This is an interview with Jake Cooper, the founder of Railway, a tool to build powerful infrastructure without the pain. During the interview, Jake got a tsunami warning, but carried on without batting an eyelid.

You are in danger. Okay. Interesting.

We cover why Railway is working on arguably the hardest problem in dev tools, why they're completely remote, how and why they're building their own data centers, which I think is a first on scaling dev tools, and why a low email open rate led them to completely rethink how they do sales.

I don't know if you ever, like, watch Minority Report. It's like precog sales. Essentially, you're kinda, like, tap into, like, when this person is actually going to, like, want to be helped. Right? So that we can potentially project, like, hey, maybe we can be around for for you to go do that.

Right? The future can be seen.

Finally, we talk about why Jake thinks you absolutely need to do hard things. Enjoy.

It's funny because I was I was talking to somebody the other night, and I was I think my maybe controversial opinion about remote despite running a remote company is that I think it's a terrible idea for, like, the vast majority of people. I think it, like, requires a lot. It requires a lot of foresight. It requires a lot of thought. You get a lot of stuff for free just literally being in San Francisco.

I mean, other I mean, it's not free. It costs you, like, chill or rent and and, obviously, all the other things. Right? But, you get a lot of stuff from just, like, being in person. Right?

So I don't wanna, like, diminish the aspect of, like, being in person. We get in together. We get everybody together, like, twice a year. We fly people out to you know, we're just in New Buffalo. I think, like, the next one will be in Portugal.

We try to, like, cycle between western and eastern hemisphere just so people don't have to travel too much. But, yeah, I think, like, there's a lot of benefits to being in San Francisco. I think, like, I personally should be in San Francisco. There's just, like, so many people here in general, whether it's, like, talking with customers or venture capitalists or, like, literally anybody in between. Right?

Like, there's just so many benefits. Right? Awesome founders who you can just chat with, stuff like that. But I think that running a road come up to me gives you some amount of advantages if you're willing to do the work, if that makes sense.

Yeah. What would you say are the advantages?

I think you can, like, you can have people be hyper autonomous from, like, day 1. Right? So you can hire people who are going to be able to go and drive various things in various different zones. If they're in disparate time zones, it's almost like a really, really strong forcing function of, like, how do you build your process in such a way that, like, if even if I have, like, minimal amount of time to interact with this person. Right?

So if I have a director like, I have 2 direct direct reports in Japan. Right? Right? Right?

I saw them.

Because they work they work still flat. Right? But the time zone overlap is garbage. Right? It's I mean, actually, in Japan, it's not too bad, but I've we have one person who's in, you know, Thailand or Australia, and that one's actually, like, super, super hard because it's like it's like you have, like, basically 1 hour of overlap.

Right? And so if the people aren't autonomous, can't drive, etcetera, all of those things, and they're gonna, like, essentially be sitting kinda waiting around, It makes it extremely difficult to run a remote company. Right? So there's, like, benefits, which is, like, you know, it it kind of, like, forces you to go and do that skew that kinda general direction, but there's also drawbacks in terms of, like, you know, potential people who you can hire. Right?

Like, we haven't figured out a good way to do, like, junior mentorship or, like, internship stuff, like, remotely. Right? Because it's like, we don't wanna be, like, basically discriminatory about, like, where we hire, but, like, at the same time, you know, those those, you know, new engineers, they need a little bit more kind of, like, touchpoint interaction. They need maybe a little bit more help and stuff like that. Right?

And if if your time zones don't overlap, then you can potentially have these people kind of, like, drifting in space waiting essentially. Right?

And, you know,

as a super lean team remotely working on really hard stuff, that presents a kind of, like, challenge that is very difficult to overcome and that we haven't figured out how to go and do yet. So I don't wanna, like, paint a picture that's, like, all roses or whatever, and that everybody's, like, doing it wrong because it's just, like, it's chocolate or vanilla. It's different ways to run a company.

Yeah. This episode is brought to you by WorkOS. At some point, you're gonna land a big customer and they're gonna ask you for enterprise features. That's where WorkOS comes in because they give you these features out the box. Features like skin provisioning, SAML authentication, and audit logs.

They have an easy to use API and they're trusted by big dev tools like Vercel as well as smaller fast growing dev tools like Nock. So if you're looking to cross the enterprise chasm and make yourself enterprise ready, check out Work OS. We've also done an episode with Michael, the founder of Work OS, where he shares a lot of tips around crossing the enterprise chasm, landing your 1st enterprise deals, and making sure that you're ready for them. Thanks, WorkOS, for sponsoring the podcast, and back to the show. Yeah.

That makes sense. I I like the way you put it in, one of your blog posts, or I think one of you said one of your team said that you're you should hire, torpedoes.

Yeah. Yeah. Exactly. That that would be Miguel. He he works on the proxy.

He's basically single handedly built the, proxy and and edge network, and removed, like, we used to be on Envoy, and and Matt Klein was awesome. But at some point, he was basically, like, you guys have millions and millions and billions of, like, these like, actual, like, edge services, that are are, like, going to need to be like, the XTS protocol is, like you know, at some point, it it slows down. Right? And you're seeing those things. So Miguel built all of that from scratch, but, you know, he's talking about a torpedo.

It's basically like, it's like, how do you get people to to realize that, like, speed is almost the only thing that matters? And if even if you're, like, if even if you're pointed almost in the wrong direction, if you can get, like, 5 shots on that, you're you're at, call it, like, a 178 degrees. Right? You're, like, not completely opposite. Right?

But, like, you run 5 iterations of that kind of, like, loop, essentially, you're mostly gonna be on target. Right? So just, like, continually just, like, quickly, recalibrate, like, where you you should be in general and then move towards that. You know? So I think that that's the the torpedo metaphor that that they were talking about.

Yeah. That's really cool. Also, just to, like, hit on the flat org structure, so there's is that there's about 20 people right now you have.

Yeah. Like, 25 people now.

Okay. And so if there's 25, 24 are reporting to you, basically. Yep. Yeah. So how how does are you, like have you seen apparently, Elon Musk has, like, one minute chats with people, like, back to back, or, like, are you kind of sat there and, like, people are coming?

Like

Yeah. So, I mean, it's not that bad at that point because I think, like, he did that with, like, all of, like, Twitter or x when he acquired us. So there's probably, like, you know, hundreds of people and, like, in and out. And, like, I think the the meme was that, like, it it's the it was from, like, the, like, Mark Andreessen talking about this where, like, you know, he went into Twitter and just, like, sat with everybody. So, no, it's not that bad, yet.

