Adam Jacob: Welcome to Infrastructure Frontiers, where automation experts run down all the AI news for infrastructure engineers. It's Monday, August 3rd, so grab a cup of coffee and let's get started. I'm Adam Jacob, the CEO of Swamp Club. I'm joined this week by my Swamp co-founders, Paul Stack. Paul Stack: Hello, how's it going? Adam Jacob: And Nick Steinmates. Nick: What's up? Adam Jacob: Our special guest is Sean Escriva, Infrastructure Engineer Extraordinaire. Say hello, Sean. Sean: Hello, glad to be here. Adam Jacob: All right, let's run down the news. let's start with Dan Lorenz. So Dan posted to LinkedIn that they're phasing code review out at Chain Guard, and they're moving over to design doc review instead. So rather than actually looking at people's code, instead they've built a system where you basically submit your plans for what you want to build, those plans get reviewed by the team, and then and then that becomes the thing that gets approved and then implemented with agents and then no one actually reads the the code on the way out the door. Paul, what do you think about Dan phasing out code review at Chain Guard? Paul Stack: So I I'm a big fan of getting rid of code reviews. Like I mean, we've got rid of them quite a while ago. We're incredibly happy with the outcome that we have. plan reviews, not so sold on, to be totally honest with you. Plan is point in time, right? Like that's the thing. And that plan's gonna change. Like regardless of what the the agent tells you what the plan is, there's gonna be drifts, there's gonna be like changes in what it's doing, so it's gonna be painful. you're gonna have to read everybody in in the context. I think it's actually just shifting the pain to another point, to be totally honest with you. And I I don't know. I I mean fair play to them. Like I l I love that they're they're embracing a new way of doing it. I have a feeling that it'll change quite fast. Adam Jacob: Yeah. I mean I feel like we tried we Paul Stack: So Adam Jacob: I feel like we tried that early on at Swamp and it lasted maybe a week where we were trying to review each other's plans. Paul Stack: Yeah, I'm like we we we were like pretty judicious and we like posted them onto GitHub issues and then we went, Yeah, looks great The same way as when you open the pull request, you went, Yeah, looks great. Yeah. So, you know, like I love that Adam Jacob: Yeah, yeah, we just Yeah. Same way we did with code reviews. Yeah. Paul Stack: they're embracing it. I and I love that they're like like trying to shift left, but I I have a feeling that they'll they'll follow through with something else pretty pretty fast after. Adam Jacob: Yeah, I think they're I think they got six months till they stopped doing that too. let's do let's talk about stacked PRs on GitHub. GitHub released stacked pull requests. I feel like I've worked with a bunch of engineers for the last couple of years who would have loved that it crawled over broken glass for stacked pull requests. I feel like at this point though, it's too little too late, you know? Like at this point I'm not doing code reviews and haven't for months and hope to never do them ever again. So why do I care if my like PRs have been stacked appropriately or whatever. Sean, do you have a thought about stacked PRs on GitHub? Sean: yeah, I think it's too little too late. Same thing. I I was one of those who would have loved it. I was judicious for years about the smallest possible pull request in a reasonable order and chaining them together. And I I just don't think it matters anymore. so Paul Stack: It's kind of funny, like the the Git Butler people have been doing this for a while now. Like they've had stacked PRs in their product. Like Sh Scott Shakon, like one of the the creators of it, like implemented this pretty well and he's he's tweeted a lot over the last couple of days of like saying, you know, if you like this implementation of GitHub, you should come and check out Git Butler. but it's it's really a case of, you know, it's it's an interesting feature that people asked for ten years ago. Adam Jacob: Totally. Nick: Yeah, and I'll just, you know, pile on a little bit. You know, they tweeted today, something to the effect of now that we've shipped stack PRs, what's what should we ship next? And you know, my feedback, everyone's feedback, is how about we keep the fucking platform stable? How about we start with that? How about we don't Yeah, how Sean: Exactly. Paul Stack: Literal quote. Literal quote is exactly what it was. Adam Jacob: Ouch. key brings the guns. Nick: How about you can't, Adam Jacob: Yeah, for sure. Nick: you know, set a watch to when there's gonna be another outage this week, you know? how about we do that? So I guess stacked Sean: Did we make GitHub actions actually good? Adam Jacob: Yeah. Yeah, let's have GitHub actions actually work totally. Nick: PRs is easier, I don't know. So anyway. Paul Stack: Yeah. Adam Jacob: All right, OpenAI reduced the price of five, six Luna by 80% and Terra by 20%. they wrote a blog post where they sort of run down the technology. They sort of allowed them to do that. I think what I think is interesting about this blog post and just this move is that there's this prevailing understanding amongst a certain cohort of people that what's