Adam Jacob: Okay, welcome to AI Frontiers, where industry experts run down all the AI news for infrastructure engineers. It's Monday, November ninth. So grab a snack and let's get started. This one's gonna come at you a little late. So hopefully it's a like a like a liner snack. I'm Adam Jacob. Nick: It all it always arrives right when it needs to, according to Gandalf, so. Adam Jacob: Right, Gandalf knows. I'm Adam Jacob. I'm the CEO of Swamp Club. I'm joined this week by my Swamp co founders, Paul Stack. Paul Stack: Happy Monday, peeps. Adam Jacob: And Nick Steinmates. And my old friend, DevOps legend, video game director, pinball wizard, entrepreneur, CTO, and CEO, the legend himself, Jamie Stormbreaker. Jamie Stormbreaker: Thanks for having me guys. Adam Jacob: thank you for coming. Let's run down what's that? Nick: Pinball legend. Where's where's that where's pinball legend come from? Then we can we can we take a little detour really quick? Jamie Stormbreaker: well Yeah, Paul Stack: No Jamie Stormbreaker: sure, yeah. We got a we got a really cool league in Seattle that we've been it's been together now for about twelve years. and you know, we just had like thirty or forty people at the at a time that were like wanna start a league and just like compete against each other. We got a really great like bar scene with a bunch of pinball games here and started a team and you know, ran it for about ten years, brought us to the championship. and you know like No learning that there's a competitive pinball scene in number of states in the United States. It's not something I had on my bingo card. and then it was a great escape from digital games, so yeah, if anyone wants to learn any tips in there in the Seattle area, I'll show you everything I know. Lessons are free. Paul Stack: I am utter dog shit at ping at I didn't even say the word right at pinball. I don't I can't even imagine that there's a a a competitive league for it. That that's fascinating. It's fascinating. Jamie Stormbreaker: yeah. Yeah. It's a phenomenal game. It does all the right things where it's kinda like how refs make games even more fun, even though you don't want them to. Not at the same time, that's rule number one. Paul Stack: Isn't that all you do? You just whack the paddles? Adam Jacob: Yeah, hitting hitting both battles at once is a recipe for death. Yeah. Yeah, definitely don't do that. Nick: Ha ha Paul Stack: See? There you go. Jamie Stormbreaker: Literally rule number one. Yeah. Yeah, we've got a great scene here though. Yeah. Nick: So fun. Sorry, Adam. Adam Jacob: Yeah. And he is very, very good at it. in a way that I am not. Paul Stack: Yeah. Adam Jacob: but but is super fucking fun. Okay, let us run down the news. So top story for this week. Robert Haas wrote a blog post about Postgres nineteen being late. In case you're not like the long time, you know, person that are, Postgres nine Postgres is never late. Postgres ships, I mean like clockwork, but it's gonna be late. And he wrote a great blog post about why basically more bugs are being found and more of those bugs are being backported to the earlier releases of Postgres. So, you know, when you find a bug in Postgres, it's not just fix it in the new version, it's fix it for all the ones that are under support. And so, you know, they're much slower to get the new features out. And then some new features, you know, they haven't even been able to sort of get all over the line because they've been sort of so overwhelmed by by by bug reports and fixes. All of that they laid, directly at the feet of AI and AI assisted bug reports and and finding issues in Postgres. And his question at the end of the blog post is basically, how are we going to keep up in the face of all of this? and I think I have an answer, but Jamie, do you have one? What do you think Postgres needs to do to keep up with the ever growing burden of their bugs? Jamie Stormbreaker: Yeah. Well I think that every especially every piece of of publicly shared software. So any open source community, anything that just says open code is gonna keep running into these. not just because you know, you can attach an entire harness and attack at endpoints, right? Like any socket that's open you can try to abuse. but you can look at the code as well. even if you can't look at the code. and there's an application running underneath that socket. So PHP, a couple weeks back, right? there's a kind of attacks and testing that you can do that you couldn't really do with deterministic programs before. So deterministic programs, we would do fizz buzz testing, just flood the system with data, try to blow it at the seams, right? but there's this new way that you can test with LLMs because they have infinite time and you did not have infinite time. They have infinite time, they're just sitting there waiting. I'm gonna get the name the word of this wrong, but it was like a temporal state. it can test all these various states to try to abuse a system that Adam, do you know what it is? Adam Jacob: Yeah. antithesis testing. Jamie Stormbreaker: Yeah. And so like setting those up is cr is really, really hard. And so I think that a lot of the bugs that we're gonna end up finding here are things that have been dormant in a system for maybe decades. Like that PHP one. I know that. Ever the world a lot of the world's moved on from PHP. I don't think it's fun to make fun of languages in any way. But that bug's old and it is simply because we just didn't have the power to find it. and so I imagine that if you're not adopting this idea that the bad guys have nuclear weapons and you need nuclear weapons to defend yourself, you're gonna have delayed releases, you're gonna have critical vulnerabilities. and people are gonna need to figure out what they're gonna do with open source software and how to protect themselves. Adam Jacob: Yeah, that was one of the things I thought was interesting about his blog post that was missing was like they're like, Hey, AI is the problem. No mention that AI is also the solution to this incredible backlog. and, you know, and to the difficulty of those bugs. Paul, I can see it. You're just you're dying. Paul Stack: I'm smiling so much because I was at a conference in New York last week and the great Eric Meyer was the keynote speaker and he basically talked about Postgres nineteen being late and he said, listen, we gotta get over the fact that it's buggy as shit software. It's been buggy forever. AI is helping them see how buggy it is and it's not ready for release yet. And he's like, any other excuse at this moment in time is like out of the window. It's just that the bug debt has finally caught up with actually what's in the system and it's more like viable than it's ever been to actually see it. So and it was it was a a stark reality because you said that's gonna happen more and more of these large pieces of software that have been around forever, that there's been manual creation of systems. Nick: it's it's fascinating that you know like Postgres is is kind of like the the technology darling, you know? Like it it it when people think databases now more and more they think to Postgres, no matter what type of database it is. And like, yeah, one way to look at this is fully embrace AI. and if you take that to its logical conclusion, I think what You end up thinking about is something like rolling releases. or you think about what the real hard gates for a stable release looks like. And I think that that's where the edges are starting to fray a little bit. It's like, what what do you want to put into this release, into Postgres 19, and what's worth keeping out? and like, I don't know. It we have this problem, Paul, where like, you know, we're Doing what, 60 releases a day or something like that? And so and so how do you how do you with it with a Paul Stack: Yeah. Nick: software like a database, continuously release so many updates while saying, hey, now it's stable and we're gonna backport fixes to this forever? And like it's like a real release engineering challenge that stable software in the face of AI kind of comes up against, I think. And so Adam Jacob: Yeah. Nick: Anyway, there's there's a lot of different ideas there, but it's yeah, I I don't envy the situation at all. It seems very Adam Jacob: Yeah. Nick: difficult. Adam Jacob: And I don't either. And I feel like the answer is build a software factory that understands your bugs, has requirements that can test the whether or not the software, you know, met meets the requirement. And like, yeah, you have to maintain multiple lines. And also like the factory can maintain multiple lines for you. And, you know, like there's which doesn't I don't mean to make that sound like it's simple or whatever. It's not. And like Postgres is complex software. And part of Part of why Postgres is Postgres is because it has this incredibly intense developer process where the people who are writing that code, they're real experts. They've they've they've cared about it very deeply over a very long period of time. It it has it is a very rigid piece of software in terms of how you can contribute to it and their engineering requirements. And like, you know, that they can keep up as fast as they are is amazing, given that they're also not going to even more advanced versions of this of of the technique. And I think, Nick: Yeah, but Adam Jacob: ultimately they're gonna have to, I think is the answer. Nick: If you think about it like as torrential rain, in terms of number of bugs and fuck bug fixes that can come into the system, the ground can only support so much before you create a lake or, you know, an ocean Adam Jacob: Totally. Nick: or whatever. And so like that's the problem is is they may be able to increase velocity on one end, but can people actually absorb that rate