I run 1 on ones with people every 2 weeks. We go and sit down for 30 minutes. I have office hours. People can put me aside. I personally bias for pretty much everything in public, so I have a little, I have a a little, like, channel inside of so we run the whole company in Discord.

I have a little channel inside of Discord called, like, Cooper Inbox. And so if you wanna pull me aside or anything else like that, obviously, like, my philosophy is that, like, DMs are private. Right? Like, they're they're personal, etcetera. If you have, like you know, you can't make your one on one because doctor's appointment and stuff like that.

Right? Where you and and it's important for people to, like, have a space for that. I think that's something that I I I switched on earlier. I was like, used to be like, no DMs. Like, you don't need whatever.

Right? And I was like, actually, the only thing that matters is that information, like, flows through the company in the most important way.

You had no DMs at the beginning.

Well, I was philosophically opposed. Right? Because I was basically like, well, the optimal way to go and do this is is to put it everything in public. Right? You just share it in public, then everybody has access to it, etcetera.

And then I kinda flipped on it, and I was basically like, it doesn't matter so long as the information kinda gets there. And my job as a leader should be to, like, skew people to say, like, if I want, you know, everything to go into public, like, what are the reasons for which people are using DMs? Right? It's like, okay. Well, then we can maybe, like, massage those and move them them towards there.

But it's kind of, like, always like a it's like an asymptotic goal. It's like, you're never gonna have, like, no DMs or whatever. Right? So I think it's, like, it's important to it's important also, like, I think as a leader, be able to statement, like, you're just wrong. I think that one that one's important too because people are, like, just like, oh, shit.

Like, they make mistakes too. Like, I can also make mistakes, and that's totally fine. You know? Because mistakes are good. It means you're moving fast.

Yeah. The torpedo is adjusting. Exactly. Yeah. That that's it's very interesting.

And you were talking about scaling with leverage. You were kinda DM ing me this, and it was this within this context of, like, technical? Is this, like, an org thing?

Yeah. So I would say it's it's, like, it's both, if that makes sense. Right? So we, like you know, we we just brought on our kind of, like, first salesperson. Right?

And I think we're in a we're in a good spot right now, from a revenue perspective where things are, like, growing well, etcetera. And we're, like, trying to figure out, like, what does that look like in general. Right? And so I think, like, the standard, you know, operating procedure for a company is, like, you know, they hit, probably what you call, like, product market fit where you're growing, like, you know, call it, like, 2 to 3 x year over year, and you're trying to, like, go and add more revenue. And so you, like, go ahead and you just hire a shitload of, like, sales reps and and go and do this thing and, like, try and throw bodies to the wall or whatever.

And you do the same thing from, like, a marketing perspective. You do the same thing from an engineering perspective. You just, like, you just say, like, okay. It's really working and, like, you know, we're pulling down this amount of money this this, you know, this month, and, we need to be able to operationalize it. Right?

Like, immediately. Right? Like, where you can just say, like, oh, cash flow in, you know, cash flow out. Right? Mhmm.

And I think the the the thing that we do that we kinda try and lean more towards is, like, basically trying to build systems over time that can operationalize that with leverage. Right? And saying like, okay. Instead of hiring, you know, a ton of, like, line rep, like, support engineers or stuff like that, where we're just like, okay. Let's, like, we can hire a ton of these people.

We can put them all in different time zones. We can, etcetera, like, have optimal coverage so that, like, our, metrics, like, trend towards, like, time to first resolution of, like, 0, basically. Right? Like, getting back to your customers as fast as possible. Right?

Like, we've kind of actually moved in a little bit of a different direction where we've said, like, let's hire 3 really, really awesome, like, support engineers. Right? Like, full on engineers. Right? Like, they're not support reps or anything else like that.

Like, they're building out the internal help station tool that we use for all of our, like, support comms. Right? And so we've, like, invested on that front essentially. Right? So that over time, right, like, when we go to our next factor of scaling, we can say, like, hey.

We do actually want global coverage because what the limiting reagent right now is is basically whoever's on rotation, they have to sleep at some point. Right? So, like, how do we figure out how to go and scale that to, like, go and drive that down? Right? And so we try and think about, like, building leverage every single step of the way so that we can hire really, really excellent people and compensate them well.

Right?

Yeah. That makes sense. And it it kind of aligns, Angelo, who's amazing on your team, wrote this really cool article about how you were looking at doing sales. And, also, I think Angelo's is actually maybe interesting as well just to briefly say that he said he was, like Yeah. Started out as an SRE and then moved to software engineer, product manager, now doing sales.

So, like, just, like, does and still writes code, I think. Right?

Yes. Yeah. He still does. Right? So we have a lot of, like, multifaceted people.

Right? Angelo is definitely one of them where people are just like, I wanna come here. I wanna do excellent work. You know, I don't want you to put me in a box. I don't want you to, like, go and do all of these other standard kind of things, etcetera, while also having the understanding that, like, that leverage and that compounding comes from actually, like, building in a space.

Right? And so I think Angela has chosen kind of, like, the sales support solutions kind of area to, like, make their own, and really, really go deep on that. Right? So yeah.

Yeah. It's it's really interesting. And I think, what what he was saying, and that you can correct me on this, but from the article was that you were looking at, like, sales, learning sales, following the following the playbook, and you were like, nothing none of this is gonna work. Learning to play golf is not gonna work. You were looking at the email reply rates were, like, no point 3 percent or something for, like, even your existing customers.

Whereas, like, outbound cold emails from from your recruiter was, like, no point 5. And you were like, this is just not gonna work. So I guess the non scale answer would have been just hire loads of salespeople to send a lot more emails, but then you had a quite unique approach that maybe you you can share a little bit about what you built.

Yeah. I mean, so we basically like, you know, we noticed this, and then Angelo was the first to pull it out because he was like I think he pulled, Ian, our our, you know, the talent lead aside. And he was like, hey, dude. You know, I'm getting, you know, I'm getting, for lack of better with my ass beat on these these emails. I send so many of them out and, like, nothing comes back.

Right? Like, nobody responds. Right? Like, how do you craft a really, really good cold email? Right?

And I think he got some tips and he got a little bit more kind of, like, leverage out of it, but it was still, like, not very good. Right? And so I looked at this and I was like, you know, I'm an engineer by trade. Like, why would I ever respond to this, like, email? Like, what's the what's the what's the bonus or what's the boon or, like, what's the what's the reason that I would actually go and do this?

Right? And, like, just as a more, like, philosophical question, not even, like, in the current context. It's like, why would I respond to this email? Right? And the reason I would respond to this email is, like, if this person was giving me something that I want.