gonna happen is we're all gonna get addicted to LLMs, we're all gonna get addicted to these coding agents, we'll lose our ability to do any free thinking thought. And then they'll just spike the prices on us as if like the whole thing's gone wrong. I think it's interesting that like what's actually happening is pricing pressure is pushing it down. And and you know, I think there's arguments that people make about, well, they couldn't possibly be turning a profit or whatever. I suspect they probably are, certainly on the on the on the cost of individual inference. I mean, we'll know soon enough when they become public companies. But like I think the I think that combination of Frontier intelligence like five six Luna is as good as the models that we saw changing our lives six, eight months ago, you know? and it's faster and it's and it's cheaper. Keeb, you got thoughts? Nick: Yeah, man. I think we're at the era of choosing the right model for the right task. I think there's going to be a lot of intelligence surrounding that at like the routing layer up top. You kind of see that with some of the models today. You know, and I think there's a real benefit to, you know, what I read in that is not only the reduction in price, but the massive increase in tokens per second that you get on the outside. And, you know. Frontier intelligence costs a lot of wall time. and sometimes you don't need it for simple evaluations or something like that. And so yeah, being able to encode that at the right layer, I think is is the interesting part. And I can see immediately how we're gonna take advantage of this. Adam Jacob: Totally. Paul Stack: Yeah, and like there's a ton of people who are like token crazy. Like genuinely. They they don't give a fuck. Like they're just spending tokens like a like an absolute lunacy. And then you have people who are really efficient. Like we're really efficient. Like we're incredibly efficient. Like for my two hundred bucks a month with Claude Code, I spent eleven billion tokens last month. Right. I'm okay if that price goes up because the the the productivity I got in that eleven billion code tokens was actually pretty great. It's worth seven thousand dollars. still cheaper than an engineer, you know. So even if even if it goes up a little bit, it's okay. But I think overall that there's a lot of like profit in in fairness at this moment. There's people buying Cloud Max Pro Prans that could probably get away with like a 20 bucks a month plan. And they're like, you know, they're that that's profit. So I I I Adam Jacob: Yeah. Paul Stack: think it's the bubble isn't as big as people think at this moment in time with tokens. Sean: I think I agree. Adam Jacob: All right, Paul, you wrote yourself a blog post about how you know, best practices come and go, but architecture is forever. Give us the like give us the ten second summary. Paul Stack: like every month there's a new term of something you have to do in AI. Like it went vibes, it went spec engineering, it went agentic like loops, it went into graphs, like graphs is a thing now, like all of a sudden as the open claw dude said recently. my the the point of my article is is that good engineering practices, good architecture, like will survive all of that. And if you're lean enough you can keep changing the methodology that sits on top, but you still have those same guardrails, those same architecture principles. And you can beat nimble and and move around. Adam Jacob: Sean, I feel like you've been doing exactly that now for months. Sean: I've been trying. How well is is a question. But yeah, that's exactly what I think we need to be doing because most of us come into this space already, we hope, with good system thinking and good understanding of architecture practices that have worked well and we don't need to throw that all out just because now we can ask a simple question and let and spend a bunch of tokens to have something else do our work for us. So I really feel like some one of the things I liked in the post actually was how execution stays reproducible. Like that phrase kinda stood out to me and I feel like that's one of the things I've really loved so far in my experiments of getting to that place where you do pull inference out and you just have good architectural patterns that are reproducible. And it's just code and it just works like you expect every time. Adam Jacob: Yeah, right? Sick. Okay. OpenAI released Codex Security for automatically finding, validating, and fixing security vulnerabilities. I think this one's interesting because the the storyline that's been going around about frontier models finding security vulnerabilities, jailbreak escapes, sort of layer upon layer, you know, open AI had that had the storyline of compromising hugging face, right? and I think I think it's interesting to think about how do we how does that market evolve into something where we're building agent harnesses whose job is specifically to sort of find and validate and fix security vulnerabilities? Do we think that this is gonna lead to like a broader category of specialist AI tooling specifically around security that we're trying to tune? Or do we think it's just whatever, it's an easy way to like get a security dollar in the door? Paul, what do you think? Paul Stack: Me me I think