of change in, you know, business critical systems is a real Adam Jacob: Well, you're not gonna release Postgres more often. That's the whole point. Nick: Yeah yeah, you you either have to release Postgres more often or you have to like have a real sturdy wall of what it means to be Postgres nineteen or Postgres twenty or whatever and be able to withstand that rate of change. Like it's a really interesting conundrum that I don't think we've f figured out as an industry Paul Stack: Yeah. No, Nick: at all. Adam Jacob: No, we haven't. Paul Stack: no. Jamie Stormbreaker: I have a hot I have a hot take. You have to release Postgres more often. Like you cannot live in a world where you're gonna try to put constraints on a system. Other people are going to be filling your system and trying to attack it. You really need to release on the edge. You know, like I I'm I understand I'm the crazy one that just pushes directly into production a lot of times, or like Adam Jacob: Yeah, yeah. Jamie Stormbreaker: one stop, but like you must have deterministic gates that predictively prevent and understand that what you're about to push out is reliable. And you can you the same thing that builds those predictive gates is the same thing that is attacking you and finding your vulnerabilities. It's a software factory. It's machines that build machines to rage against the machine. And so if you don't have the ability to like build that into your system and you don't adopt it, your software is going to get forked and replaced. I love Postgres by the way. this is not a great part of the story. I just feel like a lot of us for years most of us did not program an assembly. Adam Jacob: Yeah, yeah, yeah. Jamie Stormbreaker: and so you don't remember those days. but right now, my hot take is that you need to improve on all of the systems that deliver software. You need to improve on all the systems that run it. And if you don't do that, you're probably gonna be at risk. Adam Jacob: Totally. Jamie Stormbreaker: so personally, as someone that has operated some of the biggest games in the world, I would take edge updates from Postgres in this world if I knew that they were good. Adam Jacob: Yeah, for sure. all right, well let's move on. here's a place where you don't want to take edge updates. Supermarket freezers across Europe hit by another suspected cyber attack. the food attacks will continue until morale appro improves. I think I still haven't seen any update on who the source is of the attacks on the refrigerators. but again, we're back to like everything's getting owned. We talked about these food attacks the first time they happened a couple of weeks ago. This was a second round of similar attacks. So it just, you know, I feel like we're gonna wind up in a world where we just need to disconnect a whole lot more stuff. because I don't know that the supermarket fridges need to be online, you know? Like for whatever, Paul Stack: Ha ha ha. Adam Jacob: for whatever advantage we got out of being able to centrally manage those fridges, and I have no doubt there was an advantage. ooh, that now would be a good time to unplug your magic fridge. I was looking in my own fridge and it has a little sticker in the back with a Wi Fi signal on it and it was like, Hey, you should turn this on and I was like abs like I had not even noticed it was there until I saw it as I was prepping for this week and I was like, I should remove that sticker. let us never turn that on. Paul Stack: Called it, called it, said at the time Adam Jacob: You did. Paul Stack: this was the start of the attacks, this is the start of the wave. It's funny. UK Prime Minister has told everybody here that we should have provisions for between three and five days. like at home. Of course, that's sending some people into flat panic and other people not. but at the same time, you know, they've the expectation right now that this can continue for like electricity plants. Water plants, all sorts of stuff. everybody's freaking out. genuinely, some of the government officials are freaking out about this. So Nick: I'm surprised the conspiracy didn't come come out and say, hey, if you're operating a shitty supermarket and like your refrigerators went bad or whatever, what a time to, you know, just blame it on the hackers or whatever. And and also, yeah, Adam, I hope I hope your vision of of and and Jamie's vision of like, hey, if there's a port open, people are gonna exploit it. and therefore what you need to do is turn all that shit off. I hope that comes to consumer electronics where we get our Adam Jacob: Totally. Nick: machines back and our systems back without being spied on continuously. That'd be great. I'm not gonna I'm not gonna hold my breath for it, but man, Paul Stack: Except your ROM. Yeah. Nick: would that be great if that was an outcome. because you definitely Paul Stack: You you shouldn't be allowed any RAM in your machines. So if there's no RAM Adam Jacob: Yeah, we need it. Paul Stack: then nothing can run and communicate to the internet anyway, right? It's fine. Yeah, exactly. Nick: Definitely, yeah. Absolutely. Adam Jacob: We need it for super intelligence. yeah. Jamie Stormbreaker: this is where the right to repair community comes in. I should be allowed to rip the operating system off of my TV and put whatever I want on it. if I don't trust the company to do it and I want my life to be fully connected, I should be able to control it. I bought it. you don't control my TV. I'm not in. I do love that statement though, that's like, why did we ever put the refrigerators on Wi-Fi. maybe there's a reason, but if it doesn't do anything for me, I don't want it. and maybe they should be, and I don't know why. but I would say that I think that you're gonna see more and more of these odd devices getting attacked because there's the possibility of side quests is so big now. You can just go Adam Jacob: Yeah, yeah. Jamie Stormbreaker: build whatever you want and you're like, I bet it would be fun to see if I could turn that supermarket off. I'm not suggesting that you would do that or I would ever do that. But you're gonna have a lot of people with these tools that are gonna do that just to see if they can. And I think a lot of the market, like everyone in this room, probably everybody watching, is nobody that built the the Wi-Fi and you know, operating system for those ri refrigerators. You know, it's like a thing that had to get done and then gets forgot about. you know, when you're operating something that has something to do with like national intelligence, you're gonna be thinking really I really need to make sure this is good and protected and and monitored and you're constantly updating it. But I think that Paul Stack: Maybe. Jamie Stormbreaker: that Nick: That's the hope. Jamie Stormbreaker: that was a that was a hope. Yeah, yeah. Adam Jacob: Yeah, yeah, for sure. Nick: Mm. Jamie Stormbreaker: But if you know you are a a lot of folks don't have the power that to protect themselves and build it into these systems that they're gonna start seeing people on side quests coming and disrupting it. And I expect to see more of this. I I read this ahead of time and I was stunned that anyone even thought that would be a fun side quest to go on. No gain. Adam Jacob: Right. I f it feels like an obvious it obviously is a state actor thing to me. this is Jamie Stormbreaker: Yeah. Adam Jacob: this is this is this is preparation for broader like broader attacks surfaces of civilian populations. Yeah. Nick: So Paul Stack: See? Conspiracy theories. Fucking conspiracy theory. Paul here, like he's like just saying. Nick: Yeah. Adam Jacob: Yeah, sometimes it's me, but not very often. Okay, let's talk about Nolan Lawson. So Nolan Lawson wrote a great blog post about the cost of doing AI builds of various JavaScript frameworks. This was a good solid study of how much it costs to build across different frameworks depending on what you chose. you know, the the need for skills, the need for things not being designed for AI first has really shows itself in the results. So, you know, things that were boilerplate heavy or that were like really pushed hard as like, hey, this was being built for humans so that humans don't have to think so much about the abstraction, like they really suffer when you look at the total amount of time it takes for AI to solve for it. So, the cost per run for the equivalent features of just Hono and HTMX, which is, pretty thin, was, half the cost of doing it in Angular, for example. And so as you climb up the framework structure of like, you know, how much am I providing? How much of the framework is in here? Like, Some of that is, the code that gets written. Some of it's just waiting through the documentation, waiting through the explanations, waiting through the examples, waiting through all of those pieces in order to understand how things would work. I think it's an interesting way to think about how we choose, excuse me, how we think about and choose our frameworks. you know, because historically it would have been according to like developer productivity or dependency trees or features or those sorts of things. And here it's sort of showing that like. perhaps one of the cost optimizations that's sort of hidden in our in our midst is choosing the technology that does a little less for you, that works a little more specifically, because it turns out that the agent will sort of go the more direct route. what do you think, Paul? Paul Stack: I the measurements here were fascinating. I I am not a fan of Angular. I've never been a fan of Angular. I think it's incredibly bloated, but it consistently did bad across the board in this. But when people became or when they do become proficient with Angler and becomes their way of life and how they work it, they're so