Right? Either, like, you know, in in a in a kind of, like, more predatory way, like, like, the the feature is gated. Right? And I want to unlock that feature and go and do those things. Right?

And so that's why, you know, some people respond to, like, traditional sales emails, right, and saying, like, I need this quota or I need this thing or whatever. Right? But what's the, like, best possible experience that we think that we could offer to somebody? Right? It's like, they get to a point where, you know, either they need some, like, reassurance because they're like, hey.

You know, like, I'm scaling on the platform. I wanna talk to somebody. You know, like, at some point, you do wanna actually, like, know who's behind your kind of, like, infrastructure so that you can be like, oh, these people are trustworthy, etcetera. So that's one reason. Right?

If you have a technical problem that you're, like, looking to go in and solve, that's another reason, etcetera. Right? So, like, at what points are people actually going to, like, want to talk to people, and at what points, which is the vast majority of points, will they have absolutely no interest in in talking to anybody on, like, the sales solutions, etcetera, kind of side. Right? And so what we've done internally is we've, you know, we keep segment events for, like, every single interaction that happens.

Like, we don't track, like, users from, like, a cookie perspective or, like, we don't have any front end tracking or anything else like that. But we do admit when people get to, you know, get to green. Right? So their first their first deploy or anything else like that. Right?

And so we can actually try and figure out, like, you know and this is, like, a lot of trial and error. Like, when do people actually respond to Outreach, and where do they respond to it most. Right? And it ends up being that, like, if you can meet the user where they are, right, and say, like, hey. You've already got some stuff, like, spun up.

You're getting to a point where you're, like, you're starting to create tickets. Right? We have all of that information, and we can add new segment events because we run our own help station. Right? Like, you've you made a couple questions and stuff like that.

Maybe it's a good time for sales to get involved. Right? And say, like, hey. What are you actually trying to build here? Right?

Like, how do we get this inside of your organization? Right? Like, maybe you shadow ops IT did in in here, and your VP of engineering has no idea that this tool exists. Right? Like, what are the what are the kind of, like, blockers that we would need to, like, overcome to potentially move that through into the organization?

Right? And over the years, we've kind of, like, done that. And over the years, we've kind of removed the objections. And over the years, we've gotten a little bit better at kind of, like, doing the, like, product led growth with also the, like, sales kind of, like, assist motion. You know?

Yeah. It it's really smart, honestly, because I I think you illustrate it really well in the article where it was like, you know, there's times where you don't the salespeople are just so annoying, and then there's times when they're all caps emailing you, like, what's going on? Like, you know, why or, like, how do I fix this thing? And it's like that fight that you don't want them to email you at that point. You want them to kind of you you're trying to anticipate when maybe that kind of point is gonna happen.

Yeah. I think Angelo's used the, the term, like, I don't know if you ever, like, watched minority report. It's like precog sales. Essentially, you're kinda, like, tap into, like, when when this person is actually going to, like, want to be helped. Right?

So that we can potentially project, like, hey, maybe we can be around for for you to go and do that. Right? And it it helps if your time to resolution is really quickly or really quick. You have people in disparate time zones that can help on this. Right?

You have multiple people, like, rolling through this rotation, etcetera. Right? So there's a variety of different things, but, like, at the end of the day, it comes comes down to, basically, how do you meet the user where they are when they are, like, when they're ready to, like, basically say, like, please, somebody help me. You know? Like, I have a problem or, like, whatever.

I need I need this thing. Right? And you're like, excellent. We can we can do that. Right?

But the worst possible thing is, like, you send sales cold outbound with no, like, personalization or anything else like that. And then when the user says, like, hey. My my queries are slow now. Right? It's like crickets for, like, 24 hours.

Right? Yeah. And, admittedly, there was a period where we're just not doing support very well. Right, where we where our time to resolution was, like, it was in the, like, 12 hours, like, you know, pushing on to, like, you know, 24 hours. Right?

And since since, like, then, we've, like, over the last year, we basically just drilled that all the way down. Right? And so now we're getting back to people, like, way, way quicker. We're rolling out, like, ways for, like, people to be able to, like, page us. We've had that from, like, a enterprise level, but, like, how do you actually determine, you know, severity of impact on, like, the support side?

Right? Like, these all these all, like, engineering problems. Right? And that's why, like, we we hire support engineers. We hire go to market engineers.

We, like, we hire, you know, obviously, product engineers and and platform engineers. Right? But, like, we wanna build out these solutions with leverage so that we can actually kind of, like, scale them over time really, really well without having to just, like, continually add people to kinda, like, almost, like, backfill and hold this whole thing up. Right?

Yeah. It it it makes a lot of sense. It's still quite hard to imagine with, like, such a small team how you're able to kind of actually have, like, such a quick support function. Is is is there any, I guess you've got a lot of kind of, like you're trying to resolve their problems preemptively, and you're trying to have, like, really good discovery in terms of, like, help center and stuff like that. Was there anything else that you did to kind of reduce the time?

Yeah. So, I think this, like, one of our, like, support engineers, has this really, really awesome quote that I think he's put into a blog post, but that I, like, you know, steal all the time now. Essentially, support is the, like, the catch of, like, the try catch. Right? Like, users try documentation.

They try to figure it out on their own. They try to whatever. Right? And then when nothing is working and they can't figure it on their own, they, like, essentially, they throw up that exception, and they say, like

Yeah.

Help. You know? And and so support essentially is designed to almost like you know, I think people, like, look at it potentially in a myopic view. It's like, well, there are there are tickets. How do I go in and handle them?

Right? And then you just, like, focus on, like, ticket resolution, all those other things. But, like, if you think about it a little bit more holistically, there's all of that other stuff that, like, leads up in front of support. Right? It's like, is your documentation clear?

Like, are your deployments, like, are going green? Right? Like, we've built our own kind of, like, internal build and deploy system to, like, statically analyze code, and then we take some of the outputs and figure out, like, hey. This was a Django app. It didn't get off the ground.

Why did it not get off the ground? Is it something we can go and fix in any of these, like, next packs providers to go and make this better? Right? All of these things sit in front of support. Right?

So support is not just like, I have a ticket and and I want a resolution. It's actually, like, I have a problem and I want a solution. Right? Which, like, expands the scope for which you can operate in. Right?

And that that, like, goes all the way up to our community. We have, like, awesome people in our community who've, like, built stuff out. We've incentivized them, you know, with swag and credits and all these other things. Right? And so we, like, go and push people to basically say, like, hey.

