:25:52.720
There's so many things that didn't work.
00:25:53.599 --> 00:25:59.839
A lot of things throwing money in the sense of like, oh, I'm gonna go to conferences and then conferences will generate me X, Y, Z.
00:25:59.920 --> 00:26:01.119
Maybe yes, maybe no.
00:26:01.359 --> 00:26:10.079
When we started doing those, what we found is that we have to be very selective of which conferences we go to and if we can get, you know, potentially a speaking slot in this conference.
00:26:10.240 --> 00:26:17.200
You can't get them always, but if you can get them, you get more people coming to your event, you get the thought leadership around that, and all that works.
00:26:17.440 --> 00:26:25.759
The things that don't didn't work for us were we do happy hours meetups, hackathons, conferences.
00:26:26.240 --> 00:26:28.559
We're also doing a big Daytona conference.
00:26:28.720 --> 00:26:41.440
And so what we try with these things, all of them, is we'll set ourselves a budget and a few of them, like not just one, but it's like who are we inviting, why are we inviting them, and what are they getting out of it?
00:26:41.599 --> 00:26:45.920
So it's for us, we always think about what the attendees are
getting out of it.
00:26:46.000 --> 00:26:50.559
So it's like you're not making a meetup for a meetup's sake, you're not doing a happy hour for the happy hour's sake.
00:26:50.640 --> 00:26:56.000
It's like, can I create something that the people that are there will actually get a benefit of?
00:26:56.160 --> 00:27:01.519
A trivial example is like, if I'm inviting founders, can I invite also an investor or of investors and vice versa?
00:27:01.680 --> 00:27:09.839
Or if the founders I'm inviting, can they all share problems in a, you know, in a very open way that they're all you know going through?
00:27:10.000 --> 00:27:12.000
Or are they all potential customers of each other?
00:27:12.079 --> 00:27:15.200
And so I think about those things when I put them together.
00:27:15.519 --> 00:27:26.559
And when I create a few of them in a row and there's not a positive outcome of it, it's just like you're either burning money on this or people don't get anything extra out of that, then we sort of cut that off and then we try the next thing.
00:27:26.640 --> 00:27:33.839
And so we will try multiple different avenues of getting people together and talking about stuff.
00:27:34.160 --> 00:27:37.680
How important is it, do you think, to hire a head of DevRel?
00:27:37.839 --> 00:27:41.119
Or is there a moment when it makes sense to bring that persona on?
00:27:41.440 --> 00:27:43.200
So for us, it's now.
00:27:43.359 --> 00:27:45.119
Like right now is the per time.
00:27:45.279 --> 00:27:58.000
The reason I say for us is not because I think a lot of the things should come from the founder because you can actually authentically articulate what you're trying to say and show, and you are the company, you're a part of the company.
00:27:58.319 --> 00:28:02.319
And what has happened for us now, we have so many different events.
00:28:02.480 --> 00:28:08.319
So we had a meetup two days ago here in New York, we had a hackathon last week in San Francisco, and we're we have so many of these events.
00:28:08.480 --> 00:28:12.000
Like, I can't attend, I physically cannot attend them, let alone do my entire job.
00:28:12.240 --> 00:28:17.519
And they've come to the point where no longer do I have to be a part of it for it to exist.
00:28:17.680 --> 00:28:25.440
And so now that that has sort of been operationalized, we can now bring in a DevRel person that actually can now speak in the name of the company.
00:28:25.599 --> 00:28:26.319
I don't have to be there.
00:28:26.400 --> 00:28:30.880
But I think there's a you have to set the culture or the vibe or whatever.
00:28:31.359 --> 00:28:34.720
Your question is like, is the 996 thing that's something we did purposely or not?
00:28:34.799 --> 00:28:40.559
Like, same vibe is like our events now have a vibe and the type of person that comes and the type of person that speaks.
00:28:40.640 --> 00:28:40.799
Yeah.
00:28:41.039 --> 00:28:43.680
And so now we can sort of add a dev route person for that.
00:28:43.839 --> 00:28:46.640
I feel that bringing a devil person too early might also be risk.
00:28:46.880 --> 00:28:49.920
I also find it to be a really hard persona to hire for.
00:28:50.160 --> 00:28:50.880
Yes, very hard.
00:28:51.039 --> 00:29:00.000
I know a piece of advice we've given our companies that someone shared with me was going through the Discord and seeing what engineers are most engaged.
00:29:00.160 --> 00:29:08.559
But I also think it's this balance of you want someone that was pretty high EQ and charisma, but also it's like deep technical and understands the product and can articulate that well.
00:29:08.799 --> 00:29:09.759
It's not always the easiest.
00:29:10.079 --> 00:29:12.960
The devil you can also break up is like, is it a technical writer?
00:29:13.119 --> 00:29:14.640
Is it someone that's gonna build demos?
00:29:14.799 --> 00:29:15.599
Is it the speaker?
00:29:15.759 --> 00:29:17.279
Hopefully, it's all those things.
00:29:17.440 --> 00:29:21.839
And so people will, and I've done this historically, you'll just hire because it's a problem.
00:29:22.000 --> 00:29:23.519
Like this morning we had this whole conversation.
00:29:23.599 --> 00:29:25.599
We have so much content that we want to write.
00:29:25.759 --> 00:29:25.920
Yeah.
00:29:26.160 --> 00:29:29.119
Like we've recorded, we've done whatever, but there's no one to write it.
00:29:29.200 --> 00:29:30.319
Let's just like hire anyone.
00:29:30.400 --> 00:29:32.480
And then it's like, you probably shouldn't.
00:29:32.640 --> 00:29:32.880
Yeah.