quick and been able to do it. So it kinda to me leans into the moat and the fact that the moat is gone. You know what I mean? If you built if you built like your your skill set around a particular technology set. That's up for grabs at this moment in time, right? It's like solid engineering practices and solid engineering principles is what is what wins out under this dot under this blog post for me. It's like it doesn't matter what framework you choose. There are options now that are efficient in tokens, efficient in speed, efficient in data transfer, all of these things. And you're gonna you're gonna choose the one that's the most efficient in all of these pieces. You're not gonna choose the one that you think because I wrote Angler for twelve years, I'm gonna stick with Angler. it's it's it's kinda it it's put everything up for grabs. I'm like you're gonna see similar things when it comes to like programming languages as well, because again, people have staked their life on like writing a in a particular programming language. And it's possibly or more than likely not the most efficient for an for an LLM. Nick: Well, that's a really interesting question, which is what is the most efficient? And it could be in terms of tokens generated, it could be in terms of grokability or understandability or minimal lines of code or whatever. Like I think that that's the that's the question. Like I I think we all remember the era of a new web framework coming out every other week or whatever. And what's been interesting over the last year is that stopped pretty much in like Pretty much its in its tracks, Paul Stack: The tokens have all gone in sandboxes now, yeah. Nick: that's not happening at all anymore. I do wonder back to Jamie Stormbreaker: Wonder that. Nick: increased efficiency or something like that, if there's a framework yet to be built that is gonna be more optimized for AI agents and what that would look Jamie Stormbreaker: Yeah. Nick: like in terms of training or whatever, but yeah. Paul Stack: I honestly think a bunch of people are just going to play in JavaScript. literally just plain JavaScript. we didn't like Adam Jacob: Well yeah. Paul Stack: it. We genuinely hated it as an industry because there was no level of abstraction and there was nothing that like it just to make our eyes bleed writing it. But it's incredibly efficient and if an agent can write it then you know what Adam Jacob: Yeah. that's that's that's it actually. Right. I think back to what would it look like if that's a thing you were trying to do. my suspicion is that one of the things we found building Swamp was that, we've been building automation for a long time designed to make it easy for people to think about their systems and then baking the the the principles into the framework so that a person wouldn't have to think about, whether or not that package needed to be installed or that server needed to be rebuilt or whatever. And Turns you don't need to do any that anymore. That actually it's better if you just tell the system what you want it to do and then it does it. And if you wanted it to be resilient in a particular way, tell it to. And if you didn't, then don't. And and then by not doing it that way, the system can actually solve this much wider number of possible problems and it's more efficient as it gets to its path. I I feel like the same is happening here where we're learning, like, yeah, it turns out if the robots are riding it. Everything I everything I thought I and we're not reading the code, everything I thought I'm I meant about like, why I would use that framework or why that shape matters just doesn't. Jamie, what do you think? Jamie Stormbreaker: Yeah, I have three thoughts here. One, DH, the guy that made Rails, right? they just had a talk, a conference, and he gave it he he gave a keynote. And one of the things he noted was he famously hated Rust, right? I guess. I haven't Adam Jacob: Yes. Jamie Stormbreaker: followed Rails in a while, but I I was part of that community. I loved every minute of it and you know, it was a great part of the or the my early part of my career. and he said something along the lines of Rust is inhumane for humans, but it's great for AI. And we need to really embrace the idea that there are some languages that we did not like writing as humans. And so when you talk about frameworks, a lot of it is patterns for us to communicate and share with each other what we want. and we needed that. And a lot of times it would add extra bloat. But really what we're coming down to is, you know, I I'm literally I'm gonna go back to duck typing, which is a concept from small talk in Ruby. If it looks like a duck and it quack like a duck. It's a duck. And so if you think about a program as a thing that runs on a machine, uses resources, and it maybe has open sockets, you know, sockets of file, port, whatever. as long as it does the behavior you expect and you measure it in outcomes, I do not care what is inside of that program. It does not matter what language you wrote it in. I never cared that Nginx was written in C. It could have been written in anything else. I do not care. As long as it was as I mean, it solved the 10K concurrency problem, right? That's all I cared about. So We I'm an I have been a language snob. You know, I've programmed in like thirty two languages. I really care about weird ones like Erlang and Lisps and Rust. I actually do love Rust. and one of the things I've had to let go of, I mean, I literally was rewriting some of my LinkedIn yesterday. I'm like, I don't even care about languages anymore. It's like whatever the LLM wants. And so we had these things called language shootouts a long time ago. I bet some of the folks watching this remember where it's like on these cores, these languages, and I was always sad that Ruby, was always losing and then Python would be like, you got your gill still, it's losing. but like you were always measuring for performance. And so I've tried to come up with a way to like language shootout for an LLM. And this is really hard because it's not deterministic the same way that you can measure it. But what I have, and I will totally be proven wrong next week. It's going to great. But what I've come to the conclusion of is that Languages which are strongly typed typically are more performant in that less issues. I don't even know if you should write test code. I think Paul wrote a whole thing about this. I think I agree with him on this point. t t unit testing, but like end to end testing, yes, that's a whole separate thing. But I think typed languages are where you want to be. and then Oddly enough, languages that don't have a lot of history on the internet seem to perform better. So I wrote a Lisp that described Habitat's packaging language. There's no documentation Adam Jacob: yeah, yeah. Jamie Stormbreaker: except for my specification. It never gets anything wrong. I talked to one of my friends that is like a game engine programmer that I've worked with for over a decade, and we were talking about the new C standard. He's like, Yeah, I literally can't get it to focus because it keeps pulling up old specifications, old ideas. And so there's this weird idea here, or this weird thought that perhaps not having a body of ideas around the way a language works might be beneficial. These are not actual items. I have no idea what to do. The I I don't think the thing is go make your own languages. Maybe maybe it is, maybe it's not. Adam Jacob: Maybe. Jamie Stormbreaker: but I know that type languages it's very, very good at and it happens to be really good at Lisp, as long as it's declarative. it's a dec declarative language, not a full, you know, full thing. Adam Jacob: Yeah. Talk about things that people won't want to read is LLM generated Lisp. Sorry. Just I just that gave me that gave me gave me heartburn just sitting here in the chair. Paul Stack: Chills, literal chills. Yeah. Jamie Stormbreaker: I did not design it for people. Adam Jacob: yeah, obviously. Okay. And I and I and I am someone who liked reading Lisp. I wasn't mad Paul Stack: Yeah. Adam Jacob: about it, but whoa. Okay. So OpenAI has lowered their fundraising target after decreasing their possible valuation. So they cut their potential fundraise from 30 billion down to 20 billion. And lowered its valuation to 1.25 trillion. I think what's interesting here, we're almost in the anthropic IPO post time, it's it's all gonna happen for us. and I think OpenAI, we're in this moment where, you know, they too are gonna wind up going public and we're gonna have to figure out sort of what exactly is that shape. you know, I think we're still going back and forth about like is the bubble gonna pop? When will it pop? How does it look? You know, But it just to me comes back around to when you look at those dollar numbers, you're like, I I struggle to care that that you're only gonna raise twenty billion dollars instead of thirty billion dollars. Do you know what I mean? I'm like, so you had a one point two five trillion dollar valuation instead of one point five trillion, I just I don't I don't know. maybe, maybe I don't care. Key, what do you think? Nick: Was all this hype, that who was gonna be first, anthropic or open AI, to go IPO or whatever. And I don't know, I'm sure there was a first mover advantage for anthropic or maybe a first mover disadvantage to see how the market would react to that. You know, I I think I was I think we were all fully expecting them to IPO around the same time, and so the fact that that's not happening, and instead what they did is raised a new round. Yeah, it's it's an interesting that's an interesting development. I don't know how to think about it yet. other than that's a Paul Stack: It it kinda feels like it's Nick: lot of money and we'd all want to have