If you have an issue, in all likelihood, somebody else also has the issue. Right? Like, how many times have you just gone to Stack Overflow or GitHub or anything else like that and be like, oh, this is literally my issue. Right? And so, like, we're building a corpus, over on the, like, help dot railway.com site, of all of these kind of issues.

And so those users can, like, run into that issue, figure out it's an issue. Either it's a platform issue and it has to get bubbled up to us, at which point we can, like you know, we built the whole systems. We can attach a linear ticket to it. We can go and, push it through triage. We can go resolve it, and that whole life cycle is all automated.

So whenever the ticket goes to in progress, these these will be notified. Whenever it goes to resolved, it'll they'll also be notified. Oh, that's cool. Yeah. And then yeah.

So then you can just, like, group and aggregate these these things. Right? And, again, those are only things that we can go and do because we get built out our support experience so that we have the ultimate control to basically say, like, how do we build the ultimate amount of leverage?

Yeah. That that that was really, really cool. And it it kind of actually makes me, puts me on to a question I kinda wanted to ask you because I feel like you're actually addressing, like, the hardest problem in dev tools, building railway. It's like I I I heard you on the, dev tools FM podcast talking, which is really good, talking about how, you're like, why why someone's asking you why why are you building railway? And your your answer was like, why wouldn't I?

This is, like, this is, like, the hardest problem and probably the big biggest limitation for most people that wanna build stuff is I I've had it myself where, like, I wanted to build something for the first time that was a lot more complex than just, like, something I could build with, like, a, you know, classic kind of, like, you know, Next. Js, whatever. And suddenly you wanna deploy this really complex thing that dot, like, process videos, in my case. And, you know, suddenly you have to spend so much time on infrastructure, which is really hard. And, you know, it kind of solves with AWS in the sense of, like, you can do it.

It's like you could build your whole stack on AWS, but it's very, very complicated. And you I what I think is so hard is that you've got this challenge where you've gotta be, like, you've gotta be able to do everything that you do on AWS, but it's gotta be, like, significantly more simple that, to to do than to to do than with AWS, I I would feel like is your is your challenge, which I I've spoken to a few startups that are doing this, and I always think this is the hardest the hardest problem to be solving. I'm my my question is, like, how are you yeah. Yeah. Yeah.

So there was a question in there, which was just, like, I I guess, like, how do you think about that kind of, like, the complexity and the simplicity, like, the power and the and the simplicity?

Yeah. So I think, like, my view on this has changed a little bit over time, but it's, like, mostly just kind of, like, shifted a bit. Right? And I think there's a couple like, we've been talking about, like, design principles internally at railway and, like, what our design principles. Right?

And I think one of them that, like, you know, I've I've I've really, really liked is, like, almost expose so the the principle, if I had to put it into words, would be, like, expose complexity as depth incrementally. It's kinda wordy, and we'll we'll work on that. But, essentially, what it means is that, like, the user should always right? So if you talk about, like, Donald Norman's, like, design principles. Right?

Like, visibility feedback and mapping. Right? And so mapping is really, really important. It's basically, like, where does the user kinda map this in their brain? When you think about the complexity, it's essentially like when you're designing a feature, you should really know where your user is almost going, like, next.

Right? It's like, if I am in the current user life cycle, what am I trying to go in and do? Right? Like, and and where should that kind of, like, feature or toggle or anything else like that exist in the kind of, like, visual space. Right?

And that's why we've, like, invested really, really heavily in the Canvas. Right? Because we're basically saying, like, hey. We wanna make it really easy for you to only kind of, like, think about the things that you really wanna think about right now, and the other stuff can be removed. You can kind of, like, know where it is, like, vaguely in space in, like, kind of, like, an age of empires, like, fog of war type thing.

Right? But, like

I was just thinking that's literally just what I was thinking about, like, that kind of, like, age of empires where you get dropped in and you've just got one little village. You got the town center and nothing else. And then Yep.

Exactly. Right? So I'm I'm, like, a I'm reasonably, I would say big big gamer. And so, like, the the 2 that I pull from from, like, gaming is, like, it's either factorial for infrastructure or it's, like, you know, the age of empires come like fog of war or anything else like that. But, yeah, essentially, like, how do you figure out ways to, like, give your user the next step and only the next step.

Right? And so a second principle that I really, really like is, like, make the user make no choices. Right? And so I think, like, people kinda mess this up generally when when they kind of, like, say, like, hey. Here's a model and, like, here's 2 options.

You have to make a choice. Right? Like and the the my my thing about this is, like, people think that there's, like, 2 options and you can, like, look at conversion and stuff like that. Right? But you have to also look at the fall off in the funnel.

And when you give a user 2 options, they're gonna pick a 3rd option, which is close the tab. Right? Even if you give a user one choice and say like, okay. Cool. Right?

Like, we have our terms of service. Right? We have 90 98.6% conversion from, like, terms of service to, like, sign up. And the terms of service is literally like it's like a it just goes through to the like, there's no actions that you have to do. You literally hit a button, the big button at the bottom of the page, and you still get 1% drop off in that funnel.

Right? And so if you make the user make choices, you're gonna, like, you're gonna overload them with complexity. Right? Like and so that's what we think about is, like, don't make the user make any choices and, like, whatever the next possible hop is in that user journey, make it as close as possible so that the user can actually say, like, oh, that's, like, that's where I was going. Right?

Like, it's so easy. It's, like, really, like and then you draw, like, a very, very thin line for them to say, like, that's the thing. Right? Like, here's the button. Right?

Because in an ideal world, that's the best possible experience is you hit a button, you get the resolution. Right? You hit a button, you get the resolution. Right? I think this is thing something that, like, you know, AI for all the type or whatever is, like, you know what the next possible step is.

It's like you move to natural language and you say that was wrong. Right? And, like, getting more accurate over time is a thing that, like, will happen with AI, but there's a very tractable next step, which is, like, you see the thing and you know where to go.

Have you got any examples of, either things that you see other people other startups, like, getting you to choose between or things that you were getting people to choose between initially and you kind of now is like a there is no choice just to kinda illustrate?

Yeah. So I would say that, like, you know, I think something that people like, other people potentially make mistakes on is, like, okay. Let's do provisioning as an example. Right? You're trying to provision a service.

Right? I know based on your repository, like so I could I could do I could do 2 things upfront. Right? I could do the easy thing for me, which is basically say, like, tell me that this is like a Docker file and that it exists here and, like, you know or otherwise, you know, tell somebody to select, like, a build pack, like Cloud Native build pack or whatever or anything else like that. That's, like, the easy thing to build.

