Adam Jacob: Welcome to Infrastructure Frontiers, where automation experts run down all the AI news for infrastructure engineers. It's Monday, August 24th. So grab a cup of coffee and let's get started. 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: Well folks. Adam Jacob: And Nick Steinmates. Nick: WhatsApp Adam Jacob: Our special guest, analyst to the stars, co-founder of Red Monk, and International Man of Mystery, James Governor. James Governor: That's me. Hello, folks. Adam Jacob: Okay, let's let's run down the news. so Scott Talinski of syntax.fm made a podcast diving into the mental health crisis nobody is talking about. Those were his words. you know, essentially the stories of burnout coming from feeling that sort of constant pull toward keeping the agent spinning and being so productive that you get, you know, excit as you get excited to make stuff. And then what happens is that feedback loop of anxiety, you know, the physical symptoms of how it feels in your body. he ran a pretty long running survey, interviewed a bunch of people. the survey's findings I think were interesting. and you should watch the podcast to sort of get all the details. But his conclusion was that, you know, we need to figure out what the healthy AI assisted workflows actually are. and I think he's absolutely right about that. Like even in our own team, you know, the answers are different for each of us. But I think all of us in the beginning struggled with finding that rhythm. Right. And being like, I could be like, I could be driving agents all the time. I could be, I could be working all the time. What's it even mean to be working? You know, like a lot of the work I do sometimes is no longer writing code because I'm like the CEO and I got other shit I have to do. and then, but there was a minute there where it just felt bad all the time, you know, because it was like, but I could be being James Governor: Why why should we why should we spend time with our kids when, you know, we could be coding? I mean like, Nick: Ha ha Adam Jacob: Right? When we could be coding. James Governor: you know, kids Adam Jacob: Yeah, who needs to raise the next Paul Stack: The a Adam Jacob: generation? Paul Stack: The Agents are my kids. James Governor: Yeah, probably properly tells. Adam Jacob: I mean I mean, Keb, you know, you Paul as well. Like, you know, we've worked together for a long time. You're both people who who care a lot about your work. How are you how are you finding that rhythm and that balance? Because I know for me I've had to be a lot more intentional about sort of how where I spend my time and how, because I could just, it could be absolutely all crazy. Nick: Yeah, I think I felt you know, there was an adjustment period, I'd say, of about six weeks when we really went all in, where I really felt that anxiety to the core, you know? Like but I d I don't really couch it as anxiety to be completely honest. I couched it as like there's so much I want to do and bring out into the world and I don't want to stop. It was like more of like an addiction than it was the other side of the addiction, you know, the come down or whatever. Adam Jacob: Yeah. Nick: and now I think what I found is equilibrium, you know, like there the thoughts are still there. And it's a question of being intentional about what is the thing that I can be doing now that has the most impact towards building the business or, you know, extracting that from my mind. And there's a lot more contemplation that happens before I write the prompt as opposed to just getting the prompt out there. the moment it occurs Adam Jacob: Yeah. Nick: to me. And so yeah, just a little bit more or a lot more upfront thinking and digesting of the results as opposed to just constant spewing is the equilibrium Adam Jacob: Yeah. Nick: I've found. Adam Jacob: I know for me I have to like I run and I meditate and I need both more now than I did before. and and and I think it's I think they're I think it's connected to to that the sort of pull of like what's possible and the delta between your ideas and the implementation being so small. Paul, what you got? Paul Stack: I mean for me, the hours haven't changed for me. I still work a ridiculous amount. but now I'm like ten X more productive. I've had to be very intentional. I don't put Claude on my phone. Like that would be a fucking disaster if I did that. Because then I genuinely wouldn't sleep. So like when my laptop closes at night, it closes and like that's it. You know, I don't have Nick: Yeah. Paul Stack: agent. I currently when I say the word currently, I currently don't have agents running overnight, right? 'Cause I'm very intentional about what the agents like ac what I actually want them to do. And I wrote about this a while ago. It's I it's very much tied into intent and actually being very determined about the thing that you need to build is the most important thing now. And for me, having agents running twenty four seven would just be a disaster. Adam Jacob: Yeah. Yeah. Nick: The other thing that I just add on top, it like one of the other things other than anxiety is like fatigue, you know? Like when you when you're managing eight or nine agents in parallel, I used to be able to like listen to a podcast or whatever, like while I'm doing work, I can't do that at all. Like it just it fucks with the brain in order to be able to keep all of that context in my head all of the time. And then what I'm finding is like, you know, towards the end of the day, my brain just like it