Railway adjusts pay and interviews as AI changes engineering work

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 28:55

   And then last night when we were talking about recruiting, we also talked about how comp has moved up and taken away that, you know, you're competing with labs and some of the most well-funded companies ever. So how have you thought about comp for a company of your scale? And yeah, how do you compete for talent in today's world? Yeah, so I think the like remote thing definitely helps us, right? In terms of like being able to get talent across the globe.

2. 29:20

   you know somebody else at the dinner last night was like yeah the the bay area talent area market is kind of cooked because like from the labs i don't necessarily disagree right but i think that like great people with absurd amount of leverage and agents whatever they are uh today like constitute leverage should continue to see their like salaries go up and up and up and i think like you know as as we go in and see more and more and more levels of like production it's you know it's it's completely absurd that like somebody like andrew uh you know moving from like meta like anthropic with like a 1.5 billion dollar pay package you know that like it's absurd right but so too was a acquiring instagram for a billion dollars or whatever right like it still ends up being a net positive ultimately there and you're just seeing this like enormous like there's no

3. 30:04

   ceiling for like how good somebody can possibly be with these machines especially right now because the frontier is so so jagged and you have this diffusion of like some people are running 50 agents and creating, you know, the worst thing ever that will just destroy your entire code base. And some people are running 50 agents and they are creating the equivalent of the Sistine Chapel, which just actually you should just bulldoze your entire code base and replace it kind of with that.

4. 30:29

   Right. And so it's just and it's it's tough to tell sometimes, too, with the whole thing. So I just think that like our approach for talent is like never be stingy with with talent. Right. And just look like, you know, I'm evaluating comp like, you know, every half is probably move in every quarter, right? Where it's just like, are we making sure that these folks are in the best possible situation, right? Like from both an equity ownership perspective, as well as like, even for your own employees, existing employees, you're, oh, yeah, yeah, especially like, and I think people forget that like all the time, right? Like you need to be able to make sure that like people are being retained. People are excited about the work that they're doing, all of these other things, right? Like I'm a big, big, big believer in like servant leadership, right? Like,

5. 31:06

   and that's what we see in player coaches too, right? Like where they're just like, all right, here's this, you know, pardon my French, dog shit work that I'm just going to go in and grab, right? And just make sure that you guys are able to go in and do those things. And then also set the expectation that like, you're gonna do those things really, really well, right? We're going to talk about your work, we're gonna push it in public, we're going to go and make sure to chase down users to be like, what is the feedback? Tell us the feedback, you know, and just making sure that that environment is not just like excellent from like a, you know, comp, whatever, but also fulfillment, right? Also, they're working on interesting problems, right? Like part of the reason to start the company and like you know it's calculus that goes on in my brain early on of like

6. 31:41

   do i want to run this company or do i not want to run this company is like there is so much interesting stuff like so much interesting stuff right like we're doing like crazy bare metal networking stuff we're doing like on the go-to-market side we're figuring out how to like incept the models to like like recommend us more and more and more and we're figuring out how to go in and teach the models to essentially put its like you know hand or tendril or whatever you want on the infrastructure that we've got, right?

7. 32:07

   Like they're all such interesting problems, right? So I think you just need to make sure that like, you know, in today's age, right? Especially like, you know, if we get to a point where like maybe AI is so good that people have to work less. Like if you want to be productive and you want to move, you know, society forward and have purpose, like you have to be really, really good at all those things.

8. 32:24

   Has AI changed how you hire engineers in terms of the process itself? Are you applying a different process to the point you were making? Like somebody could spin up 50 agents and destroy your code base, somebody could do the exact opposite. How are you testing for this? Yeah, so I think there was a step function where actually, you know, we almost blinked after a bunch of the AI stuff. You know, it was like really, really good.

9. 32:46

   And we realized we hadn't updated like our technical interview. And when was that actually just for... So that was like, you know, we like there was a couple moments, right? There was like moments in like 2024 where we're like, okay, like you can definitely get mileage out of this thing. And then there's like kind of like, you know, Opus 4.5 moment, which I think is like the one that most people are familiar with where it just like hits everybody like a ton of bricks and you're like, oh, this is crazy, right?

10. 33:07

    And so for us, that was like, you know, I had plans going into December about, okay, cool, we're gonna do enterprise sales and all of these other things. We just raised our round. This was December 25th. Yeah, December 25th, right? And I was like, okay, well, we got this agent roadmap. We got this enterprise roadmap. And I go to like, oh my, we do two weeks off at winter break. I come back from winter break and I'm like, yeah, the enterprise stuff is gonna have to wait.

11. 33:29

    You know, we're all in on these things. like, you know, we just have to move a lot of this stuff forward. And so I think to answer your question, it changes the ceiling, it changes the ambition, it like, as you as you kind of go through these steps, right? And the new like opening eye model Astra, I think it's so like, it's still another step forward. And it's so good, right? And so you have to consistently raise your bar for whatever the ambition is of the thing that you are doing, right? And so we look for that we always do, you know, in the technical interview, we will say, bring whatever you want ahead of time, You can do it cold in the interview if you want or whatever. If you're not like, I'm not going to prepare, you can spend an absurd amount of time preparing for it. It doesn't really matter to us. And we just like go through that prototype, that thing, that piece of work, et cetera, with you. And so I think we just have to consistently almost be like, oh, man, the models are so good, like what you can do in some period of time. We got to like move it further and further. Right. And then you have to actually like evaluate people on things inside their work. Right. Not just the, oh, my God, that's impressive or whatever. But and this was always the case. Right. It's like, are they using the right patterns? Are they like doing the right things?

12. 34:27

    like all of this kind of other stuff, right? So it's just a deep amount of scrutiny. Has accordingly all of this exploded your roadmap