Like, the UI is, like, pretty easy or whatever. Right? But it requires the user to make a choice. It requires the user understand what the Dockerfile is. It requires the user to understand what the build pack is.

It requires and that's one step. Right? Like, this is why the AWS, like, configuration thing is, like, kind of a mess. It's like, you go through it and you're, like, okay. I'll select, you know, my name.

Okay. Cool. That's good. That's not really a choice. Like, it's fine.

And then you move on to the next thing, and it's, like, do you want a bunch of other operating systems or stuff like that? And you're, like, what the fuck? Like, what am I doing here? Like, what do like, I I have no idea. And then you go you close a tab to, like, go and figure out something else, and then maybe you never come back.

Right? And so that's the attrition in that funnel. Right? And so if you can actually do all of the hard work to basically, like, statically analyze the code and stuff like that and be like, this is probably this. Right?

This is these are these are these are probable outcome or whatever. Right? And then, you know, maybe not even make the user make any sort of decision. Just say, like, that was the thing. Right?

And then kind of, like, usher them forward by saying, like, hey. We weren't able to build your thing. Right? Like, we statically analyzed it. We look for any of the stuff, etcetera.

Right? And then try and figure out where they're trying to go next. Right? Like you know? And then, you know, build that into your experience and say, oh, actually, it looks like, you know, we have a higher percentage of, like, failures when it comes to, like, a monorepo.

Maybe, like, our monorepo detection needs, like, work for the stuff like this because people are trying to, like, you know, push a front end and a back end into here, and they don't know that they've got to, like, you know, kinda select in into the, like, front end folder, right, and set a root directory.

It's

like, okay. Well, like, how do you move those things a little bit more forward so that the user doesn't really need to, like, make decisions to say, like, give me your root direction? They just say, like, I wanna deploy my repository and, oh, like, it is my front end. Right? Perfect.

You move that forward.

Yeah. That that makes so much sense. And I think, like, that's kind of was, like, the magic with, like, Vercel was, like, you just you just do it, and it just works. And it's like, you don't even have to make any selections around it. So it's like it's like you're trying I guess you're trying to you're you're bringing that kind of experience to, like, much more complicated, like, infrastructure setups kind of thing, where it's like you're you're get you're guessing or not guessing, you're, like, kind of figuring out what they probably need and doing that for them.

Yeah. And then they can back out of that later if it's not the thing they want and, like, displaying more.

Yep. And, hopefully, you can give people things that, like, are almost like double duty. You know? So, like, they they solve multiple problems. Right?

And so, like, an issue with, like, call it, like, monorepo deployment or, like, microservices or any or, stuff like that is basically, like, what is the ordering of my service? Right? Like, I think Lyft had a PR open on Kubernetes for, like, years, where they were just like, we need, like, ordering or anything else like that. Right? And they're like, no.

I don't know if that I think that that might be the PR, but, anyways, you need ordering and stuff like that, and many people will integrate, like, Basel or, like, custom workflows or anything else like this to go and and make this happen. And so well, how do we, like, derive, like, ordering? Right? Like, as a just an implicit thing. Right?

It's like, well okay. Like, we need an ability to, like, kinda, like, store variables. Right? And so, like, we know to know what your, like, back end URL is. Right?

And so if we build a system that allows you to, like, drag pieces on the canvas to say, this is my back end and it needs to consume this variable on this database, then inherently, like, this service depends on this service. Right? And so this service should roll out before this other service. Right? And so you can construct that like DAG, and that's like a feature we wrote all that recently.

But it's like, if you're already using these features, you can just, like, get compounded benefits. Right? And I think that, like, as you figure out how ways to, like, go in and compress that experience to say, if you use a couple of these primitives, you get all of these outputs. Right? Then, like, you can incrementally adopt those things.

That's what ends up moving people towards, like, oh, this feels like magic. Right? Like, that's awesome. Right? It's like magic, you know, and and people are like, oh, what is magic?

Like, how do you like, magic is just doing less shit than you thought you would have to go into. Right? Like, that's that's it. That's that's the only thing that that that, like, quantifies magic. Right?

It's like, oh, wow. Like, it already works. That's awesome. Right? Like, it feels like magic.

Right? And so if you can figure out ways to just have the user do do less and get more, like, your product your product will feel more magical. Right? Like, a lot of the kind of trend of, like, moving stuff onto your landing page and, like, allowing people to access your product, right, is, like, you know, that's that's magic. Right?

Because you're just like, oh, normally I have to sign up. Normally I have to, like, whatever, do all of these other things. Right? Like, so you can just think about, like if you think about that compression, then a lot of, like, magical experiences just open up to you. You know?

Yeah. Yeah. That that's really cool. We're defining magic. I love it.

Yeah. And, okay. So one one question I've got, which is actually, I don't know if I if I caught that part, but Paul Butler, in case we did, Paul Butler is, a customer of yours, a fellow Canadian, and, a great guy running drifting in space. And, he was, get it he wanted me to ask you about your dedicated servers, your your bare metal servers that you're you're now operating. So it's not in the usual dev tools playbook of, like, open your own data center.

No. It's not. Yeah. And so, like, I mean, we've got 2 DCs online, a third one coming in over the next, like, month, and then, like, Singapore should be online in in January. Right?

So we'll have, like, 4 data centers in, you know, basically 4 different regions, and we'll have a global coverage of, like, you know, less than a 100 milliseconds, like, door to door for for literally anybody in the world, which is super cool. And I would say it's, like, going pretty well. I think we've got a we've got a blog post that we're working on right now. It's basically, like, how do you go from 0 to, like, like, running your own hyper hyperscaler in, like, 6 months. And there's a lot of really, really interesting stuff in there.

And so, like, you know, there's a lot of interesting stuff when it comes to running a data center. Right? There's obviously, like, the procurement of servers. Like, how do you buy them? Like, what do you spec?

Like, what are people interested in? And then there's like, okay. Well, like, how do you find a space for these things? Right? Yeah.

You know, like, we basically have to, like, find a building. Right? And so you have to, like, you have to, like, go to Equinix or Digital Realty and, like, compare all these things. Right? And then, like, you have to kind of, like, almost take this transparency paper of, like, who's providing which services, like, Internet services at which rate to, like, these other companies.

And then because, like, we wanna move to, like, a a point where we are doing, like, really, really cool, like, you know, 0 downtime stateful failover, like, interactions with, like, any any perceptible latency, any perceptible issues, or anything else like that. Like, we're taking another transparency paper, and we're saying, like, where are dark fiber lines that we can actually, like, go in and tap into so that we can we can go and do, like, terabit, or, like, terabyte, plus, transfer rates, right, to, like, make sure that we can say, like, if this like, if a tsunami hits, like, this data center, like, you're totally fine. Like, there's no there's no issues or anything else like that. Right? And that provides a variety of benefits.