that problem if we could and also you know, and also like yeah, it's just a weird outcome. I don't think we could have predicted. Paul Stack: It feels like it's massively related to when we spoke last week about their anthropic specifically about their credit rating being junk and about like, you know, like there there's definitely something that seems to be there. Like I the suggestions were last week that it's gonna like be delayed a quarter at the at the earliest before they IPO. Of course there's no substantial evidence for that in any way, shape or form. That's just pure rumor and pure conjecture, like from the internet zeitgeist, but the fact now that open AI are also like doing some weirdness and some shenanigans with regards to another race sort of feels like there's something. You know what I mean? That there's absolutely something that's like like brewing there, but yeah. I and I'm not the person in our company who understands any of the business aspects, especially at those numbers. I mean, Jesus Christ, I cannot Adam Jacob: No one. Paul Stack: fathom thirty billion and a one point two five trillion valuation. I mean, that is just bonkers to me. Adam Jacob: Yeah. Yeah. None of Yeah. No one who works at Swamp Club understands those numb that numbers at that size or running companies at that size, just so we're clear. No one does. Not yet. Go ahead, Jamie. Nick: Not yet. Jamie Stormbreaker: Yeah, I I could I couldn't possibly imagine that a speculative company, no matter how good the technology is and how life changing it is, is worth more than Walmart. and like I know how powerful this is. I'm not saying that you know, I I I always akin it to like, you know, hammer and nails versus like an auto gun, right? A nail gun. yeah, that's life changing. You get on a roof and you have a nail gun. Yeah, I never want to not have that again. But it it's a tool. Like I feel like they're betting on subscriptions and building a a like a moat around the idea. And then you watch other companies distill and refine AI. and I'm not saying that you should do that because you'll break the terms of service and you'll lose your account. Definitely don't do that ever. but is it gonna sustain? and I mean the hottest take here is are we building something amazing that's actually for everybody and not for the people who are preferred investors? and I don't think anyone watching this that is a preferred investor wants to hear that, but like are we just in this race where you need to offload to the public an IPO so that way it saves your investment? Or do we actually build something accidentally which empowers all of society and not just billionaires? Adam Jacob: Yeah. I don't know. We'll find out. It's the thing that keeps me up at night. Jamie Stormbreaker: I have no idea. Adam Jacob: let's talk about Deep Seek. or as Keebe sometimes puts it, the Chinese homies. they have Paul Stack: Ha ha ha. Adam Jacob: they have unveiled an autonomous coding agent prototype called DSH, their super agent prototype. what I think is, you know, when we think about ways that we could insert, we're gonna talk about China again in this podcast for sure, but when you think about interesting ways you could insert things, like owning the harness is certainly one good way to be able to insert whatever you would like in the process. What's interesting about the the prototype that they built is its job is to take as little human attention as possible for as much value as possible. So it only acts when you when you ask it to, or when a wit written rule allows it. It only speaks when a person must act or can use a result. So if you can't, it's silent. You know, it's just like working on it, Bob. It asks one question at a time in a structured form. So it doesn't like embody, if you've ever used Claude and it's been like, here's the three things you should think about. And they were spread out across five paragraphs. it's gonna ask you one concrete question at a time. it thinks about impact rather than effort. So, so instead of like saying, Hey, this is how hard this thing will be, which you see a lot in Claude, where it's like, it'll take me five days. And it's like, no, it's gonna take you ten minutes, because you're Nick: Yeah. Adam Jacob: a robot. but this thing's gonna be like, this is like a this is a big change or it's a small one. and and it tries to sort of report honestly. I haven't actually run it because, you know, my desire to check it out a get and like figure it out is low. but I think it's an interesting piece of the puzzle where when we think about sort of what are we seeing come out of China and how are we thinking about how those models move, like, you know, it's obvious that they need to move into the coding harness, they need to move into that sort of into that cycle and that what they're doing is actually starting to push the R and D of like what is that shape and how could it work? I think that's super interesting. what do you think, Jamie? Jamie Stormbreaker: Yeah. the Chinese models are interesting. I also am skeptical only because I'm I mean I'm not even putting code on GitHub anymore. Like I'm I'm now fully just a hermit that pulls everything in and I don't wanna share because I don't know what's gonna happen, right? I think that the behavior Adam Jacob: Is that the moral equivalent of a of a technology prepper? Nick: That's that's why you need the blinking Jamie Stormbreaker: Yeah, bas yeah, I think I'm a open source prepper. Yeah. I've I've re I've taken Adam Jacob: Yeah. Nick: lights, then you can have it all. And you do have the blinking lights, so yeah, you just make Jamie Stormbreaker: Yeah. Yeah. I Nick: it all run local. Jamie Stormbreaker: Yeah, I've taken the the open source idea and just moved it into Discord rooms instead and like never share this with anyone. so I'm a little Adam Jacob: Yeah, yeah, yeah. Jamie Stormbreaker: bit like that just for anyone watching right now. Like I don't know what's gonna happen, right? But I do love the idea that they are exposing behaviors that the community is ex i is already exploring and using anyway. you know, like my s Powerline Bar and Tmux all tell me, when my my cash is gonna expire. it tells me how big the context is, There's people have figured out ways already, to identify when they're gonna start spending more money on tokens and when they're gonna start, increasing the cost of what they're paying to their service. they also have been building tools to try to simplify that. you like by default, Claude's gonna say it's gonna take five days to do something, but you can Put guardrails around it if you cared enough to tell it stop with that shit. Don't don't tell me that, right? So I like that at the bare level, instead of me putting an overlay on the context that, you know, you could get the the the the actual behaviors that you're looking for. and I'm I like I know that these other companies and models could do it. It's not impossible. I feel like they're choosing not to. So this is a selling point for the Chinese models in their companies, they are saying we have a better user experience. That is a little weird. yeah. Nick: You know, the more customers I talk to, it it it it turns out like a lot of them are thinking really heavily about lights off factories and what that means. And what I really like about the this deep sink agent is one, you know, what are the Chinese known for? They're known for their deep manufacturing expertise in factories in particular. And two, like, I think the key line is take as little human attention as possible and give as much value as possible. And to me, what that means is lights off factory in our in our parlance. what they're trying to do is deeply integrate into the places that the operator lives and then you know, try to create lights off factories to get code generated and through the system. And so I think that's really interesting to think about purpose built. harnesses that accomplish, you know, purpose built goals in that way. And so for that I think it's really interesting. Paul Stack: I have a lot of thoughts on dark lights off, dark factories. I taste is like kind of what's important in software and a perspective. And like I I could switch all of our stuff at this moment in time to a like a dark factory and it would be horrendous. It would be fucking horrendous because every single feature that come in would be implemented and it would be we've all seen that that meme where we've got this huge Building block of all these weird things, and then there's one tiny little stick that's holding up the entire thing. That's effectively what we'll end up building. We still have to have people who drive systems who have that systems knowledge and have that systems thinking. There are definite places where lights off factories are useful, but I think the move. I I I don't get this idea where more and more people want them just to be dark factories because in six months time they're gonna walk that back and I think they're already seeing that like a bunch is gonna change. Nick: Well whether whether Adam Jacob: Yeah. Nick: you whether you get the motivation, I think we all get the motivation, which is like, let's have this shit autonomously generate shit for us. and Paul Stack: Yeah. I get I get it. Nick: also whether you think that's practical or not in the end, I think those two things are separate, yeah. Adam Jacob: Yeah. Yeah. I might the reason I don't want dark factories is because I don't want to have to debate in a Google Doc or in Jira or in your or wherever your issue tracking. I don't wanna design everything in Markdown and then have to get it all right so that my dark factory doesn't mess it up in the middle. Like that sounds worse. So like Nick: Well, we did learn through waterfall that that's not even