is the second one. I I mean, there's no coincidence that last week there was a security incident where they jailbreak and like breach the a a system and then this week they they launch a security product, right? Yeah, absolutely. I love a bit of conspiracy, right? Adam Jacob: we're going full conspiracy theory. Marketing stunt. Paul Stack: But like you know, at at the same time, like there there are gonna be specialized models that are well trained for this type of thing. Like I already know people working in this space within those companies, but I don't know if like they'll be specialist tools because i if you have to bring in another tool into the tool chain it's just gonna feel painful. Nick: if you know this, but they hacked one machine autonomously, but you know, anthropic they hacked three, so they're three times better, by the way. Paul Stack: Yeah. Adam Jacob: Well and and and also today, the Tailscale folks wrote a blog post about how Tailscale was involved in that vulnerability because after it had breached Hugging Face and it found the security vault, it found the Tailscale token, which then it spawned a bunch of sub agents and connected them to the tailnet. Which, you know, super fun. and Nick: So smart. It's so smart. Paul Stack: Fan hot. Adam Jacob: also just in case anybody's wondering, it's not like this happened accidentally. Like OpenAI told the thing to do whatever it could had to do in order to solve this problem and it could go and do whatever it wanted and it did. It's not like someone didn't prompt it to do this craziness. You know what I mean? Like. Nick: My f my favorite thing about that is nothing novel was invented. This is just Sean: Mm-mm. Nick: an agent using every tool as its disposal. And so really it's just as much a hardening story as it is, you know, a novel use of technology story, right? Like Sean: When I feel like when when your intent is ten thousand miles wide, don't be surprised that it just goes off and finds the right path through that forest to to do something interesting. Like that's just kinda how it works. Adam Jacob: Totally. All right, speaking of Paths Through the Forest, Kimmy K three is up on Hugging Face. It has Frontier Intelligence. That's caused a whole bunch of drama about, you know, how did Moonshot actually get a model that is that effective? similar sort of co joined story, you know, Dario, CEO of Anthropic. Posted a blog post talking about their position on open weight models, basically trying to say, well, we like open weight models. We see why that's useful. The real problem is distillation attacks against American ingenuity. And that, like the fact that, you know, the that these Chinese models are running sophisticated distillation attacks in order to catch up to frontier intelligence is a threat to like American hegemony and sovereignty. How are we feeling? about Dario's position on open weight models and Kimi K3 being up there with Frontier Intelligence. You wanna take it, Keebe? Nick: Yeah, I mean what's what's amazing is how do these models get get as good as they are? It's through distillation, you know, like that's literally how it happens. and I think you know well yeah, and then they they go and they do post training, which is a distillation Adam Jacob: Right. They distilled all the world's knowledge in order to get us what we have now. Yeah. Sean: Exactly. Nick: of the model targeted to a specific task, right? So like that's why they're so good at coding. And you know, the What I think is is that ultimately, you know, what this proves is that no one model is gonna rule them all at the end of the day. Really it is attention, as they say at the end of the day. So like the fact that someone can build a cheaper model based on some distilled input means that open source is here to stay forever. Basically. Like you're never gonna be able to block this block this. Pandora's box has been opened. And so the question is just like how do you maximally exploit that for everyone's benefit, you know, acr across the whole show. Adam Jacob: Do do we have no concerns at all about like national security, any of those sorts of things? Are we like not everyone on this call is American, so you know, like Paul Stack: Hey, take my data, like do what you need to. Like I have Nothin sensitive in my in my system. I don't mind. Like just Adam Jacob: Ha. Paul Stack: just keep my subscription at two hundred bucks a month and it you can have Nick: Mm. Paul Stack: it. Adam Jacob: Yeah. Thank you. Thank you. Thank you, government basilisk. Okay. you know, in a similar vein, Toby Naup I hope I said his name right. If I didn't, I've met Toby before. So if I said it wrong, Toby, I'm sorry. He was the co-founder of Mesosphere. and he had a post this week saying that open weight AI is is having its Kubernetes moment that sort of with that launch of of open weight models with frontier intelligence, that this is the time where the whole industry shifts toward no longer focusing in on sort of best of breed, you know, vertical houses like OpenAI and Anthropic. And instead we're all going to shift to building, you know, the the open platform collaboratively together. you know, I kind of felt like this was a a reach in terms of both what kind of collaboration we're talking about here and and the sort of