Right? Whether it's, like, image transfer sizes or anything else like that. And we've hooked it into our orchestrator so that, you know, and I think, like, this is the the thing that's, like, really interesting. It's, like, ultimately, the only thing that matters is that, like, your compute is close to your storage. Right?

Like, like, it it so, like, that's that's what matters because, like, your database queries. Right? If you have to go speed of light between these data centers, it takes way longer. Right? So, stories end up being a really interesting problem that we've, like, you know, we've, I would say, solved on the cloud stuff, and we're now solving on on the kind of, like, bare metal stuff.

Wow. Yeah. That that that is extremely interesting. And so who like, that seems like multiple people's, like, full time job, but you're now and we've already established you're doing a lot with, like, 25 people. How how the heck have you managed that?

So that is basically, I would say right now, probably, it was 2 people's full time job, our, head of operations, Christian, and our one of our most seasoned kinda infrastructure people, Charles. And so they've been kinda, like, mostly building this out. We're getting more people involved, you know, the person who built our orchestrator, Pierre, person who built our proxy, Miguel. And, yeah, that's another awesome person that we've brought on named Finn who's, like, the best like, one of the best networking people I've ever like met. And so, you know, we're pulling in more people to go in and build this stuff out.

Right? But it's still, like, one singular person basically, like, driving these DCs. Right? So there's a lot of different stuff that goes into it. But, yeah, it's it's it's been cool because we've just, like, we've continued to operate with Levered and say, like, you know, like, philosophically, how do we how do we attract really, really excellent talent.

Right? It's like, well, we give them really, really interesting problems. We, you know, give them the capacity to, like, push these forward and, like, give them the the space where they can build the best you know, like, do the best work of their life, right, for the lack of better word. Right? So, yeah, that's it's it's close to, like, 1 ish person doing doing these things.

Right? And there's a lot of stuff that gets involved. Right?

What kind of stuff have you learned about running data centers?

I mean, it's like hardware procurement is like the difficult thing is, like, actually, almost like procuring the hardware, if that makes sense. Right? Like, we've learned that, like, first of all, a lot of the systems on that side of the house are just, like, super, super archaic. Right? Like, if you want people to, like, go and rack your stuff, it's, like, you send them, like, an Excel sheet or a PDF of, like, here's how, like, you go and and wire these things together, basically.

Right? And so you have to you have to, like, punch in this format, basically. It's like you're doing punch cards, essentially, where you're just like, here, put these, like, servers here and then put these cables here and then plug in these, like, fiber lines here and then, like, put this, like, spine here, essentially. Right? And we're like, wow.

That's like a lot of paperwork that we, like, don't wanna go in and do. And so, Pierre has built this thing where it's like we've essentially kind of like virtualized the racks because we're like, well, we need to know where these things are anyways so that we can kind of like say the storage needs to be close to the compute. So we'll keep all your stuff together so that your your everything is super fast. We've, like, virtualized those racks. So it's, like, well, like, what if we just had, like, an output thing where you could just, like, add more stuff to this, like, virtual rack and then just say, like, here's the diff, go and and, like, send it to these people.

Right? And so, kinda taking that arcade format and, like, moving it a little bit more forward so that we're getting just, like, almost get to a point where it's like, how do we copy paste data centers? Right? So that's, like, one thing that we've learned, that's, I think, super neat.

That's so that's I feel like if someone's looking for startup ideas, they should just go and look at just go and talk to railway and see what they've built internally.

Yeah. I mean, the whole the whole space over there is, like you know? And I think there's, like, there's a little bit of arb in terms of, like, stepping into, just hard things, because people are like, oh my god. That's so hard. It's, like, really impossible to, like, go in and do that.

Right? And you just, like, go and look at it. You're like, wow. There's so much low hanging fruit. Like, yes, of course.

Like, a lot of these things are hard. Right? But it's like, oh my god. Like, a lot of this is like a kind of a bit of a shit show. How do we, like, go in and refine a bit of this process to make it kind of less of a mess?

Right? Because it's hard because it's, like, esoteric and difficult to navigate and, like, there's problems that, like, pop up and stuff like that over time. Right? But if you just keep solving those hard problems, right, like, you get to a point where you're actually like, oh, okay. Cool.

We can we can just you can just do things, you know, the meme or whatever.

Yeah. I I believe it about you guys. That's very cool. Okay. So I guess, is it like, do you see I guess you see yourself as, like, taking on AWS at some point, eventually.

Like, it's like to that scale where you're gonna build out this global global scale of, like, of, like, hardware, and then be able to do anything and and be able to just, like, run completely off AWS?

Yeah. So, I mean, this is an interesting question. Right? So, like, I guess, like, at the heat death of the universe, everything is competitive with everything else. And so, like, at some point, we'll be competing with AWS.

Right? But, like, they're so gigantic, and they built, like, so many cool, awesome, excellent things, right, that, like, we actually work with, like, companies on AWS. Right? We work with companies on GCP. We work with, like, companies, like, wherever.

Right? So we're not dogmatic about, like, where we wanna run this. Right? Like, again, we built our own orchestrator so you can just, like, curl the binary into a box, pair into the WireGuard network. Right?

Just like bada bing, it works. Right? But, like, we wanna make sure that we can offer the user the best possible experience. Right? And for us, like, there's only so much stuff you can go and do on the cloud.

Right? And so, like, you know, we've built, like, some in house, like, storage stuff that's allowing us to, at this point, like, start to roll out, like, copy on right pull request environments. Right? So, like, we already currently do, like, your full infrastructure is like a preview. Right?

But, like, you know, the database is empty or whatever. Right? And some people are like, oh, how do I, like, seed this with, like, data that I can actually use and stuff like that. Right? And so now we can actually give you, like

Oh, cool.

Full copy on right production environments. But if you make any changes, then, like, they just kind of exist in this parallel universe. You know, you can test against your staging data, etcetera. Right? But you're not gonna break anything for anybody else.

Then you merge, and we'll go and clean all those things up. That's only possible because we have our own DCs, and we've, like, built all of those things. Right? Like, being able to do, like, really, really fast image transfers. So, like, so you get, like, instantaneous, like, builds.

Right? Like, you can do it on cloud. Right? If you, like, get really, really good about, like, data proximity and all of those other things. Right?