possible. the whole idea Adam Jacob: even the original waterfall paper thought it wasn't possible. Waterfall was all about speed of iteration timing. We just we Nick: Yeah. Adam Jacob: just never most people never read the paper. They just took the one diagram and were like, Aha, I know, yeah. Mm-hmm. Nick: Yeah, we're gonna get the spec perfect. Yeah, it's said everyone they want to do always and then never actually happens in practice. And so Adam Jacob: Yeah. And said never the waterfall paper. The waterfall paper never said that. It was the opposite all the time. It was like, Hey, you know what you should do? You should then go through this waterfall cycle as fast as possible and loop around is what literally what it said. Anyway, okay, let's move on. Go ahead, Jamie. Jamie Stormbreaker: Well no, real fast, just for the listeners and me, 'cause I don't know. What's a dark factory? Adam Jacob: Paul hit it. Paul Stack: No human interaction at all. it you just set it off from the start with this the exact spec that you want and it pops out the other side. Literally a Adam Jacob: Yeah, you're lit. Yeah. Nick: Yeah. Paul Stack: factory that, has the exact specification Adam Jacob: Ticket comes in. Paul Stack: and inputs like that it needs. Nick: To give to give the devil its due, if a bug bug comes into Swamp Club and says this word is is incorrect on this page or whatever, you can run that shit as a dark factory, no problem. And Paul Stack: Yeah, totally. Totally. Nick: so the question is just a matter of degree, you know, like where is that threshold that makes sense for you and your products that what you can do is let the agents do the busy work and what is the work where you need to focus your attention is the entire point of the game. Some people are moving all the way to the there is no spectrum, it's all lights off, which is crazy. I think we would all agree. and also having lights on for you know, in a very stringent process for shit like that, like doesn't make any sense at all. And so anyway. Paul Stack: kinda silly yeah. It's kinda silly. Yeah, totally. Totally agree. I Jamie Stormbreaker: Yeah. Paul Stack: gave a talk about this last week in New York, so yeah. Jamie Stormbreaker: Yeah. bug fixes and things that are d deterministically de defined. for instance, I don't want mu the game to crash. yeah, like that's all actually automated. It just p comes into a a crash system auto yeah, it just does the whole thing, pushes it out. It's like, yeah, great, bug fix, right? I think taste is the right word there. Paul, you might have accidentally been the person that put that in my brain, or somebody else had, but I think that is the right way to describe it. In the games industry we're seeing Paul Stack: I blame Adam. Jamie Stormbreaker: it might have been Adam. Yeah. Someone in this room probably Paul Stack: Ha ha ha. Jamie Stormbreaker: put it in my brain. But In the games industry, what we're seeing is a lot of and if you're on YouTube or, Twitter or any of the socials, you're seeing someone say like, I gave Astra this one prompt and it made this entire game and it's Grand Theft Metal Six. I'm like, Bullshit, that is ad revenue that you are you're farming for clicks right now. It literally cannot. It doesn't have taste. The same way that, you know, a million monkeys with typewriters, maybe I guess could make Shakespeare, but I pretty sure that it would never happen and we'd never be able to prove it 'cause we'd all be dead by the time maybe it happens. It's never gonna happen. it is not novel and you do need humans. And so yeah. Adam Jacob: Yeah, they're averaging machines. Jamie Stormbreaker: So I think that you really need to do here is just build great developer tooling and empower people, especially if you're making something creative. Adam Jacob: Yeah. All right, let's talk more about software factories. So John Cutler wrote a good blog post on his Substack. Is that do you call it a blog post if it's on Substack? I think you do. Feels like a blog post on Substack. A Substack. He wrote a Substack about how software Nick: Substance. Adam Jacob: factories are themselves becoming siloed systems that are then causing other problems. And, you know, the quote here is like we're all enthusiastically building systems that preserve our professional identities. And potentially give us a new source of power, but we're still organizing those systems around the very functional boundaries that cross-functional product development was supposed to transcend. And I think his point was basically: look, maybe it's just cope, that software development's cooked. And now we're all talking about how it's not cooked, because we build the factory and we build the machine that builds a machine, and our taste is important. And so maybe we're cooked, or maybe we're not cooked, but what we're still doing is sort of baking. the organizational structure and our understanding of our roles into the way that the new system is behaving. I think that part is very true. So in our experience, we went, we speed ran a lot of this problem. And what it did was break a bunch of the traditional roles about about how we work together and, you know, what our expectations are and when do we talk to each how do we intervene and when do we talk to each other and like how do I how do you control for quality when people can move at this rate of speed, like, so many things start to fall apart. And so I think I think it is an interesting perspective that says, hey, is it just cope and we're we're we're just lying to ourselves by trying to level up one level to prove we're still relevant. Or yeah. What do you think? Paul Stack: It's Conway's Law. Conway's Law is playing out in real time faster than we've ever seen it before in any organization. And that seems to be like what's happening. I see this I when I talk to people like in the ecosystem, in the industry, on Twitter. It's like there's people who have dealt with this really well where they have a very sharp, simple process that they iterate on and they know that it's never finished and they're like extremely. Extremely effective. We have had that, right? We didn't start off with the software factory that we have today. We've evolved it over time as we've learned more, as we've wanted to put more pieces in it. At the weekend, I put in a change to the software factory because I was thinking of the governance layer and I actually leaked how a piece of governance was actually happening when I wasn't able to do it. So we're continually evolving it. There's a bunch of people who are trying to overthink this way up front, and it's exactly the same thing. They're trying to carve out like this structure that may or may not exist. The same thing played out. with DevOps. The same thing played out with SREs, platform engineering, all of these things, right? This was a very pure set of goals and outcomes that we just ended up fucking up in a massive way because we overthought it and vendors became a big thing and all of this drive. So it's it's just it's playing out at a faster rate than we've ever seen it before in my opinion. Nick: I talked to a customer today and you know what they what they have is they have a platform team who created who created a software factory like primitive for all of the slop cannons that the rest of the company did. And so like if you're just a random user inside of the company and what you want is like read-only access to a data lake or to some, you know, product analytics or whatever. They have a pipeline that you can vibe in a Docker container that mocks all that shit out and creates creates the dashboards or creates the views or whatever that you want. And then they ship that container to a production infrastructure and it actually loads all of that data up. And so like they went all the way to the other end of it of like, you know, really facilitating people writing all of these apps. And they're like, it's been so successful that now we have the opposite problem where we have. thousands of bespoke apps that people have run because they're going wild. And now it's like, well, which one's up to date? Who's actually looking at it? Was it a one-time thing or was it a like a, is it a full-time thing? how do you how do you make this stuff real? I just think that's fascinating. That yeah, I think the name of the game is gonna change over time. and what you need to provide and how you need to provide it. And this was just an interesting example of Adam Jacob: Conway's law for you listening at home is that you basically you ship the org chart, that whatever the technology system is you're designing will be a reflection of the communication structures of the organization in which it was built. Paul Stack: And we're lucky because we're super small. So our chart is incredibly easy. It's it so it it shows it in the software structure and the and the process that we've built. Other people, and we talk about this a bunch, don't have that same affordance. They're stuck in larger enterprises where they have incredible levels of of detail. And I I think that there is there is another bunch of gatekeeping in there where there is a bunch of mid middle management that are really trying to hold on to the structure that they have, they're moat. And that's causing issues with regards to how software factories are created. So I could rant on this all day, just saying that. Jamie Stormbreaker: Yeah, I'm I will double down on the idea that the transformation that we went through from cloud and DevOps is happening again probably at like a 10X or 20X scale. It's really hard to measure it right now. One of the things that I'm very happy and proud of is like watching anyone ever use my developer tools in figuring out how to do something better, especially if they never came from that field. so