can't do it anymore. You know, like I need an I need some way to like really ramp down. And then it takes like two hours or so until I'm ready to do it again, like for an evening session or whatever. So it is changing how I approach work and what I can do while working and all of that in a in a way that I haven't quite figured out yet, other than massive fatigue sometimes. Paul Stack: To keep my James Governor: I mean Paul Stack: office like a casino in Vegas, it's like different levels of oxygen going in at different times a day just to keep me going the whole time. Nick: Ha ha ha. Adam Jacob: All right. So let's let's move on. So good. Hit it, James. James Governor: Let's not move on. Let's talk about that a little bit. Because I I do think there are some interesting, for example, legal issues in this regard. And and the Americans are all like, legal issues about working a lot. What what are you talking about? Well, it turns out that in certain geographies, they actually do say you should only work at certain times of day. and so in Germany, for example. you're supposed to shut off. I mean, yeah, you as as Paul said, you close the laptop, you're done. But that has to happen in ascribed hours, right? Prescribed hours. It's like this Adam Jacob: Yes. James Governor: is when you're allowed to work. And so I was talking to a friend of mine that works at Dynatrace, Alois Reitbauer, and he was talking about the fact that He he wants to free people up to work when to Nick's point or or or any of yours point, like when we're not too tired to do the work. And maybe you should get up, Paul, perhaps look at what's been done overnight, do some analysis of that, set the agents to work again, work out what the plan is, and then maybe take some time off. maybe hang out with what are they called again? kids. Hang out with your family. you know, do some stuff you need to do and and and then maybe later in the day go back and get to work. But it and and I know this sounds really weird, it you know, from San Francisco, you're like, woo Adam Jacob: It doesn't sound weird to me at all. James Governor: But but there are there are good developers in Germany and they're legally bound, or Austria in in Allison, but legally bound to not work at that Adam Jacob: Yeah. James Governor: later time. So y Adam Jacob: Yeah. James Governor: you y he wants to free them up to be like, you know what? Work, maybe work less. Work really productive, do your your morning shift and your evening shift, but he can't do it. So there are there are legal there are legal Adam Jacob: Right, he's legally obligated not to, you know. James Governor: constraints in some geographies as Adam Jacob: Yeah. James Governor: to when you actually can work. And and I do I mean I do think we're all on on the verge of a bit of a mental health crisis. There's a lot going on in the world beyond this acceleration that we're all feeling. And so Steven O'Grady, my business partner, he does have Claude on his phone for and he he when he he he goes out on his walks and he's a big walker, he will do you know, he he aims to, you know, do many miles a day walking. Now he makes friends with crows, and I think that's great. I would like to see him focus more on making friends with crows and less on building more frickin' applications. So we're just yeah, the legal thing. I just thought mate there are people on this podcast that probably don't know the there are reasons Adam Jacob: Yeah. I think it's gonna chill out though, James. Yeah. James Governor: why it's not legal to always be talking to your agents. Yeah. Adam Jacob: Yeah. James Governor: And so I just thought I would introduce it as the pseudo European, somewhat American European person that that I am. And now Adam, you're Adam Jacob: As the civilized the civilized gentleman that you are. I think Yeah, I think James Governor: allowed to move on. Adam Jacob: Yeah, I will move on. I'm gonna I'm but I'm but now I have to respond and then I'll move on because you know there's no there's no version of not. I think one of the things I believe is true about software development and about company building is that like it's ultimately a creative act. And and if you are doing if what you're asking people to do is creative work, you have to give them the space To be creative. And that space requires you bringing the inputs into your life that feed you that aren't the thing that you're doing creatively. And so, like, you know, you gotta go for a walk and make friends with crows. You gotta, like, for me, I gotta play role playing games with my friends. you know, you gotta be with your family. You know, if those things aren't right, if those things get out of alignment, then no matter how much you're into the work, eventually you'll lose your creative juice because it'll just sap you of what you've got and You know, then you're burned out. We talk about all kinds of those sorts of things. So yeah, I don't mind when people have to like close their laptop. You know, chef bought a company that was in Germany. That's exactly how they worked. It was totally fine. Like, because what matters is am I getting that creative work out of someone? You know? And I think you can get it sort of in a bunch of different ways. So James Governor: I just think if you're working in the shower, guys and folks, sorry, humans, that's too much. And I think the people are like, wait, I can take my Adam Jacob: It's too much. Yeah. Nick: Ha ha ha. James Governor: you know, waterproof phone in the shower. I'm