00:21:02 · Jonathan Brebner, Anurag Goel
[00:21:02] S0: Really? It's kind [00:21:03] S1: of reached a tipping point. Yeah. And because of that, as we talk to these customers, we find that new problems are emerging that Render is actually uniquely positioned to solve. So for example, something that we excel at is running long running compute workloads incredibly reliably, scalably for weeks, months at a time. And this is as opposed to some of the serverless architectures that you hear about where you can run something for five minutes, but that's very different from keeping something running for months. And Render has built this incredibly reliable platform that runs applications for as long as people want to run them for. And that has allowed us to attract a lot of AI workloads that are coming to us from places like Cloudflare, places like Vercel, and even AWS Lambda, because all of these places are built for ephemeral execution. Whereas when you call out to an agent, it could take a while. When you want to run some sort of processing using an open source model, that could take a while. So people are realizing that these platforms, while they might be flashy, they're not the right fit to run AI workloads, backends. And so we're seeing a big demand for that on Render. [00:22:29] S0: And in fact, even it's sort of a trend line, right? What is one of the benchmarks for, you know, agents and new models? Yeah, well, how long-
00:22:29 · Jonathan Brebner, Anurag Goel
[00:22:29] S0: And in fact, even it's sort of a trend line, right? What is one of the benchmarks for, you know, agents and new models? Yeah, well, how long- [00:22:38] S1: Right. Yeah. And so you're not going to be able to really run an agent and have it do what you want to do if cutting it off at five minutes, which is maybe the limit for your infrastructure provider. So because that's happening, now the folks on Render are saying, the customers, they're saying, Okay, it isn't just the long running stuff that we need. We actually need everything else around it. There is a need for a suite of products that form together an agentic runtime. And by that, I mean, okay, these long range workloads, but also this notion of workflows because agents, they have a series of steps that need to be executed in sequence and reliably, and each step could take really long and you need observability and you need concurrency limits and you need retries. And so you need an infrastructure primitive that allows agents to operate that way. But that's just one part of what people want us to build. They want us to build an AI gateway. They want us to build sandboxes. They want us to build a shared file system. And they want us to build these things because they need them as they build agents. And right now, because vendor doesn't have them, they have to go out to a bunch of point solutions, but they really don't want that. They want everything to be part of the same systems so you can get end to end observability, security, compliance, all of that, and it's obviously is one bill.
00:24:09 · Jonathan Brebner
[00:24:09] S0: Yeah. I I mean, is this a case where it's like just the demand, like, happened so quickly and was so clear that you're... That it's like, alright. Now you have to kinda prioritize which you wanna do. I mean, we we were talking a little bit earlier about, like, part of the difficulty here, this is... I mean, this is also true, you know, with the the, data science product that you created is, like, you have customers paying you for something else. And then suddenly there's this other thing where there's a bunch of demand for it. Like, maybe they fit together or maybe it's like they're complementary, but like maybe it's... You have to think about, okay, where do we wanna focus the company? Right? Are there ways that you've gone about sort of managing that like decision tree?
00:24:45 · Anurag Goel, Jonathan Brebner
[00:24:45] S1: Yeah, absolutely. So, I still think that building this is a bet. It's not a clear cut thing that will definitely succeed. We don't know how the demand is gonna change. And in fact, there are people at Render who are like, well, if you wanna make Render the best place for AI apps and agents, what about the AI bubble? What happens when it bursts? And my point when that comes up is, look, yeah, there might be a valuations bubble. There probably is, but real value is being created. We're all using AI every day and people are building genuinely useful apps using LLMs. So it's not like the crypto bubble, forgive me for saying anything about that, but certainly there is much more value that I see here that developers see, that application developers see. And so you could either, you know, put your head in the sand and say, Oh, we're just going to be the naysayers and we know that the bubble is going to burst. And at this point, that's actually just everyone's opinion. And we'll just be right at some point. Or you could say, you know what? The bubble, yeah, maybe there is one, maybe there isn't. But the rational thing to do right now is to assume that things will work out in some form. And people had these problems today, so why don't we solve them? And we'll see what happens when the bubble bursts. Yeah. And so that's, for us, a minus one to zero step. Yeah. For, okay, let's jump into this with our customers. Let's figure it out together. [00:26:19] S0: It's ultimately... It is still a bet.
00:26:19 · Jonathan Brebner, Anurag Goel
[00:26:19] S0: It's ultimately... It is still a bet. [00:26:21] S1: Yeah. It is still a bet because the the shape of the product is still very unclear. Yeah. You know, everyone's still trying to figure out what agentic memory means. How are you going to build an agent memory infra primitive if people are still trying to understand agent memory? [00:26:37] S0: Yeah. Know, being an SPC kinda gives you a... Like, I think a pretty good sense of the way that, like, everyone actually kind of has whiplash right now and everyone... Like, no one is able to keep... Right? We we get in people who were, like, at, you know, the the, like, major AI research labs, and they're like, oh, I can't keep up with it. Yeah. Like, right. I couldn't keep up with what was in my own company. Was... Right. Like... Yeah. And then you as a... Someone who is building for, like, people who are feeling bad, but are still, like, building, like, new products on top of this new technology, like, ultimately, yeah, you're not gonna have certainty about anything. No. Right? But you do have to try and build conviction in something. [00:27:11] S1: And I think you have to be really good at customer discovery. You have to be very, very close to your customers, but then you have to ask the right questions. You have to put your own ego to the side come, you know, surely you have a point of view about what should happen, but what customers tell you might completely change that, and you have to be okay with that. [00:27:36] S0: Anurag, I I appreciate you joining us. Of course. It was a pleasure catching up and
00:27:36 · Jonathan Brebner, Anurag Goel
[00:27:36] S0: Anurag, I I appreciate you joining us. Of course. It was a pleasure catching up and [00:27:41] S1: Thank you. [00:27:42] S0: And Likewise. Talking more about yours and [00:27:44] S1: and Render's journey. Thanks for joining Minus One. Thank you for having me.