what happened with Kubernetes in the scheduler space. You know, like, I as much as anyone love to find examples of things that have happened in the past and sort of replay them into it. But I have trouble seeing how what's gonna happen is, you know, we Moonshot launches Kimmy K three and then what we all decide to do is turn Kimmy into Kubernetes. You know? Like it just that's such a huge leap to me. Sean, what do you think? Sean: I would agree. I mean, I'm actually really excited for the open weight models direction as well. Even though I don't think we're there yet, the idea of getting to self hosting something like Kimmy K three, whatever the future generation of it is, if it's K four or something else, like sounds really great. I don't think it's a Kubernetes moment either. I think the the foundational basics of Kubernetes were so much more generic and in the open at the time for me that it made it easier for other things to come along and imitate it. And then it led it sort of was not as large of a leap for operations Adam Jacob: Mm-hmm. Sean: from what we were used to doing. to go from, I have this EC2 instance that I can spin up so fast and need to orchestrate around it. And then I have this Docker container that I need to orchestrate around it and then automate and then okay, these thing platforms kind of do this thing for me. And the the leap to what the frontier models are actually doing feels so much different for me than that right now that I don't think just having an open weight model out there like K3 means now we're just generalized to this abstraction that's consistent everywhere. Adam Jacob: Totally. Paul Stack: I think if this if this brings the same complexity to run in models as it does to run in software and Kubernetes, I think it's like a it's gonna be pain. Like I mean, my god. Like you know what I mean? And I Adam Jacob: Ha. Paul Stack: Jesus, I don't wish that on anyone. Sean: Yeah. Paul Stack: Like Adam Jacob: Yeah. Paul Stack: Don't get me wrong. Like I I understand that the fact that a lot of these larger companies, it's gonna give way and people are gonna start like looking to use their own com compute to do it. But if you can't do it in a way where it's actually scalable and maintainable, then it just like I i is it even worth it? Like really. I I i call me in six months when I can when I can have the hardware to run a model locally and still get the same results as I do right now. Sean: Yeah, and like Adam Jacob: Yeah. Sean: have you looked at all of what it takes to like deploy Kimi K three, for example, on SageMaker with Hyperpod and NEKS? And it's just like this this complexity death spiral right now that is like Paul Stack: You know what you should do? Just just create like a a Kubernetes manifest and just deploy it and it'll work, right? Sean: Yeah. Adam Jacob: A home chart. Nick: I th I think it's the wrong layer, you know. The AI will have its Kubernetes moment when, you know, everyone tries to d to politicize a harness that is opened by everyone, you know? I think that's gonna be Adam Jacob: Man, I think AI already had its Kubernetes moment. I think OpenAI and Anthropic are literally the Kubernetes moment. It turns out that API is basically identical, you know? Like Paul Stack: But the AI foundation will will will solve it for us, right? That's that's that's their movement. Nick: Yeah, you got it. They're gonna create a yeah, they're gonna create a harness on top, right? So Adam Jacob: it's a spicy podcast. Yeah. Of course they will. Okay. Let's talk about this one's a little self serving, but Keith Townsend, I was on his podcast last week. he's the CTO advisor. For people who don't know, he's one of the longest sort of interpr independent enterprise architects, CTOs, kind of advisor to the to many large enterprises in in that game. he wrote a good blog post today about Deterministic AI being an architecture problem, not a model problem. And, you know, certainly for me, this feels really true. Where I think the beginnings of this journey for everybody, me included, was that you were going to see the frontier models continue to improve. And because they were improving, they were just going to get so good that it didn't that that you wouldn't have to think about asking them to use deterministic systems or to build a system around them that the models would just sort of do it. And I kind of wound up referring to that as the idea of like the God model. You know, where you can just throw whatever you want at the God model and the God model will sort it all out. And I think what we've seen is that that there's this emerging architecture of trying to think through which pieces of the system need to do what and how do I build the most effective and efficient machine in order to get the outcomes that I'm looking for. And that's an architecture challenge, right? It's just like it has always been a software architecture challenge, right? When we think about this is the problem I want to solve, we think about what the architecture would be that would help us solve it. You know, so we can scale the way we need to, so we're secure the way we wanna be, so that we can deploy it the way we want to, so we can manage it over time. I think those same questions sort of come into AI. Sean, do you have thoughts? Sean: yeah. I mean obviously I agree. You don't you're not gonna get determinism out of a randomized system, which and the frontier models are probabilistic and just searching for the best thing, right? You're never gonna get consistency out that. and trying to do that is just such a losing battle. If you think of it as a God model for those who believe in that that direction, like you st it's still beneficial to be specific with your prayers. Like and so so Adam Jacob: Totally. Absolutely. Sean: S so you're gonna get better results, you're gonna have your faith strengthened, you're gonna go a certain direction, like like like whatever that is for you, right? I Adam Jacob: Yes. Sean: i I think that's the thing that you just it isn't it isn't a model problem where eventually these things just get so good they're gonna suddenly behave consistently. Yeah. Adam Jacob: Yes. Yeah, I think that's right. Okay. Our last story for this week is Mitchell Hashimoto started a new company. Mitchell, of course, of Terraform fame. one of the one of the greatest of all time in the in the starting infrastructure companies and writing infrastructure code, automation game. back at it again. His company is now called Super Logical. And Superlogical is trying to build a multiplexer for all work. He got together a pretty he stole the design team from Poolside. I'm sure Jason Warner has miffed about that somewhere. but they were great designers, Poolside's designers were. And so yeah, when he talks about a multiplexer for all work, I think it's an interesting idea because one of the things he did was hopped on a video on X where he sort of explained why existing terminal multiplexers weren't quite enough and And he talks about sort of the double duty that they play, where every time you go through a multiplexer, they have to process the entire terminal. Then your terminal gets a chance to do something with it. And that inefficiency actually causes a bunch of problems in how we think about what we can do and how we can operate and sort of how you could rethink how terminals could work at all. and I think they have a bigger vision for how they're gonna then expand beyond terminal multiplexing to thinking about all of the different streams of work that come in. Thinking of that like a terminal, which then gets you sort of this multiplexer vision of the world. so of course, I'm stoked to see Mitchell back in the game. I think Mitchell's a good dude and has obviously done great work over the years. Paul worked at Terraform. So, like, you know, you owe a little something to to to Mitchell at some point. So, like, how do we all feel about super logical? you wanna go, Nick? Nick: Yeah, man. listen, I feel like I'm right in the target demographic for something like this as someone with seventy Alacrity open instances open right now. you know, like literally. Anyone who's seen my desktop knows it's a fucking mess. yeah, it's gonna be interesting. You know, I I think in this era of, you know, Paul, this one is gonna be like I think right in your wheelhouse, right? In this era where you can build you literally the exact harness that you want, you know, like in the way that you work. It's gonna be interesting to see how that how those two worlds coincide, you know? you can build the tool that you want, bespoke for you. And so I don't know how someone else's opinion on how you should do the work, you know, matters in this age. Maybe it does. We'll find out. Paul Stack: I built my own tool, like specifically what Nick's saying here. And I I and it's not because I I'm brilliant at doing stuff like that. I literally opened an agent and I said, Hey, I work exactly this way with Claude. I use like Claude work trees everywhere. I want keyboard shortcuts. ITERM is a little like too too much. There's like not enough like very specific features that I want. And I just like lent in and like iterated to the fact that, you know, it saves like the sessions. It uses it it uses Tmux server in the background. I can close it, I can reopen it. There is a bunch of tools that have spawned in the last couple of months where people are putting their own flavor of this. So, like, you know, it's definitely an an ecosystem that's wide open. Like it's definitely something that people are really like want to get something in, especially as, you know, we're kind of heading towards the death of the IDE. You know, and and as Adam Jacob: For sure. Paul Stack: you know, that's and I think that's really like there there's a a space that's like kind of wide open. So, you know, he he's a like Mitchell's smart, fucking smart. You know, he like look at how he started of the the projects, right? He started Vagrant, started Terraform, started Packer, and then look at what those projects became, look at what HashiCorp became. So you know he definitely has has great ideas and there will be a lot of eyes on it just because of the fact that he is like he's had such a track record of building stuff like this. Adam Jacob: Yeah, and let's just say it. Mitchell has Mitchell has tastes that people like. You know, like I sometimes disagree with Mitchell's taste. Like I it's actually not everything he builds is to my particular taste. But boy, there's a lot of people where his taste is theirs. You know? Like it just aligns. Yeah. Paul