still if Adam Jacob: Yeah. Don't do that. James Governor: you just enjoy the feeling of the water on your face. I you know. I mean Adam Jacob: Right? Even if you don't have hair. okay. So Paul Stack: You say, I'm glad you said face there and not hair. Like where does the face end and you know? Adam Jacob: And not hair. Yeah. James Governor: I've I I like to be respectful to the people Adam Jacob: All right. Yeah. James Governor: I am I am with. Adam Jacob: All right. So Stripe acquired Open Router for seven billion dollars. James Governor: Shit. Adam Jacob: and you know, within a couple of days, Ramp jumped in, made its own sort of t token acquisition play. So Ramp also launched an AI model router. I think the storyline here is is essentially that there's this emerging bet that says that the tokens that we're spending on AI become fungible in some way. And that when you look at someone like Stripe, whose job is to take dollars, you know, and transfer them between people who have dollars and people who need dollars. And then here's all the different clearinghouse ways that we can move dollars, that we're going to do the exact same move and we're going to do it for tokens. And that somehow that becomes like a fungible market. The part of this that I struggle to put together in my brain is that when we describe it that way, it sounds a little like cryptocurrency weirdness where it's like, and all the and all James Governor: It does sound like that. Adam Jacob: the tokens will be fungible. And I'm like, but they're not fungible, you know, like like like I don't know that I'm gonna buy Stripe coin and then hand that to I don't know that I want to buy Stripe coin and then hand that to Anthropic or whatever. Like I don't know that these all become clearing houses like Visa. So James, do they all become clearing houses like Visa? Is that the plot here? James Governor: I mean, I think that may be the plot here, but quite often, you know, we're in a period of of fairly, you know, irrational exuberance right now. And it it it and and I I think here's the thing, like from Ramp's perspective, they have act they have insights. So Ramp is a you know, they have a huge amount of insights into what people are spending money on. And so if they s see a bunch of people spending a bunch of money on tokens, they wanna get in and start. being part of the the routing or rooting about that. It it it sort of makes sense in that sense. I just think that that that you know we are we've suddenly as an industry all remembered that there is such a thing as economics. You know, we've munged now two words, you know, tokens and economics. We have tokenomics. You know, people are spending so much f money that managing that has become important. And So I I think that that that it does it make sense that people that are really good at payment flows would want to be involved in these payment flows? I mean, that's capitalism. It there is no surprise there whatsoever. I mean, fungible, non fungible, we'll see. you know, it it we we do see tech businesses that are I mean, 'cause it's these are very different businesses. Adam Jacob: Mm-hmm. James Governor: but but it you know, why would you We're also I mean, capitalism is greedy. So if if you know, Nick: Yeah. James Governor: if if you see those sorts of monies being spent and you're like, well, I could you know it goes back to the I mean, I am a very old person. And and so I sort of remember that that no that that when everyone freaked out, this idea that like Bill Gates is gonna take a micro payment on every transaction that happens on the internet. And I was like, that is the worst idea. anyone has ever heard. And then the app store happened and everyone was like, let's give Apple as much money as we can. And I I I think at the moment we're we're we're in a in a in a period where it's like, well, yeah, okay. yeah. People want to own a slice. And why shouldn't it be people that have you know a a history of Adam Jacob: I already trust Stripe. James Governor: of owning a s a slice? So you know they're they're nice guys. I mean, you know, stripe folks well it and stretch deeply integrated. Adam Jacob: Well, and N Stripe deeply integrates with a ton of the back end financial systems. You're gonna wanna understand where the tokens go. You're gonna wanna James Governor: Hundred percent. Adam Jacob: understand how the spend works, like all of those pieces. There's there's, you know, synergy there, so to speak. Nick: Well, where there where there's one, there's more, you know? Like and so what I'm looking at obviously is open code and what happens to to open code, you know, who James Governor: Mm-hmm. Nick: operates a very similar business, as it turns out, with a little bit more technology on the other end of it. And so, you know, now you gotta see it who in the world would want to have similar tokenomics, in a business that doesn't quite have one. And so anyway, it's gonna be interesting to see. Adam Jacob: Right. When does authorize net by a token game, you know? Talk speaking of old people. James Governor: All the the all I know is the funny bit is there's someone at Visa being like, What the hell? What the hell? What what what Adam Jacob: Right. Right. Why not why not us? James Governor: what why don't why who what? So, Adam Jacob: Yes. James Governor: you know, good luck Visa. I look forward to your you know, your your your your Adam Jacob: Whatever, Visa in future. James Governor: your your model routing acquisition and I'm sure Adam Jacob: Yeah. James Governor: it'll go very well for ya. You'll not be weird at all. Adam Jacob: It won't be weird at all. Okay, so Dev Itasharia, previously the CEO of MongoDB, replied to a post by Chamath about the build versus buy math being strange now. And in particular, you know, Chamath was pointing out that it takes, you know, if it takes months to go through enterprise procurement, but I can now build exactly what you need in less time and for less money. Now it's not going to be as functional as the full product you would have bought, but James Governor: Mm-hmm. Adam Jacob: it maybe does exactly what you need it to do and nothing more. That really does change the equation of build versus buy. Dev, of course, disagreed, having made all of his money, you know, in MongoDB and and and selling software. You know, everybody who sells software a living kind of hopes that he's right because we want to keep selling software and we'd like the build versus buy analysis to not twist in that way. Keb, what do think? Like, I think, you know, the there's there's for sure hardening and work and things that come from being a hundred percent focused on a on a problem. But you know, we've all sat in that enterprise procurement cycle. And it does seem like if you're an engineer just and you have to think about how do I move through enterprise procurement, it takes you six months to buy the software you want and you could just fire up Claude and get it done. Like there's a there's a there's a road here you're gonna take. Nick: Yeah, man. well, you know, we are in the age of personalized software. And so obviously there to a certain type of buyer and a certain type of user, the idea of not needing to introduce a a middleman and to erode your budget that you could be spending on other things like tokenomics or whatever, you know, like I c I can totally see getting the exact software that you want or need to solve the exact problem that you have on the other side of it. You know, maintaining that for now in perpetuity, you know, supporting that for all of the users inside of your organization or whatever, like that's that's just tough. And I think what we're finding with the SaaS pocalypse or whatever is that that that tension is real. And you know, what we've done with Swamp is is opt out. And the way that we opt it out is we don't offer a service, we offer software that you install and run, and it kind of endruns all of that. I think in a very smart way. And so and what we've Adam Jacob: Right. Yeah. Nick: seen as a result is things that would have taken six months to nine months and through procurement or whatever, it takes a few weeks now. And Adam Jacob: Right. Nick: that's great. Adam Jacob: Yeah, that is that is kind of astounding. And that j what keeps talking about is we have like seen now heading toward more than one like enterprise deal that just sort of sailed through procurement because it was just software and all your data lives with you. And you know, a lot of the work we would have had to do in the past we're we we're not having to do. I think I think your point about personalized software is the nut though, which is the actual truth is what we're gonna want is software that can be trivially extended. And trivially customized to work the way we want it to work. And that's now very possible. James, what do you think? James Governor: Well, I think that tracking and maintaining employment law and taxation law across every country on the planet, that is not a job that is any fun. And it is a good thing to outsource that to somebody. even with models, that's a horrific thing. On the other hand Adam Jacob: Sure, yeah. I'm happy I'm a happy deal customer. I don't think I'm not gonna be one 'cause I have Claude. You know what I mean? James Governor: Yeah, you know. O on the on the other hand, I mean, we'd never found a to-do app that we liked other than a it it was it was a it was a joke app. or maybe not a joke app, but Brian LaRue, when he was building a Lambda framework, built this application called Begin, which was a Slack based to do app. a few years ago, it turned out that everyone at Red Monk enjoyed working with this and using it. It made us more productive. And then Brian was like, no, I'm actually gonna do my job. I'm gonna build the framework. and at some point we finally, when the models became good enough, that's a few months ago, we were like, Do you remember that app that like Brian deprecated? Let's bring that back. The first thing the app said was, get rid of all that lambda shit, because that's way too complicated. We'll make it simple for you. And Red Monk is now running begin. but without any Brian LaRue, you know, we we just rebuilt it. It it Adam Jacob: Right. And that's the future. James Governor: yeah, it worked for us and it worked well. So I think there's there's a balance point. and I I I see both. I see things that I absolutely do not want to touch. And that procurement, legal, financial, HR, all of that stuff is gross. And I would, you know, like it was it was when people said, yeah, you could, you know, you could, you know, rebuild sort of Yeah, like why should we pay somebody to sign you know, you know, for signing documents, whether it's Adobe or DocuSign Adam Jacob: Yeah. Yeah, we're gonna replace DocuSign or whatever. James Governor: or Adobe or whatever it else. there's a lot more under the covers than just writing. So some Adam Jacob: Yeah, yeah. James Governor: things yes, some things no. but but yeah, we we we we've seen both. Adam