And, like, you know, you can build your stuff beside where it's gonna go and and have to run. But you can also just, like, take advantage of of, like, you know, 100 gig lines that you happen to run between these DCs or, like, multi like terabit lines that, right, like, that you run between these DCs or stuff like that. Right? So, like, it allows you to unlock this kind of additional layer. Right?

And so, I mean, AWS is huge. There's so many other companies as well. Right? Like, we just wanna give people the best possible experience.

Yeah. That's it's really exciting, honestly. It's it's super cool. Mhmm.

And

there's so much there's so much opportunity, I think, to to do it and to do it well.

Yeah. I think so too.

I think AWS's revenue, someone told me was, like, the quarterly revenue was, like, 500,000,000,000 last quarter. That that sounded insane. Is that, like, is that true? Is it

I think it's in the I think it's in the 100 100ish 100. You know? It's it's generous. Right? Like, I mean Yeah.

I, you know, I think if we go back to, like, the the kinda, like, start of the interview and you're saying, like, oh, this is hard, and this is kinda, like, the problem to solve. It's, like, I I think that this is the the problem to solve. Right? Like, just from, like, a it's complicated. People really, really wanna work on it.

Like, people need it from, like, an ability to move forward perspective. The market is absolutely gigantic. Right? And so there's a massive amount of, like, leverage to be had here. Right?

Like, I I think that, like, if we do this right, like, there's almost this kind of, like, kinked toes right now, which is, like, how do I go from, like, idea to, like, thing that is running that everybody can go and access. Right? And if we do our job right, we kind of unkink that hose. Right? And a massive, massive amount of kind of, like, you know, call it GDP.

Right? If you wanna use, like, Stripe terms or whatever for the Internet, just kind of, like like, is unlocked. You know?

Yeah. Because you're just gonna enable all these people that were tied up with, you know, running Kubernetes, can suddenly just go back to shipping features and shipping shipping stuff.

Yeah. Exactly. Right? And, like, org level initiatives to, like, make developers faster by, like, you know, buying things like, I don't know, like, space lift to go and manage their Terraform and CubeSat or anything else like that. Right?

Like, there's this massive, massive machine of, like, how do I go from idea to production? In most organizations, it's it's kind of like, you know, use that leverage, quote. It's kind of almost like held up by, like, all of these, like, DevOps, SRE people, like, literally just doing God's work to just, like, go ahead and just make sure all of this stuff just

Hard work.

Stays up. It's so hard. It's so difficult. Right? And they built really awesome tools.

Like, you know, I I think that, like, you know, I I'll I'll rag on Kubernetes because I think, like, you know, it's it's it's kind of a lot of complexity and stuff like that. But, like, you know, it's a good tool. Like, it works, and it solves the problems that they need. Right? I just think that, like, if we can take a bunch of these, like, verticalized industries that people have done, like Terraform and CUBE and Ansible and all those other things, and you can kinda, like, merge them together, you can get a lot of outsized benefits.

Right? Even especially if you start to, like, look at things like observability, like Data dog. Right? If you start to look at, like, things like CDNs, like Cloudflare. Right?

You start consolidating a lot of these things. It's really, really it would be really, really great if you could just learn one tool and just apply it wherever.

Yeah. Yeah. You're not uncool.

No. I think that was, like, a amber alert or something. Thing. Tsunami warning. So topical.

Hopefully, our our DC is

Where are you?

In the I'm in San Francisco.

Wait. Where's the tsunami then?

I don't know. Let's see. You are in danger. Oh, okay. Interesting.

Well, I think this will be fine. You know? I'm a very laissez faire person. It's like, I'm sure it'll be fine. Okay.

So yeah. Well Yeah. Yeah. Yeah.

Okay.

If if there's a wave that crashes through, I'll be like, oh, sorry. I gotta go. But I'm on the I'm on the 3rd floor right now. So, presumably, I'll see water rolling around at at the the, like, you know, Geary Road, and I'll be like, oh, shit. I really have to go.

Thanks for thanks for having me. But, you know.

That's we got we got your lessons on what you've learned about data sets.

Exactly. Right? Like You can go go take the paper, like, carry on carry on the dream. You know? Yep.

Yeah. I I actually because I I'd done a lot of, like, mobile app stuff, and and then that kinda led me to doing, like, you know, lots of, like, Firebase. And I kind of, like, gradually got more and more now where I'm actually, like, touching more, like, what is properly infrastructure. And it's really amazing how long I think I I was, like, coding without really having to do all this stuff, and it's, like, it's kinda shocking, like, once you leave that kind of, like, happy path of, like, you can just do a CRUD build a CRUD app or whatever. Like, once you leave that happy path, like, how it it is surprising how much of a gap in, like, complexity you suddenly get.

It's like a chasm. Suddenly

Yes.

You leave that path, and it's just a chasm.

Yep. There's a massive, massive chasm there. Right? I I I call it kind of the DevOps trough of sorrow, because, like, you have to kind of move through this chasm of, like, figuring out what servers are, figuring out what a reverse proxy is, figuring out what Yeah. Like, what is a proxy versus a reverse proxy.

Right? Like, you have to figure out all of this. Right? And then you, like, go and essentially, like, apply it. You go and mess it up.

You go and figure out, like, ways to go and build these workflows, right, like, in such a way where, like, you're like, okay. They work for me. Right? And then you go, like, bring people on on your team, and you're like, wow. That doesn't work anymore.

Right? Like, I guess I need different workflows, and I need to replumb everything. Right? And then you have this kind of, like, you know, n squared complexity of, like, I have all these microservices and I have all these environments now, and I have all these, like, you know, all of this stuff I gotta, like, plumb together. Right?

And it pretty rapidly becomes like a full time job, which is why, like, it gets consolidated into those areas where you just, like, don't run into them until you really do. And then you're like, oh, I've, like, stepped almost, like, off this cliff into this, like, amount of stuff that I feel like go and pick up. Right? And so, you know, that was kind of, like, one of the main things where I was like, okay. Like, something has to exist here.

Right? Because you have this almost, like, happy past, etcetera, over here. We're like, I'm building my nice, like, React app, and, like, everything works or whatever. Right? And then you're like, what do I do when I wanna go host a back end?

And then, like, go see that guy down the street. It feels like you're walking down, like, a seedy alley, and you're like, what am I what am

I doing? Yeah.

Right? Like, you know?

So Craig's Craig's So,

like, it was just one of those things where it's like, you you need something here. Right? Like, something has to exist. Right? There needs to be, like, a lamp a lamppost and a light over here.

Right? So we're we're building that.

