All right. Thank you, David, for joining us today. I think that Sentry is a very iconic product, especially when we're at an infrastructure conference. But for maybe some of us that are not familiar, would you mind just giving a quick intro on you and Sentry?
Yeah. David Cramer. I started Sentry a long, long time ago. If you're not familiar with Sentry, thank you. We capture all the bugs in all the software in the world at this point. Pretty much every company uses us. We do a little bit more than that, but for better or worse, that's still most of the business. But very much like software observability, developer focused.
Awesome. So this name of the session is called Layers Not Revolutions. And I think that the kind of genesis of the name is that we kind of are talking about how with this new age of AI, all the layers and every piece of the stack is going to be replaced. But you've actually been quite both vocally, but I think I believe that you also believe that a lot of this is overhyped and you're healthfully skeptical about the durability of a lot of these new trends. So can you talk a little bit about what you believe will continue to be sustainable and what might just be a fad?
Yeah. I think every two weeks I'm told that we'll be out of business because there will be no more bugs and then I point them to LLMs and how there are infinite more bugs than there used to be. I used to tell people early, so Sentry's been around a long time, but like 2015 era when we were venture funding, I had to describe sort of my view of the world and my view of the world at the time because clouds, if anybody has forgotten, 10 years ago not every business was running on a cloud provider. It was still like you would have some that were like no cloud whatsoever. I don't know that that's true anymore. There's still some that run data centers, but if you go back in time, you're like, oh, for me it was very clear data center to cloud. But if you step back and you're like, well, does that mean data centers are less valuable or they make less money or if you were one of those networking people, your job's less valuable? No. It just meant the growth was plateauing of that sort of industry, which strangely now is coming back. And I think of everything the same. Like everything in technology is like a building block fundamentally. Like one new advancement doesn't really replace anything before it for the most part. At the very least it builds upon it and then sometimes it like evolves along the way. But when I look at that, like for example in our business, we capture a lot of logs and errors and traces and all these other things and you could argue well like if AI writes software for us maybe it doesn't write bugs then why does our business exist? I'm like probably doesn't but I also don't see that happening. And I think there's sort of a meta thing of like if that does happen, it's over a very long horizon and anybody that's stagnant and you can call out some three-letter company names that you can come up with, you're kind of always advancing with what the technology is, but again, it builds on each other. And so, I don't know, that's how I see things. But you can look for infinite prior art in the industry that very rarely does technology go away. And if it does, it's over a very long horizon.
Yeah. I think it's very interesting that you're both kind of on the one hand being healthfully skeptical but on the other hand you are tinkering on Twitter, you are kind of like trying out new products and also have built a couple really AI native products like Warden, Seer that are adjacent to Sentry. So how do you kind of keep both headspaces?
So we used to always say, like many of you, we have some random bad tagline for our company, I don't know what it is offhand, but it was something about we catch bugs or we find, I don't know. And we sort of evolved that like what if we could find and fix them, which we cannot to be clear today. But you can imagine what becomes possible when there is new technology. So one thing that's really exciting that we can do today that we couldn't do before is we can do root cause analysis with the click of a button, right? It's not cheap mind you, but that was like impossible to do before. If you've ever worked in big systems, like correlation of anomalous events was like awful, some mathematical cool thing and it was like flaky and then what you'd have, you'd have like these anomaly detectors and one would fire but then a hundred more would fire and so you're like I need an anomaly detector for the anomaly detectors. And what's cool now is like LLMs unlock a new set of functionalities that we can build on top of the thing, again it's like a building block. And so I always describe to people I'm like a very healthy skeptic of everything by default and even LLM I was at first to be fair, they were not great at first. But once you can see the value in something you can kind of like then unlock like well what could I do with that. And so you know I'm in the fortunate position where Sentry is a later stage company and I've hired some executives in so I get to write a lot of code these days and there are quite cool things that you can do but you get grounded very quickly in what the state-of-the-art is. And so for example, everybody's favorite subject lately is there's a lot of code that is being generated, it's not necessarily good. And code review is for example one of the big areas is that people like to play in now. And there literally it's like week to week the conversation changes, but like the conversation of like this week might as well be like, well, what's code review becoming? And even Anthropic, I think, launched code review today, which is absurdly expensive by the way, but also everything is absurdly expensive. And I think it's just interesting the way we create new problems and then choose to solve them. I forget what they, I'm not an academic. There's a theory or something where it's like the more compute we have, the more we'll consume.