00:29:33.119 --> 00:29:34.160
You probably shouldn't, right?
00:29:34.240 --> 00:29:37.519
So it's just a cost, a time and energy cost as well.
00:29:37.759 --> 00:29:38.079
Exactly.
00:29:38.319 --> 00:29:44.000
So one of the last questions I have is related to this whole go-to-market strategy, conferences, happy hours.
00:29:44.160 --> 00:29:45.440
You blogged about integrations.
00:29:45.599 --> 00:29:46.880
I think one being with Langchain.
00:29:47.119 --> 00:29:51.039
What does that ecosystem strategy look like for partnerships?
00:29:51.519 --> 00:29:52.960
I think that's a really big one for us.
00:29:53.039 --> 00:29:54.079
So we've been lagging on that.
00:29:54.240 --> 00:29:55.279
Langchain has been great.
00:29:55.359 --> 00:29:56.880
Congrats to them on their new round.
00:29:57.039 --> 00:29:59.519
They've been like just great for that partnership.
00:29:59.599 --> 00:30:00.799
And I really enjoyed that.
00:30:00.960 --> 00:30:06.640
I think it's like surface area of the way I think about it is surface area of availability of Daytona.
00:30:07.119 --> 00:30:14.640
So if you're building agents through like Langchain or anything else, and your agent needs a computer to do something, ideally it's already there.
00:30:14.720 --> 00:30:18.559
So the person like building the aging is just like, oh, my agent now needs a computer as well.
00:30:18.720 --> 00:30:19.920
Click done, right?
00:30:20.079 --> 00:30:34.000
And so of all
the not just agent orchestrators, but every type of company and or product that has an agent that can have the ability to spin up or need a computer, we should be there by default.
00:30:34.160 --> 00:30:35.440
And we haven't been very good about that.
00:30:35.519 --> 00:30:36.799
Long chain has been our first.
00:30:36.960 --> 00:30:40.960
And so we've actually hired people now, engineers in the go-to-market team.
00:30:41.200 --> 00:30:41.599
Nice.
00:30:41.839 --> 00:30:42.640
To be able to do this.
00:30:42.799 --> 00:30:47.839
So I had one originally, and then my CTO was like, I need him for a week, and then I'll give him to you.
00:30:48.000 --> 00:30:49.200
I've never got that engineer back.
00:30:49.279 --> 00:30:49.920
That engineer is gone.
00:30:50.240 --> 00:30:51.119
Now you learned your lesson.
00:30:52.000 --> 00:30:53.279
Now we hired our next one.
00:30:53.359 --> 00:30:55.200
I'm like, you're never touching this engineer ever.
00:30:55.359 --> 00:30:56.000
He's in our team.
00:30:56.240 --> 00:30:56.799
Don't share his name.
00:30:56.880 --> 00:30:57.279
Yeah, yeah, yeah.
00:30:57.359 --> 00:30:57.440
No.
00:30:57.599 --> 00:30:59.039
Do you want to end with a quick fire round?
00:30:59.200 --> 00:30:59.599
Sure.
00:30:59.839 --> 00:31:00.079
Okay.
00:31:00.240 --> 00:31:04.160
What's the one developer tool that you can't live without outside of Daytona?
00:31:04.559 --> 00:31:07.279
Oh, so you said cursor before, but that would be the one.
00:31:07.359 --> 00:31:07.519
Yeah.
00:31:07.759 --> 00:31:08.400
That's fair.
00:31:08.559 --> 00:31:09.839
What powers your work day?
00:31:10.079 --> 00:31:14.240
What powers my work day is probably like white monster energy drink.
00:31:14.319 --> 00:31:15.440
That probably powers my work day.
00:31:15.680 --> 00:31:18.559
If you weren't working in developer tools, what field would you be in?
00:31:18.880 --> 00:31:19.680
Energy.
00:31:20.799 --> 00:31:21.119
Energy.
00:31:21.200 --> 00:31:23.599
I think that's where the money's going right now in data centers.
00:31:23.920 --> 00:31:29.599
It is for data centers, but I think that energy, like if you have abundance of energy, you have abundance of everything.
00:31:29.759 --> 00:31:31.519
And I think that can change the world a lot.
00:31:31.680 --> 00:31:34.400
I think the most valuable thing in the universe is time.
00:31:34.559 --> 00:31:39.440
You can't, as far as I know, we can't manufacture time yet, but you can manufacture energy, which can essentially give you time.
00:31:39.519 --> 00:31:41.519
And then so that is something that I would do again.
00:31:41.920 --> 00:31:44.480
What's the best piece of advice you've received as a founder?
00:31:44.799 --> 00:31:49.279
This was an advice to me specifically, but I saw an interview with Sarah Tavill from Benchmark.
00:31:49.359 --> 00:31:49.680
Yeah.
00:31:49.839 --> 00:31:52.079
And when she said that, that was like, again, markets.
00:31:52.160 --> 00:32:00.720
It's more like you can have if like a market is like a river stream going and you have like an okay founder, they'll still kind of get where they want to go.
00:32:00.960 --> 00:32:06.240
But if like you're against the stream or there's no stream, you can have a best in the world founder.
00:32:06.480 --> 00:32:10.079
They might be successful, but probably less so.
00:32:10.240 --> 00:32:14.000
And so that's something that I didn't understand intuitively earlier in my life.
00:32:14.079 --> 00:32:17.279
And so I think that's one that's really like make sure you're in the right market.
00:32:18.400 --> 00:32:18.960
Awesome.
00:32:19.200 --> 00:32:20.160
This has been really fun.
00:32:20.400 --> 00:32:20.880
Like this.
00:32:21.440 --> 00:32:22.319
Thank you.