Venture with Grace
00:37:47 · Austin Spiegel
[00:37:47] S1: Yeah. Mean, I think that's a decent comp, right? If you think about what Datadog is, the core part of it is this data platform for collecting telemetry, logs, traces from software. And then there's all of these different applications that different personas in a software business would want to use to understand, root cause analysis, uptime, costs, so on and so forth. So that's definitely very similar to what we do, right? Core data platform and then different ways for customers to interact with that data. To answer the second part of your question, I think the thing that you still have to ask yourself essentially is like, would you build CAD? Would you build a CRM? Ultimately, yes, one of our one of our AEs, like, built this kind of copilot CRM, but we ended up purchasing Salesforce, and it was honestly a good decision. I don't think... Don't know how many will say it was a good decision to purchase Salesforce, but the fact is is that Salesforce has the data model and the infrastructure that that, you know, it already exists. It's best in class. It's, like, best in class data model. Why why should we, like, go reinvent the wheel there? Also understanding that, like, maybe we want different ways to interact with that data, and that's what LLMs do for I think just the same thing applies here, which is we have built a platform for ingesting very high rate telemetry with minimal latency and storing it for incredibly long periods of time. Even with AI, that's not an easy problem to solve. Is that really going to make a difference to your business if someone else can already do it? And if you have the tools to customize the way in which you interact with that data, then is that solving the problems that you have? That's generally what my take on it is, and then I still think infrastructure is the mode. You know, do, like one thing that I've been telling the team recently is I think ultimately it's really the pace of execution and learning that will keep us ahead because LLMs can build anything and they're only getting better. So, this is why our Ford deployed engineering team is so important and this is why Mike's leadership is so important in engineering. It's really about how quickly can we go from insight at customer to deployed product that solves their problem, right?
00:40:08 · Austin Spiegel, Grace Gong
And if you can make that loop really, really, really good, then you're always going to stay one step ahead of the rest of the industry. [00:40:17] S0: What would you think is like the common mistakes that hardware companies typically make when they're building their own data infrastructure? And obviously you mentioned, it's not like technically everybody's strong suit to build something in house, but I guess what makes it so hard for other people to navigate this?
00:40:45 · Austin Spiegel
[00:40:45] S1: Yeah, good question. I mean, I think the biggest challenge is just underestimating scale, right? So if you start, I mean, is for more startups I'd say, but if you kind of start, early and you have a time series database you've deployed and you've got a visualization tool and maybe you've built kind of some wrapper software around that, it's hard to understand that when you're creating megabytes or gigabytes of telemetry that if your business is successful, which you're betting on your business being successful, then you're going to have multiple orders of magnitude for telemetry than that and it's quickly going to become this quagmire. So, I think that is one challenge. The second challenge I've seen is that a lot of, you know, just some businesses don't think about software as this living, breathing investment, an ongoing investment. They think about software like hardware where, I mean, obviously now businesses are iterating on hardware more quickly and changing it all the time. Like, each Falcon rocket that launches is different from the previous Falcon rocket that launches. But I think historically, is basically really long requirements phase, really long design phase, really long validation phase, and then let's go build several or hundreds or thousands or millions of this device. And if you take that same thought process and you apply it to software, you're not going to be able to... It's not going to scale, right? So inevitably your software is going to fall out of date, whether that's just security compliance. Maybe you want to make one change to the software you can't deploy it anymore because the deployment environment is no longer existing. Maybe you scale beyond what the platform is capable of doing, but none of the people that built the platform initially are there anymore, so they don't understand the intricacies of it. So I think that's also part of the problem, which is people don't take into consideration that this is actually an ongoing investment and it's not just, Okay, we built it and now we're done with it. We actually have to keep investing and I think people underestimate that.
00:43:06 · Grace Gong
[00:43:06] S0: Since I guess you described machines where 50% of the bill of materials is software. And so basically, assuming what you meant is for the hardware company, they're spending 50% of their budget onto software. And what would you say, as a software company yourself in the hardware space, how much would you say is you're spending on, I want to say software, but yeah, software or like, you know, I live in an AI land, everybody's spending like a majority amount, you know, GPUs and something like that. So like, how do you kind of like allocate your budget as a CEO for the company?
00:43:50 · Austin Spiegel
[00:43:50] S1: Yeah, I mean, I think we're fortunate in being a software business in that most of our expenditure is people, right? So that's definitely the most important and most expensive part of the business. And thinking about, okay, well, do we achieve our goals and working back from achieving our goals to figure out what headcount needs to be and who are the people we need to have a seat? That's a majority, I'd say, of the planning. Of course, there's also software spend, right? So, we use AWS for hosting. Obviously we use other providers as well, but AWS is our primary provider. We also use LLMs. That's another large expense and then like your kind of long tail of SaaS products. I think the big thing for me that I watch out for is certainly LLM spend. I'm still not... Want people... People use AI all the time. We have like 100% uptake in AI usage. We've got engineers using multiple agents. Something that I find kind of fascinating about this is it's always been really challenging to measure outcomes that engineers create and the engineering productivity, right? There's like how many different frameworks that try to figure out how to measure engineering productivity. And now we've just kind of turbocharged like the amount of work that engineers can do. And And we're also spending much more money per head because of that. And I still don't think we have a great way of actually measuring, like, is this meaningfully... What is the difference it's making? Like I 100% believe that AI is making a difference, but I think being able to measure that difference against the cost is still something that is really challenging and something that I'm definitely keeping a close eye on as we kind continue to consume these tools.
00:45:50 · Grace Gong
[00:45:50] S0: Let's say since you guys recently opened the asset department, so I'm sure it's for attracting a lot of top engineers. And if someone getting offers from SpaceX and like getting offer from you guys, like what would be your like pitch to them to like join you guys?