for anyone watching, I watched a game in June called Swap Meet. as a third person shooter has these boss fights that are are like pretty complicated. I was the game director of it and I designed the first two boss fights and then I let the animation director, another animator, a concept artist, and our art director designed and then implemented the third boss. I oversaw it but I didn't do it. And they basically used all the AI tools to to make it happen. and so making enabling people to have their ideas come out is super powerful. And I'm just gonna say that at least for the next a little bit of time, when your organization wants to go through a transformation, you're gonna call up somebody that knows how all this stuff works. You're not gonna just hand it to somebody that's like, I have Claude. Like, yeah, I get it. Like you have all this power, but you need to know why you would be reaching for something. Do you know what GDPR is? Do you know what pri every application still needs to be secure, performant, compliant, needs to respect your privacy, it needs to be highly available and resilient. and you need to be you you also need to know how to like ship it rapidly, answer player or customer feedback, integrate it, you know, like do you know how to do all of that? and I would wager that engineers, they just kind of need to understand your primary role maybe is not opening up VS Code anymore. Instead is being a domain expert and making sure that you can, you know, explain to a really large organization, even though maybe you are working 10 or 20 times faster, why the things that you're doing are really critical. You know, we're going to see a lot of applications come out that get five coded and they are not one that you're going to want to spend money on. Adam Jacob: Totally. let's let's jump to Goldman Sachs. So Goldman Sachs very famously was like, We're giving everybody access to clawed code. Everybody's gonna go. We're gonna get it done. And now has very famously rolled it back and said, actually we're gonna do that for engineers, but for the traders and the bankers, they need to work with engineers in order to in order to solve these problems. And I think I think we're gonna start to see that happen more and more, right? Where the yes, you can vibe code. a solution and there's some real value in being the domain expert who knows what you want because you can express what you want in a way that you know you would have taken a lot of time to sort of download into somebody else's brain. But that process of like how will you build it as good software? How will you use it? How will you deploy it in a way that's safe? How will you think it through, you know, like those pieces of it clearly still require some some humans to do Nick: Yeah, that's what we learned Jamie Stormbreaker: Yeah, the Nick: at the event that we were at last week, right? Where what people are doing is using the domain experts to create a prototype to go into a planning meeting to make it real. And I think that's really interesting, as a as a pattern that the enterprise is seeing to be successful. yeah. Jamie Stormbreaker: Yeah, I I think that we're gonna see more and more of somebody expressing themselves very quickly by prototyping something. but I'm positive that no one wants something vibe coded when it needs to h conform to PCI compliance. I have it on good authority that that sucks to go through. there's no Adam Jacob: Yeah. Jamie Stormbreaker: way that you're just gonna let a random that has never had this dom never had the experience or the capability of understanding how to achieve that. Yeah, you're just you don't want that for your business. Paul Stack: The friction that we're gonna go through there though is crazy, right? think about it. What what did we do before? We implemented DevOps as a way of breaking down barriers between Dev and Ops and making sure things happen. Now we're just putting up a different wall and throwing across a vibe coded piece of slot and then saying to an engineering team, Hey, make this work. Do something with it. Yeah. Yeah. Adam Jacob: Yeah. Make this make this make this make make this pile of slot be good. Yeah. I think ultimately ultimately I I don't know that I think that's the gonna be the way. I think it's gonna be more like how are we how are we thinking about giving structure to people that makes it so their prototypes can go further, and Yeah. Totally. Paul Stack: Tool tooling's gonna help a lot of people here, like with sensible tooling and sensible options. And we're all we already see that day to day with our customers, right? Once we put that sensible tool in in place, it allows them to move at a different rate across all aspects of their org. Adam Jacob: All right, let's jump let's jump out of software factories for a minute. Let's talk about Jensen. Turns out Jensen Paul Stack: Ha ha ha ha. Adam Jacob: did in fact have an affair with the Chinese spy. Nick: You heard it here first. Adam Jacob: and yep, keep called that one. and NVIDIA has opened an investigation internally. you know, they have full confidence in Jensen, blah, blah, blah, blah, blah. He apologized, sent a note to the firm, you know, for his indiscretions with his wife, how he hurt his family. His kids, by the way, work in NVIDIA. So, you know, let's hope they didn't learn that one from the memo. I'm sure they didn't. and I think the yeah, you know, but back to that notice a couple of weeks ago that they sent to the anthropic engineers, like, honey traps, it's real, you know, like Paul Stack: There's breaking news that came out today. It's not just NVIDIA. Sources confirmed that Google's Gemini team was also leaking information to China and China said, please stop. I'm kidding. I'm kidding. I'm kidding. Adam Jacob: China told them to stop. that's funny. I see. You're I I got the joke now. Right. It's Gemini being like, do crimes. Nick: It's a joke. 'Cause 'cause Gemini sucks or whatever. Adam Jacob: yeah, I it's you know, how this is connected is interesting. Like the as we've learned more of it, like she wound up a high level executive in a company that basically exists just to siphon GPUs into South Asia so that China can rent the GPUs from South Asia. so places where like they are allowed to export. He did start lobbying harder to allow for export to China and that we were making a mistake. So I think this is gonna get gnarlier. and Nick: It's just greed. Of course, if you can sell if you can open up a market to, you know, one of the greatest economies on earth, then you're gonna want to. I don't know that it's needs to be that nefarious. but you you know, could be. You know, so Adam Jacob: Yeah. Well, but it is when it's yeah. Nick: so I I've got two I've got two pieces on this. Number one, apparently our intelligence community has known about this affair for more than a year and sat on it and didn't tell Jensen that they knew. and they've been watching it. Adam Jacob: Well, I also read that I also read that he hired a crisis comms firm a year ago in advance, being like, Hey, just a FYI. Nick: Yeah, you you're used to just slipping in the NDA. Turns out you also hire a crisis comms firm as well. So that's fascinating. my my favorite conspiracy theory, I was told that we need more conspiracy theories on this podcast, so so I'm bringing it, you know, is that normally the Chinese spy is the honey trap. But in this case, what might be happening? Is that Jensen is feeding false information, to the Chinese through the spy. And so, there's an Adam Jacob: he's a super spy? you think he's a cutout? Nick: opposite term for honey trap that I won't use, but that's the idea. that he that he is acting in that way. And so the conversation back home is probably a lot easier. It's like, honey, I did this for Adam Jacob: Had to do it. Nick: national security, you know. I I didn't do it because I wanted to. I did it for the love of the United States. Adam Jacob: I don't think that argument's working. Paul Stack: For the love of my country and the love of the game, right? Nick: Ha ha. Adam Jacob: Yeah. Jamie Stormbreaker: Yeah, I don't know. I don't think that would work with with my wife. No. Adam Jacob: I don't think also, Nick: Ha ha ha. Adam Jacob: he's been they've been married since high school or something, very, Nick: Yeah. Adam Jacob: very long time. So oof. I don't know. It makes me feel bad. And now she has to be covered now all of social media is just like photos of his wife and photos of the Chinese spy next to each other, being like, Look at what happened and like that seems awful, you know? Like, I think if I'm if I'm if I'm Jensen's wife, what I'm doing right now is moving to the country. Paul Stack: File for divorce. Adam Jacob: No, I'm just gonna hide. I'm just gonna hide. That's what I'm gonna do. Paul Stack: let's be honest, the divor the divorce settlement will be spectacular, just saying, so Adam Jacob: I bet she doesn't do it. I bet it I bet they work it out. Okay, we'll see. I know nothing. I have no insider data from any of these people, but yeah. But yeah, you heard it here. If if you meet a particularly spectacularly attractive person and they seem super into you and they're also pumping you for technology secrets, just say no. Jamie Stormbreaker: Yeah. My my favorite commentary on this was that they were hacked by C V E zero zero zero zero one, like the very first one. You know, yeah. Nick: Yeah. Adam Jacob: Yes, the original. Jamie Stormbreaker: I think this is another f friendly reminder that humans are still insanely critical in software development and building companies and teams and like teaching people about social engineering and protecting yourself, like again, taste, people and judgment. you're gonna need that still. Don't throw that away. Adam Jacob: Yeah, you're gonna need