The Exchanges

Q This is the equivalent of my mom test, right? Like, what have you done that has your solution to this?

A So internally we built a tool called Central Station that allows us to go in and aggregate all the context from all of our users. Um, so every piece of feedback, every piece of customer support, every single thing like that, uh, gets aggregated into, uh, what we call like clusters. If you have an incident brewing or like anything else like that, now we can go and determine how many users are affected, all of those other things, et cetera. And then we can actually break off a discussion based on that. And I think a lot of that is actually a lot more, more helpful and more correct in terms of, Instead of like having just these like long running channels where you're just like, which channel should I put this thing in, right? Like if you can dynamically aggregate that information and dynamically route it to the right person based on the context, right? We know, we know internally like these four people are pretty close on networking, right? And so if we see like, okay, we've got a networking thing, you can roughly like drill it down to like those four people, right? And if you're saying like, oh, okay, cool. It's actually with this part, you can just go and like look at the commits, right? And this is, like, no longer a manual process internally. Like, this is the whole point of why we built, if you go to, like, uh, station or help.railway.com, there's a whole reason we built this…

AI assessment note: “internally we built a tool called Central Station that allows us to go in and aggregate”

Q Yeah. Is there any point at which you started prioritizing Um, AI developments or agent development?

A I think like we've, so we've prioritized almost like agentic as like a, um, top of funnel thing. Um, and probably over the last six months, we've probably deeply, deeply prioritized like agentic as a, as a mechanism to go and build and deploy things. Uh, just because we, we believe fundamentally like the, the curve is so sheer and like that is the way that people are going to go and build and deploy software. Um, And it, it, it almost like fundamentally doesn't matter if it's, if like this is dot com or not, because we're all on the internet now anyways. Right. Um, and so if agents are gonna go and deploy a bunch of things and we had an inference wall at some point, then like at some point we will go in and fix those problems. But like that will be kind of the dominant species over the next like 10 years is, is we've moved from assembly to C to C++ to JavaScript to now like words. Right. And you're gonna need to be able to close that loop. Right. Um, but That's, that's where it goes, you know?

AI assessment note: “probably over the last six months, we've probably deeply, deeply prioritized like agentic”

Q How do you plan on the build out? Like, I mean, the growth chart, It's so vertical that like, you know, like, uh, are you usually a hundred percent utilization rate as soon as you're live with these tracks? Like how far ahead are you?

A Yeah. So, so, um, like we still maintain like cloud presence for like bursting essentially. And so what we can do is, um, you know, we work with AWS and GCP and a few of those, you know, other, other clouds, like we can just rent. And then the moment we kind of get space or power or whatever, you almost just like compact those off the cloud, right? Like we, cause we started on, on the clouds and then we built, There's a system to allow us to migrate to our own metal, and so there's nothing that says you can't just continually do that again, which is exactly what we do right now, right? And so we, we never want to be in a spot where essentially we are, um, you know, compute constrained, right? And at the start of the year, like, we actually got to a point where we were compute constrained because, uh, the one upstream provider that we were actually working with wasn't able to give us quota at the rate that we needed to, and the hardware was, like, slower, right? And so we had to do a bunch of different stuff. I spent a weekend rebuilding our entire, like, Network, like overlay essentially, so that we could straddle, uh, five different clouds.

AI assessment note: “we still maintain like cloud presence for like bursting essentially”

Q Is there like a balance that you ideally like 50, 50, 3030, 30, like what, what's the mental model that you use?

A There's no, there's no balance. There's like, you just, you just have to think about all these things and, and be obsessed with all of these things. Like whether it is being obsessed with like, How do people think about your product from a go-to-market perspective, or being obsessed from a perspective of, like, well, like, if I can make this change at the, like, kernel level, then I can make it so that the user's SSH connection never drops, right? Like, because that's what I want. Like, I want a universe in which I can go and, like, snapshot all these things, and it looks exactly like you would just kind of iterate on, on a VM, right? And I think you just have to be obsessed with all those things, like, at every layer of, of the stack. And I think that's what makes it easier for me. I think some people, like, They're obsessed with different portions of the, the kind of like journey custom, like the company, like whatever. Right. And I think that that's when you can get really, really good, almost like cohesion by like segmenting out these things. Right. And so, you know, in the elevator, I was talking about like, you know, you have a technical kind of like person, et cetera, and then you have the customer kind of like person in general. Right. Um, and I think like if you can segment those lines out really, really well, and you can be very, very clear about what your areas of ow…

AI assessment note: “There's no, there's no balance. There's like, you just, you just have to think”

Q You mentioned somebody brought it up. Do you see a world in which these things get automatically caught, not necessarily by your agent, but, like, your customer agent? You know what I mean? That, like, the, the cash and validation thing seems like a pretty easy thing to, to check if you know, to look for it?

A It's hard because then you almost need Well, for us to, like, determine it, like, we need almost, um, we'd have to hook in with, like, your observability infrastructure in general, right? This is, like, why we almost have the template loop on the platform is to be able to kind of roll those things out progressively, where you say, hey, listen, you know, I can roll this out to, like, Johnny Vibe Coder initially, right? Or I can push a shard and you can almost, like, uh, consume that at your own leisure and be like, oh, okay, I'm gonna update to this specific version, right? Or Have this kind of, like, roll out over a period of weeks where you're pushing a new version, and then it goes to, you know, 0.1% of people, one percent of people, early dot, like, whatever, and then rolls out all the way there, right? That's the kind of, like, non-deterministic version control that, that we've kind of, like, talked about earlier. So yeah, a hundred percent, right? And I do believe that, like, that's where most things should go, go towards, because I think, ultimately, most companies end up building that stage rollout system in, in-house, right? Um, and it's just the same Thing built again and again and again at every single one of these, these different companies. So there's a massive opportunity to consolidate a lot of like developer stack.

AI assessment note: “It's hard because then you almost need... we'd have to hook in with, like, your observability infrastructure”

Q Yeah. You only have 35 people, so I'm sure they're not all spending 10 K a month. What's kind of the distribution?

A I think I'm at about 25, uh, in, in general, um, and then we have some, you know, power users kind of all the, all the way down. I don't know, like, we came back and, or from, from the winter break, and I was basically like, if you're writing code by hand, you are doing this wrong, right? Like, they're code, the tools are good enough at this point, like, that you move extremely, extremely quickly, and like, yes, there are issues and pain points and all these other things, um, but, You should be reviewing the code that you are writing instead of trying to go in and write it by hand. Like all of those architectural patterns, all of those other things, like you're not, just don't like, you're not gonna throw them in the garbage or whatever. Actually, they matter more now than, than, than any other time, but you just shouldn't spend your time generating code that you would write. Like, if you know how to go in and write it, just like ask the agent to go in and write it and then reconcile it until it looks like you would have written it yourself. Right? And I think incidentally, like, People misconstrue, uh, my propensity to, like, push people towards agents for, like, hey, we're growing really, really fast, and we've had some kind of, like, bumps in real life. Like, they're not necessarily related, uh, in terms of that. Um, but I think people should really, really understand, like,…

AI assessment note: “we have some, you know, power users kind of all the, all the way down”