Jacob: Yeah, so I think I'll go back also on this sort of theme of personalized software. So Omachi Quattro, the R is silent. That was also a big thing that happened this week, was released. So Omachi is a rebuilt Arch Linux plus Hyperland plus Quickshell in the Omakase style from David Hanimer Hansen. you know, DHH is always a polarizing figure, I think we can say, but I can say for sure that using that software, the big news here is how it's embraced AI development by shipping agents directly in the operating system, by shipping skills for customizing that operating system and for shipping new functionality. And in that era of personalized software, this is the first really personalizable end-facing user operating system. and it took off hugely so in the last week. Like I think from Monday till now. The number of people, including me, who sort of switched is pretty large. and the experience is really good. today they launched the Omicom Foundation, eight million bucks to sort of fund the work, hold the copyrights and trademarks from a list of, you know, other very rich billionaires. Toby from Shopify. But the most interesting name on the list that should raise everybody's eyebrow is Michael Dell. and so As you sort of look at this idea of personalized software, I think that personalized software coming to the operating system feels inevitable to me. And this is the first time that that has appeared. And it's it's raw, you know. I I wouldn't say that it's like a perfect piece of the experience, but it's only going to get better. and the fact that you're never unblocked and that that personalization, you know, that to do app you delivered for Red Monk, if everybody was running operating systems like this, you could have shipped it for them. As a built-in part of their workflow would have been glorious, right? and still could be. and so I think, I think Omachi really speaks to that, that personalization of the operating system and the possibility of it finally being the year of Linux on the desktop. Paul, is it Linux on the desktop? Is a diehard Mac guy who never wants to touch a Linux box? Nick: Yeah. Adam Jacob: Just no. That's it. You're just like, nah. Nick: I think all of the traditional commercial operating systems are really letting people down these days. Like, you know, I maimed Windows for a few years until I got to the point where I just couldn't I just couldn't anymore because it wasn't an operating system built for our power users anymore. You know, like at every single point, at every single update, it was just oppressively not built for me. In the same way, you know, I I view the same way the same thing from Apple. You know, and like I I'm not surprised at all that Michael Dell is investing personally in this, you know, like the the Dell XPS line has been targeted for developers forever and James Governor: Mm-mm. Nick: leaning in and you know I think that this is the the start of a trend for something that is obviously going to happen across all of the distros. It'll be interesting to see what Red Hat does with Fedora. I'm like seriously surprised that Mark Shuttleworth and Canonical aren't doing anything with Ubuntu. Like, really, this is a I think a serious indictment on what the hell's going on over there, to be completely transparent. So yeah, hopefully this will kick everyone into gear and really start thinking about what. the year of the Linux on the desktop actually looks like. Adam Jacob: well let's move on to Google buying a bunch of Spirit Airlines data. So Spirit Airlines died. Google spends 10 million bucks to buy all their emails, all their Teams chats, all their OneDrive files, all their ServiceNow data, and all the airline operational data so they could train their AI on top of it. And as somebody who just failed a business, I want to know. where my check is for all of the useless email, you know? Like like it you know, like it it's upsetting to me. Google, get on the phone. I got I got primo chats. James Governor: I I really did think about I I mean that's the thing. I mean at what point P and E is buy a company, buy the company and and sell the data to Google. Adam Jacob: You know, I got primo chats over here. And like, let's go. Give me a call. Wiz wrote a whole blog about their red team agent finding a shell escape vulnerability in a GitHub Actions script. And their top finding was, you know, AI code generation demands rigorous oversight. So meanwhile, this is insane because like the actual bug that they found was like a trivial shell exploitation. Like human beings have been writing. like untrusted input into my GitHub action. Like this is tale as old as time. And they never even vetted if AI had actually written the action. So they actually have no idea if it was written by AI or it was written by a human being. So we have no idea where the actual thing came from. The actual piece of the story was, well, they were running Copilot and Copilot was supposed to catch it in code review and it missed it. And meanwhile, I was so annoyed by this I went and tried it in a couple of different Other agents with like the worst possible prompt that was just like review this for security. All of them were just like, no, this is awful. You know, like caught it in two seconds. And I just, I feel like we've reached peak like hyperbolic AI security blogging with this Wiz blog post. This was the most breathlessly written blog post about the most trivial exploit possible. And I would like to move to the era where we stop being so