it for sure. Okay, let's talk about Occupy SaaS. So occupys.com, they're gonna rebuild all the SaaS platforms from the engineers that got laid off from those SaaS platforms. So engineers who got laid off from Salesforce, rebuilding Salesforce with their token accounts. they've got a list of all the different software they're going after. you can apply as a laid off engineer to go like join them. And and re-implement your former employer's SaaS platform under the Occupy SaaS banner. my hot take here is that the Chinese model makers should give them free tokens. That like this is an obvious layup for DeepSeek to just be like, here's some here's a big pile of free stuff. Go get it, Salesforce re-implementers. Paul, what do you think about Occupy SAS? Paul Stack: It th this one's new to me. Wow. just wow. how many times have we said that there's no moat in software anymore? And like how like how many times have people said that like SAS is dead? Like, Jesus, this is kinda terrifying, but kinda also awesome. Like, you know, there's so much like technical understanding in these people that have been laid off that They will genuinely create good pieces of software that are free and easy to use. We've seen it with the Adobe suite already, right? We've seen a bunch of pieces that are there. This one's gonna play out very interesting. I wonder If the companies that laid them off are gonna go after their NDAs and they're gonna go after their severance bonuses and all sorts of stuff, I think this is litigation city coming up based on this. I really do. Adam Jacob: Yeah. Paul Stack: And I think Jesus, this is this is a mess when that happened. Yeah. And Sales Compel, which is Salesforce, Adam Jacob: Well, off zero's the first one. yeah. Paul Stack: is the and they've already made releases of both. Adam Jacob: They're also pretty clear about their clean room policy, blah, blah, blah. I don't think that saves you, but but yeah. Cool website, protest movement as as software development. Jamie, what do you think? Jamie Stormbreaker: Yeah. well the first is there's gonna be litigation no matter what, as we go forward because there's gonna be a lot of hurt feelings. One thing I'm gonna say, you know, well, first of all, in America you can just sue anyone, so sure. but Atari versus Nintendo of nineteen ninety two says that a lot of this reverse engineering is totally allowed. As long as you can see the source code and you're reverse engineering it, you're okay. The problem with some of these folks is they have seen the source code, so they've definitely opened themselves up. So this is gonna be really interesting. if you want to go reinvent and rebuild Salesforce or some other software and you've never been given access to Perforce or Git or whatever the version control was, you've never seen it, I think you're a little bit more safe. You can still get dragged into court. but I probably would not join this if I've ever seen the source code because you're definitely open. it is definitely an interesting idea in general. I am a secret anarchist, so like I love the idea. that people are just taking power back. I thought open source and, the entire idea of the Unix and it came from the educational institutions and the military. You know, like this is supposed to make our lives better. So I like the idea of destroying something that is really big and doesn't deserve to be there. Or like I don't particularly like that it got as big as it did. but that's just my personal opinion. I think that I would be careful though if you worked at any of these studios and then you've seen source code and you want to try to join that movement, I'd be careful. Paul Stack: They do have a cool song though. I don't know if you've heard it. It's on their it's on their GitHub org. They have actually their their own song about the back of it, which is funny. So rather than going after a Nick: I think I think this is a Paul Stack: bunch of things, they build a song first. Nick: I think this is a leading in Jamie Stormbreaker: Really? Nick: indicator of the anger that's out there around, Adam Jacob: Totally. Nick: all of those things. And so, like negative or angry energy needs to go somewhere. And I guess that's working for free on replacing your, you know, your previous employer. But yeah, I think that this is this is not a good sign, you know, that this thing exists you know, in in this way, as a result of AI. and it's you know, back to anyway. I I think that this is a problem that we need that we as a s society need to solve in the near future. You know? I don't know how. 'cause layoffs will continue but until morale improves. But yeah, anyway. Adam Jacob: Yeah, no, it's a it's a it's a it's a legit one, the Occupy SAS. Okay, let's talk about going the other way. OpenAI bought the Pierre Computer Company. We've talked about the source control re-implementers and new source control folks. Pierre was in that sort of vein of like high scale Git redesign folks. OpenAI picked up Pierre. I feel like I feel like we're in an era where the it's clear that you have to rebuild the shape of how source control works. GitHub rebuilt their back end. you've got the the jujitsu folks who are building one, you've got the entire one, you've got a bunch of others that are sort of coming online right now. And it's it's clear that source control is like up in the air and gonna change. so you know, good for the folks at Pierre. Congratulations. getting that open AI stock and And getting your getting cash, I'm sure. and like I wonder what it means in terms of thinking through the potential for those companies sort of in the market. You know, I just think it's a source control's up for grabs. It hasn't been for a very long time. All of these solutions, in my opinion, are just not weird enough at all. You know, like we could get much stranger. So I'm looking forward to the weirder ones. and I think it was smart that if you had an exit you took it. So Nick: Yeah, I'm really surprised on I was surprised on this one because, you know, we're pretty plugged in to the tools and technology that let's say our peers and people along the frontier are using because we're talking to those people every day. And like I d I don't know anyone who used code dot storage or, you know, a PR product at the end of the day, and yet apparently they have something unique enough that that's was acquirable. Or the team was thinking about it right, or whatever. who knows what the actual story is and what the dollars were, on to whether this was a success or not at the end of the day. But yeah, just fascinating that that's you know, they're they're acquired without much real external traction that you can see. But maybe they're just killing it somewhere that we don't see. Paul Stack: Well, one of Adam Jacob: Yeah, or they killed it they killed it with an open AI, Paul Stack: well one of their most famous people were lovable. Lovable were doing a bunch of stuff on top of them, bolt. So, you know, there's there's definite like, you know, big scale usage in some of those areas, which I think, you know, that's kinda why it lends to me to be that it was picked up, but you know, definitely not an everyday tool, but like, you know bunch of super smart people over at Pierre Computer Company and fair play done. I love seeing a big exit like that and people taking it people able to take advantage of that does does just show there's a gold rush and fair play, for taking advantage of that gold rush at this moment in time. Adam Jacob: I didn't see any info on its size. Jamie Stormbreaker: to to your point, Adam, Virgin Control has for a very long time not been vulnerable. virtual control for a really long time has not served everyone. The games industry has had a very challenging time because Perforce, while it's pretty good software, it does not have a great flow like we have with like a PR flow, C I C D making branches. Branch is all heavyweight, so it's like subversion centralized virtual control. powerful for artists, not great for everything that's not just a large binary asset. And then we have Git with Git LFS, Unity, you know, threw their hat in the ring, said that they were gonna work with Git and GitHub, and then three months later abandoned it silent for seven months until they eventually said, like, we actually bought this other company that we're gonna use as version control called Plastic SEM, which has a hybrid model. I think for a long time this has stood because it's hard to get it right. It's also the second thing you establish when you make a company. Maybe the first one's your domain, so that way you have email. And the second one is that you set up what version control you're going to use. and it's hard and not very it's very thankless to build a version control system. so I mentioned earlier, like I'm off GitHub. I'm literally building my own version control system at the same time. Nick: Yeah. Jamie Stormbreaker: I know exactly what I wanted to do because I made games for twenty years. and I can be on a billion side quests. It already powers everything that I'm running. Claude identified, sorry, anthropic. They also created a thing I think it's called that it's not fossil yeah, sigil. Adam Jacob: Yeah. Yep. Jamie Stormbreaker: and while I don't I they haven't showed any code, they've just talked about the problem. what they're really saying is that with a genetic workflows, I'm gonna use the highway a analogy, but everyone in LA is gonna tell me to fuck myself, right? Because the more lanes on a highway, it actually doesn't make it better. But for computers, no one's making left and right-hand turns, you know? so you have to increase the bandwidth when you have all this stuff coming into the system. And Git falls over eventually, it turns out, when you have the throughput that they