Yeah. And it's not always like we've become more productive with that compute.
Yeah. And so you got to wonder is this going to be similar? It feels like it could be similar right now. So yeah, speaking of the subject of productivity, we kind of talked about in the back room there how you actually don't find that much productivity gain just yet from AI agents and to an extent that you know you haven't personally experienced or at least Sentry hasn't really made drastic moves on headcount because some of these impacts are still unknown. Like what are some of the heuristics that you have there?
Well we budgeted more for developer tools than we ever have in the history of our company. Which is not that much in the grand scheme of things. We added 15k per developer for cloud code which is not enough by the way. My spend is $200 to $300 a day. I'm on the higher end to be clear, but it's a lot. Do I get $200 or $300 a day in ROI? No. So, and by the way, it can't be one to one. I need like $3,000 a day in ROI for that spend to be justified. And so, I think we're still early on that. But I think using the technology, learning the technology, understanding the negatives and the positives, how it works, because it's not fundamentally changed that much since it first came to be and you really got to get familiar with it. So, I think that's really important. I'm very bullish at like from a headcount, like a lot of people I'm sure in this room as well, hiring is very hard right now and it's not because people are not hiring. Has nothing to do with that. It's probably because a lot of investors are giving them money to be fair. But there's been no change in our business from like a hiring mechanics, from a productivity mechanics. Like some people are doing more I think and we're starting to be able to optimize how people spend their time but it's very little here and there. A great example, our CFO is an avid Claude, Claude Code, Work now user and he was showing me the other day, he's like, I had, I think Hex we use now or we're trying or something and it's pretty good so far. But he's like, yeah, we wired Hex up to Claude Code something and I had it generate me this report of the last I don't know six weeks of signups or something. I don't even know if it was correct, but it was like free insight, right? Did we do anything with it? No. But you can imagine if we did and we didn't have to like go ask somebody on the data team to build a new thing, you know, all these little things that you can optimize out, but that's always technology. I always tell people like technology is only a productivity tool. Has no other purpose to exist at the end of the day. And so this is just another version of like could we be more efficient in X, Y, and Z kind of things. And we're just like obviously still super, super early in it. Engineering probably being the most mainstream version now, but even that you can see all the flaws of it.
Yeah, like to, because I think you strike me as someone that's like very opinionated about those technical flaws and potentially what, despite kind of the seeming rising adoption of AI agents and these coding agents and the kind of code quality they have, like from your perspective, like why do they suck?
If you've tried building something and you actually know what the result should look like, it is very hard to get it to the result that you know it needs to be. And so I always tell, like there's always this fun thing people like to say which is like oh like you know like some version of you don't need to be like a software engineer have programming experience to be able to build software. Now that is so wildly false. Like if any of us work in that field we know how false it is because yes you can build like a cool MVP or something, it doesn't work. You don't know how to fix it. You don't know what questions to ask. And I always tell people if you can't give the LLM the answer, it's not going to be able to give you it back. And because all it is is roughly just complicated pattern matching. And so I look at the outputs I have and they are so all over the place and there was a point where I cared more about time spent doing this and I would look at sort of my contributions. Some days I'd be like, wow, I got so much done. And it was usually a new project where I'd get a lot off the ground right away. I'd feel really good and then the next day I'd spend like eight hours trying to prompt the thing to do the right thing and it wouldn't ever kind of thing and it's way too easy to get lost in that like the dopamine hits of that and it's not productive by any means. But I will argue that I think that the genuine gain so far in engineering is getting a project off the ground is like so easy now. But I at least personally and most of my peers who've worked with me over the years, maybe less so at Sentry, but know that I am like a, I'm pretty high throughput when it comes to engineering. And I look at, I actually looked at my contribution graph on GitHub about 10 years ago when I was Sentry prime if you will to today and you see two mountains but the current mountain is much smaller than the old, much smaller. It's like 20, 30% smaller. And I will tell you the code is definitely not better. And there's not necessarily more of it and so it's interesting. It's just like one data point, but I find it very hard to build good software with LLMs, but I find it very nice that it removes the barrier to start things, you know.
Yeah. One very interesting data that I recently found actually, there's this website called basically What Does Claude Choose? Basically it shows a default stack if you just like give Claude code like oh choose like an observability tool and logging tool and actually Sentry was like the default stack for observability and logging which is obviously amazingly positioned for these new class of potentially vibe coders or YC founders that are trying to iterate really fast. Do you see like their use cases and like this, you know, problems that they encounter as someone that potentially is vibe coding their first product and are plugging into Sentry potentially different from all the other companies and existing potentially enterprises that you have?