Yeah. That's super cool. So just before we wrap up, last couple questions. So the first one is if you had, like, one kind of big takeaway for other DevTools founders, what what would it be? If if it was like your the tsunami's hitting, and you're just like, the world is rising.

You've got one message for the dev tools founders.

Just do harder things. Like, hard things are almost like I think there's, like, a Sam Ullman quote, which is basically, like, risk is not as risk adjusted as, like, you think, basically. Right? And so, like, more risk is actually, like, less risky, especially early on. Right?

Like, do harder things. Right? Like, harder things are not actually as hard as you think that they are. Right? Especially when you get really, really good at, like, pairing down complexity.

Right? And I don't wanna diminish any of the stuff that we're doing. There's a lot of stuff that we're doing that's really, like, really, really, really difficult. Right? But the nice thing is is that when you have these difficult problems and you have these, like, audacious goals or anything else like that, turns out, like, people like working on hard problems.

They like working on interesting things. Right? Like, it's actually so much easier to, like, get people who are interested in, like, really, really core, you know, dev tools, etcetera, things than, like, you know, I don't disparage any company, but, like, you know, DocuSign. Like, some sort like, how do we send forms or anything else like that. Right?

Like, it's not that interesting. Yeah. Right? But if you have all of these hard problems, obviously, you have to go and find the people who who have the capacity and desire along with that desire to, like, go and do those things. Right?

But, you know, you can attract really, really awesome people to help you go in and and build your vision. Right? And so if you're like, oh my god. It's so hard. Like, how can I go and do it alone?

Like, first of all, there's always a clear path to, like, getting some really, really garbage version of the thing off the ground. Right? Like, first word version of railway that that that, like, I showed to people was, like, you can you can get a Postgres in the version, like, instantly. Right? And do you know why it was be instant?

Because it was the same Postgres version. It was the like, I just created new users for people. They would hit a button, and it would create a user and give you a password. Right? And, like, that was it.

Right? But people were like, oh, I really want this. Right? Because, like It's so fast. It's so fast.

Like, there's no provisioning. I can just, like, connect to it or whatever. Right? And then you end up with all of these, like, problems, like, you get serverless things connecting and, like, flooding your database and all of these other things. Right?

You're like, okay. Well, I actually have to solve all of these problems. Right? But doing that hard thing quickly and then figuring out if, like, there's actually a market for it. Right?

Like, I would say is the main thing. It's, like, find your 80 20, find your fast fast possible thing, figure out how to, like, get your first 100 people in, like, a Discord. Right? And, like, build something there, and then try and figure out how to, like, make your first dollar. Right?

And that's, like, I think the, like, the way to, like, cut and, like, do the the harder things is, like, just get really, really core to the thing that you're trying to go in and do, but, like, also really lean into those hard

things. Right? Yeah. That's that's really good lesson, I think. And the other one I wanted to ask you is if you have any dev tools that you're using right now that you're really excited about.

I mean I can't say.

Yeah. Well, I mean, the confidence that it was gonna be Claude, because I think Claude is just really like, it's it's good. There's a guy on on Twitter that I follow. His name is Nick. And I I think he worked at OpenAI or anything like that.

But he's basically saying, like, people are we're talking about AGI. People are talking about AGI and what it, like, looks like, and and it seems to more look like I just talked with my friend, like, Claude, like, 11 times a day. I'm like, oh, that's like so right. You know, it's like this this thing is like here and it, like, works, you know. Dev tools wise that I think that that are are solid.

I actually quite like Space Lift, as, like, a tool in general. I think that they're they're really good. And so if you're in, like, the DevOps Espace, I think they're they're great, and you have to manage those cube clusters.

Because what we do is generation space lift.

Well, it's not it's not like Terraform generation. It essentially, like, manages the application life cycle. Right? So, like, Terraform, you know, you run you apply it locally or anything else like that, kind of not a great idea because, like, you know, there's you need production permissions to, like, go and destroy these things. You go to manage I'm permissions, all this other stuff.

Right? So if you have, like, you know, something you wanna consolidate into a bastion somewhere or anything else like that that, like, runs essentially on PR environments or, like, on merge or anything else like that, like, this tool works quite well. It's, again, like, a bit complicated. But, like, we dog food, like, state of the art stuff internally. Like, we have a cube cluster, and we, like, have all of these.

And we've used service meshes in the past and, like, all of this other stuff. Right? So we dogfood some of our stuff, just to host, like, our web app. And so, like, this is a tool that, like, actually is, like, okay. This is pretty nice.

That actually works, you know, and does the thing. It's a bit of a pain in the ass to set up. You have to write, like, policies if you wanna do anything complicated. Right? So, like, rewriting Rigo.

But, again, Claude is actually really good at writing Rigo, so you can just do that.

Yeah. That's that's, it's it's funny how it just unlocks all, like, all these languages. I I I work for FFmpeg quite a lot, and it's just, like, so good at writing f of mpeg. And it's just Yeah. Suddenly, it's not a barrier.

Yeah. Exactly. Right? Because it's it's really, really good at compressing all of that knowledge together. Right?

And, like yeah. I think it'll be really interesting. I was talking with somebody a bit because, I mean, I live in SF. So, like, there's only 2 topics of conversation. It's where do you work and and what do you think about AI or whatever.

And so, like, I was talking to somebody the other day, and it was like, you're always I think you're always gonna have, like, a human in the loop because, like, by definition, the AI, like, is kind of, like, a fuzzy search. Right? So it'll always be 99, 99.9, etcetera, kind of thing. Right? So I'm actually pretty bullish on those, like, human in the loop kind of experiences.

Right? Like, Copilot, etcetera. Right? But, like, at the moment, you get to, like, AI for some random esoteric thing here that's gonna, like, go and solve your stuff, I'm just like, there's no way. Right?

Like, it's always gonna have, like, any issue or anything else like that. You need the the person there. So it's it's it's leverage. It's compression. It's all of those, like, lovely things.

You know?

Yeah. Yeah. That's that's amazing. Okay. I think we're at at time, Jake.

Thank you so much. I really, really enjoyed that. So thank you so much for your time.

Yeah. That was great. Likewise.

And people can, I was gonna say people can go to railway.com to see more?

Indeed. Yep. Just launched our new dot com recently, so that's excellent. We're now at railway.com, so check us out.

And, what what is your Twitter handle if anyone wants to fully

my Twitter handle is just Jake. So, like, the word just and then Jake, if you wanna go and have a look at that. That's pretty good as well. Yeah. Thanks, man.

Yeah. Amazing. Thanks, Jake, and thanks, everyone, for listening.

Thanks so much. Cheers. Bye.