Stack: I still know people using Vagrant today, like in Sean: I Paul Stack: twenty twenty six. You know what I mean? Sean: I do too. I do too. Paul Stack: And like he built that like so early on and it grew and grew and grew and it's still around all these years later just 'cause some people just absolutely love it. So yeah. He has a track Adam Jacob: Love it. What do you think, Sean? Paul Stack: record in it. Sean: Yeah, I I would agree on his taste. Like I remember having a conversation with him actually at the very first chef community conference about Adam Jacob: Totally. Sean: vagrant stuff and it was just like, Yep, he's he's on the same wavelength that people love right now. I it's funny you had mentioned taste 'cause I think of the like there's that like Rick Rubin gif Adam Jacob: Mm-hmm. Sean: of him, like people or video of short of him being paid for his taste. He knows nothing about music. Adam Jacob: Yeah. Yeah. Sean: And in this case, you have somebody with taste and somebody who is actually an engineer and knows like what engineers kind of want. So it's gonna be interesting to a lot of people. but I also feel like I I'm more and even though I don't work the same as Paul or Nick, those alacrity windows are crazy. And I am not exactly a GUI guy. I live in TMUX, but I can just build the thing I want. No. So I don't know it's gonna have to be really compelling for me to be interested in it. Adam Jacob: Yeah, I think if there's if there's a if there's a thing I'm looking at with skepticism, it is that. It's just in a world where we can build the software exactly the way we want to, what does that mean for how we think about consuming software, think about our own workflows? I don't know the answer to that, but I think it's I think it's one of the more interesting questions of of this era. Okay. Sean, that's the end of the news. So now it's time to ask you five questions. Are you ready? Okay. So of the Sean: Okay. Adam Jacob: stories we just covered, which one will you be thinking about later today? Sean: I'm gonna be thinking about open weight models. Probably the most. Adam Jacob: Cool. which one is noise? Which one are like meh? Paul Stack: You can say it, it's okay. Like you can totally. Sean: the rest? The rest? No. I already the i I'll say okay. I agreed already a hundred percent with what Paul said, so it's kinda like, yeah, that one's off my table right away. Adam Jacob: Which one? Paul Stack: I didn't pay him. I didn't like for what I Sean: D no, it's true. Practice is the half arc architecture shouldn't practice if has have a half light. Adam Jacob: You think that one's noise? Sean: No, I think it's I think it's not noise, it's just I'm not thinking about it anymore. So Adam Jacob: Okay, that makes sense. what's a common belief in the industry that you think is won wrong? Sean: That tokens are expensive. Paul Stack: Ho ho ho ho ho ho you've gone for it there. Adam Jacob: yes. Okay. Paul Stack: You are gonna have people at your door. Like, I mean, Adam Jacob: Let's go. Paul Stack: my god. Adam Jacob: what's the one thing you wish you could teach everyone in the world? Sean: system thinking. Hands down. Well I Adam Jacob: Okay. Paul Stack: I'm just gonna say swamp, but Nick: Ha ha Sean: I am not the advertiser. Adam Jacob: Okay, what's the best what's the best and the worst moment you've had working with AI agents? Sean: the best. Well, maybe the best and the worst right now was when I took out the telemetry for you guys. That was really fun recently. It was it was Adam Jacob: Yeah. That was super fun. Paul Stack: Ha ha ha. Nick: Great blog post, by the way. I loved that. Sean: it was totally accidental. I wasn't clear about my intent or what I wanted it to do, and it just went crazy. And then Nick is like, Hey, what's are you doing something insane? And I was like, yeah. But then at the end it was Paul Stack: For those for those who are not aware, it's si he was throwing six hundred requests a second at our telemetry server and we had no idea it was coming. Adam Jacob: Yeah. Sean: It was fun. So but it w Nick: We were built for that scale, man. And I said more, you know? Like and that's what we need. So thank you again publicly. Adam Jacob: Yeah. It un it uncovered a bug in our architecture. And now that now you can go way past that and we don't have to think about it, which is good for everybody. Okay. Well, that's all the news that's fit for infrastructure this week. If you want to dive deeper into any of the things we talked about, you can find all the links and the summaries over at Swamp Club Swamp Dash Club dot com slash podcast. Sign up to get it delivered straight to your inbox every Monday morning. If you're trying to figure out how to manage production infrastructure and applications or to scale your use of AI within your organization, Swamp is here to help. Visit us at Swamp Dash Club.com to get started. Using AI agents safely and effectively by building repeatable systems you can trust and verify. I'm Adam Jacob. Paul? Paul Stack: Stack. Nick: I'm Nick. Sean: And I'm Sean. Adam Jacob: And Sean as our special guest, you get to take us out. So Sean: All right. Adam Jacob: say goodbye and plug anything you want. Sean: Thank you all. Happy swamping. Adam Jacob: All right, we'll see you next Monday. Thanks. you