have in their system. so I'm really interested to see what the future of version control looks like, especially because for years I've just watched especially artists in the game industry suffer. Because either you had to use Perforce and all the engineers are suffering, and so do the players, because you can't actually test features. You can't release as fast as you possibly can. Or using Git and then you're watching artists accidentally change things in a scene or, large files overriding each other's animations. and even though we have Git LFS, it does optimistic locking and not pessimistic locking because it's decentralized, there's no central place that you can connect to and say, This person has this lock. Now it's released. Adam Jacob: Totally. Jamie Stormbreaker: nothing integrates with Unreal, nothing integrates with Godot, nothing integrates with Unity. so yeah, hopefully this is one thing I probably will open source if it's worth it. also it looks like other people are doing it too, so maybe I can give up my side quest at some point. Adam Jacob: Yeah, the weirder the weirder we get on source control, the better off we're all gonna be, I think. Okay, that's it for this week. so let's move on to some questions. Jamie, you ready? Got five questions for ya. So of the stories we just covered, which one are you gonna be thinking about later today? Jamie Stormbreaker: man, I think the most interesting one for me i it happens to be the version control because I'm working on it. I'm quite Adam Jacob: Yeah, yeah, yeah. Jamie Stormbreaker: literally right attached to it, right? The idea that things that have gone stale for long periods of time, even if they're good, like I'm not saying Git has not served me so well or GitHub, but that we can reinvent software based on the the trans the transformation that we're gonna have to go through for organizations and teams. like we built software with the idea, and this even goes back to Angler, that humans need to use it in a very specific way. And I think we're gonna start seeing a lot of software, including virtue control, come up where I have a bunch of shadows of myself that are working alongside me and I'm the human gate on top of them. and other people have other shadows. That's at least what I call it inside my system. They're just agents, right? and you know, I think you're gonna see this across not just virtual controller, you're gonna see it on issue tracking, planning software, wiki, build services, I mean, the the the the whole gamut. either these companies are gonna reinvent their software or someone's gonna c come along and remake it for them because the lanes are too tight. you you it's it's not the problem that agents hallucinate over time. it is very true that if you let the if you let more than three of them talk to each other and you let them talk over time, more entropy gets added to the system and it's a nonlinear function. But if you isolate them, you still have the same problem where a thousand things are doing work faster than a human could have. and if your software that supports development doesn't prepare for it. It's gonna fall over and you're not gonna be able to you're not gonna win the race. Adam Jacob: Yeah. All right, which one's noise? Jamie Stormbreaker: Jensen. I don't give a shit. Paul Stack: Ha ha ha ha. Adam Jacob: Don't even care. It's a it's an international spy scandal. but you could get Jamie Stormbreaker: I know, but the the best part of it was the conspiracy that Nick hit me with. dude cheating on Nick: Ha ha Adam Jacob: Ha ha ha Jamie Stormbreaker: his wife that is a person I didn't really care about in the first place. I don't care. Adam Jacob: Yeah. Done really. I got you. I got you. What's a common belief in the industry that you think is wrong? Jamie Stormbreaker: that everyone's gonna lose their job and people should take less salary and that you'd have not no defense and that you you know, you you should be scared that you can't get jobs, especially if you're coming out of college right now. If you're listening to this and you are a college student and you were graduating, I know it's gonna be painful, but there are people like me and the people in this room that are working to make sure that you're empowered. and that organizations will understand that cutting headcount is n might might lower your current operating budget, but it also is going to stop you from being able to compete against the people that you don't even think are competitors that are in the shadows right now because they have power you did not even understand. Adam Jacob: Yeah. All right. What's the one thing you wish you could just teach everyone? Jamie Stormbreaker: to this one is particular is however you've been doing something right now, just reconsider it with an open mind. take every feeling you have, all politics out of it. Just think about how you were doing something before and try something new. Even if you think that you know generative AI is destroying the world. yeah, sure, there's a bunch of slop. Yes, there's bot networks that are all over Twitter and it blows, right? But give it a shot because the same tools ki in the right hands can can do good things and not just bad things. And if you are not considering that and you there's only there's three things stances I think you'd take. one, move forward. The box is already open, it's never gonna close. Two, put your head in the sand, three, run away. two of those are the wrong thing. adopt it, move forward, just bite your lip until your shirt is red. Right, it's okay. but remember that we are the ones in power right now. You can make anything that you want, and you can learn anything right now easier than ever ever. and I think that would probably be a mindset shift that I hope people would take. Adam Jacob: Great. What's the best and worst moment you've had working with AI agents? Jamie Stormbreaker: the worst moment was when I took all the guardrails off because somebody said back in June that they could you could just prompt it, it'll just do all the work and I let it run for a week and it just blew through a bunch of tokens and chewed itself up like an engine that was low on oil and I just kept driving it. it does not work like that. There's no fucking way that you can just prompt something and it just solves solves the world. You cannot do that. those are those are fake ads on YouTube. that was the worst time. I would say the best time I already mentioned one, which was watching, not engineers design an entire boss fight that included all of the AI like that had to be defined. You know, AI in games is different than the AI that we've been talking about, right? but I'll choose a different one. I went down to downtown Seattle to visit a friend that is a CEO of another game studio. And I was talking to him about what I've been doing, which is like helping local businesses with their software solutions, just con continually to c like, you know, I'm the CEO of a game company, but I also like to walk around and get lunch and help people with problems that I couldn't I I never had the time to do this before. There's so many side quests now, right? So we're in his office and I look and I'm like, what is that? And he's got like this what looks like an old Steam, you know, like friends list. And he shows it to me and he replaced Slack in Zoom and made his entire collaboration suite on his own. It's like an electron app. And I'm like, You've never programmed ever. What are you doing, man? This is amazing. And then I w looked at the looked at it, click through, play with the whole thing. He's like, Yeah, the entire team is on this. And I worked on it for like a month and it's exactly what I want. And I'm like, Can I have that? It looks better than everything I have. And so the the story there and what's really rewarding is watching people that have great ideas that are product owners, project managers. UX UI designers and now they just have the full capability to not just prototype it, but you could really make a thing that matters. I would just worry if you needed to scale and be compliant and private. You know, this is the internal tool, so it's okay. Adam Jacob: Yeah. Okay. Well that's it. That's the news for 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 swampdash club dot com slash podcast. You can sign up to get it delivered straight to your inbox every Monday morning ish. Depends on the day. If you're trying to figure out how to build AI systems that are repeatable, token efficient, solve real problems, Swamp is here to help. Visit swampdash club dot com to learn more. All right, Jamie, take us out. What do you want to plug? Jamie Stormbreaker: Yeah, actually, I I this I think this is the first time I've other than blogged a little bit about it, but like you should if you haven't checked Swamp out yet, you should go check it out. the power is unbelievable. It's just a primitive that it i I I really truly believe that you have these three gates, deterministic gates, non deterministic gates and humans. And it allows you to model every single thing in a system and make it more deterministic, which lowers your token costs, makes everything more efficient, everything more reproducible. and It truly is a game changer. It is a another thing in your toolbox that you want to add in and no one I'm not here for sales and not giving me any money. I'm just telling you check it out. if you've never tried it, literally attach claw to the website and just say, What is this? and have a discussion with it. Adam Jacob: There you go. Paul Stack: Jamie is not a spy for us, I'm just saying. Yeah. Jamie Stormbreaker: I'm not not welcome. Nick: Thanks, Jamie. Adam Jacob: Thanks, Jamie. Listen to Jamie, everybody. We will see you next week. Bye. Nick: Yeah. Jamie Stormbreaker: See ya.