Yeah. 100%. Like I mean you know I don't think I could, I've yet to like meet somebody who truly just vibe codes and doesn't know enough to be dangerous I guess if you will. But like again, if you don't know, my opinion of what a senior engineer is is somebody that can theory craft really well, there's this kind of problem going on. Oh, I know what that is. I know the shape of that. And then I can ask questions. I can ask them super effectively now with an LLM, but I still knew what to ask. And the problem is like if you don't know what to ask, where do you start? Oh, it's, I got a, like I launch a website or something. This one's like a real problem people I'm sure have all the time. It's Next.js and there's these complicated ways Next.js works and maybe you just have a white page. It's just like a blank page. My page is blank. What do you do? You know, like well what's great is you need logging and then the agent needs the logging. So, you know, we're safe. But like you don't know where to start and the agents are like they're just pattern matching. So if it can't get the right data with the right answers, it's not going to be able to solve a problem for you. And so then you think about that it's like, well, is there an evolution of Sentry or completely different product to be clear that maybe isn't even us that caters more towards those types of people. And I always think of it like, well, if you're on Squarespace and something's broken in your website, you're not installing Sentry. It made no sense. Do you now install Sentry if you're using like V0 or Bolt or something? I don't know. But the cool thing that also can build on top is you can actually make products easier to reason about as long as you can afford the compute. And so maybe there's a world where it does bridge a little bit more and sort of entry level software gets easier and it's unclear about the other side of the spectrum, but I don't know yet. Like we'll see. I'm skeptical that, you know, the flower shop down the street is going to start vibe coding their own e-commerce app. I think they're just going to go buy Shopify and not care, personally. But we'll see.
Yeah. Given that you, like going back to some of these experiments that you've built with like code review, with potentially this like one-click kind of getting insights from your logs. How do you contain maybe some of these experiments that potentially could or could not be part of the Sentry product? Like what are some benefits of keeping those separate versus like potentially integrating immediately? Yeah, how do you think about that?
It's hard. I had to send around a thing at the company and be like, hey, I built an open source competitor. Technically, it's not open source. I built a source available competitor to our paid product and you should be okay with it and here's why. I don't know if anybody agreed with that. Here's why was like if I could build it in a couple days, it's not defensible, so it doesn't matter. That was roughly my explanation. And it's not trying to be a one-to-one competitor, but it roughly it really is if you squint. And I think that's very true is like you can monetize anything under the sun. You can at least attempt, but is that a good spend of time? And like it's really easy to think like fully in a business kind of lens. Whereas like my career was always like public code, open source, all this stuff. And developer tools especially are a lot of that. And now it's a little bit weird these days because a lot of it's not open source. I'm pretty sure I built the first public like at least semi-open source code review tool, which is like shocking to me that that is true. Especially since there's dozens and dozens of businesses and people don't like paying money for third parties. But I think that's like very real and we always think about this because like the existential crisis of like, you know, are you going to be obsolete, especially when you're a later stage company? It's like, well, if somebody could build something in a couple weeks, we can too. And so it's fine. But I do think there is a lot to be said about you need to experiment a lot right now and you need to be thoughtful because like software isn't free. It still has a massive maintenance burden. And I feel this all the time because I have too many active projects. But it is, you know, the way you solve a lot of that is the same as before. Like our version, my version especially was always like build the software, if it's valuable, it'll become obvious and then you maintain it and if it's not, cool, you can shove it at some point. And so even things like Warden, which is like this code review sort of code review thing, it's been valuable for us. It caught a ton of security vulnerabilities like completely hands-free effectively. And that's great. If somebody builds something better in like a few months, we don't actually care. We'll just throw that away and use whatever the better thing is, whether it's ours or a paid service or something else. But the learnings we got along the way that we might be able to apply to other things we do are like really, really valuable. And as an example, we took that and we benchmarked our own product with it. And unfortunately the benchmark was not useful but I think the most important thing I can tell anybody and because I'm in the fortunate position where I create budget and remove barriers at our company, people have to be using the technology and they have to use it as much as they can right now in the sense of like you can't have like guard rails on a lot of this. But at the same time, a lot of the way you approach your problems has not changed significantly. And even here at like Daytona's event, it's like sandboxes aren't a new idea. The value of them has changed. I would argue the value of them is a lot higher than it ever was before. And I think that's going to be true for a lot of businesses.