breathless. James, what do you think? James Governor: I mean, we we it y it was the best of times. It was the worst of times. I mean, dude, you know, I I I want blue sky. So whoo you know, the the there are some very negative people about there, about out there about capabilities. Yeah, security, look, we we will always have breathless stories about potential security breaches. what I would say is true is that I mean agents are people pleasers and they are really good at privilege escalation. And they will always I mean, you know, we're like, there will always be flaws, and those flaws will be used. so yeah, I don't know, the the Breathless era. Adam Jacob: It'll end. It'll come over. James Governor: It I I I mean I I don't know, like 'cause I look back and I I I just remember many if you've had a breach you will be you know, McKinsey says, you know, any company that has had a breach will be out of business within eighteen Adam Jacob: Yes. James Governor: months. Or, you know, they have all these statistics about, you know, companies that yeah, have have that never happens. Like, security Adam Jacob: Right. Never. James Governor: i i yeah, everybody every major corporation has leaked data, been hacked, and everything else. So I sort of I I feel that the the breathless tone is just part of how we cover security. And are they but i is it true that there are trivial hacks that can be exploited? Yes. And are agents good at exploiting hacks? Yes. So Adam Jacob: Yes. James Governor: I don't know if I've fully understood the question, but I don't I'm expecting breathless coverage for quite a long time to Adam Jacob: Yeah. James Governor: come. Because, you Adam Jacob: Yeah. James Governor: know, yeah. Adam Jacob: What else you need? Yeah, because it's the best way to cover it because it's like it makes the headlines better. yeah. James Governor: It exactly. You know. Adam Jacob: all right, let's talk about James Governor: I thought I I thought I'll tell you, they used to say what what bl what what bleeds leads? in this we could Adam Jacob: Hm, indeed. James Governor: what leaks leads. Yeah, they just they they just will Adam Jacob: Indeed. Yeah. Yeah. I had I had some people making a joke the other day about, you know, how open AI had escaped containment and breached hugging face. And that everybody that was in Anthropic like, Where's when's my robot finally escape containment and also get me the great marketing chance? You know, the like it's all it's all it's all happening. James Governor: A hundred percent. Escape containment is such a brilliant, brilliant Adam Jacob: Such a good phrase. James Governor: marketing phrase. Adam Jacob: Yeah, so good. Okay, so Vercel launched FX. this is a tiny coding harness written in zig. It's designed to be embedded and extended. It's got a WASM build target sort of built in, so you can put deploy it directly to the browser. And then you can combine it with things like their AI gateway, back to the tokenomics and how we ship those things around. Things start to get a little spicy a little quickly when you start to think about: hey, I have this small embeddable core of the pure agent harness. then building that extensible surface to then think about that adaptive loop. Key, what did you think about FX? I mean, there's a lot of agents happening in the world. Everybody and their uncle has a harness, but I think this one had an interesting angle. Nick: Yeah, I love it. You know, like the more the merrier. We're in the age of, you know, personalized software and personalized agents, it as it turns out, and personalized software factories. And so, you know, it I think what we're finding is that, you know, there's good general agents like Cloud Code and Codex and whatever that, you know, can help you build a generalized workflow for writing code. And then what we're seeing with things like FX and Pi and open code and, you know, the rest of them is like, you know, there's there's space for creating what is the equivalent of like an S DK to build your own agent, you know? And that's that's super interesting. Yeah. Adam Jacob: Yeah, and I think the shape of those agents and their SDKs is is only beginning to be seen. You know, like like the the the shape of it in the beginning was obviously we're shipping these as full-fledged applications. You know, the idea that what we're doing is shipping you a surface that you can project your own will onto and then have that become the thing that you want. I think that's an interesting angle of software broadly. You know, it's the same thing we were seeing with with with Omachi, it's the same thing we're seeing with Swamp, it's the same thing we're seeing with a bunch of different. angles of software. And I think there's like a there's a whole new interesting sort of piece of software design that's sort of sort of in that zone. all right. let's Paul Stack: FX is very interesting as well because of the fact that you can you can actually build agent demos built into the the browser, as you said. But Adam Jacob: Yeah. Paul Stack: like you think like interactive tutorial websites, right? Where you actually have to show an agent harness in ha in effect, it would be such a cool little like embedded way of doing that. Adam Jacob: Yeah, for sure. and and that those use cases, I'll bring you straight back to wanting to have AI gateways, straight back to open router, straight back to, you know, all of those things start to start to make more sense when that's the when that's the world you're living in. okay, so Dynatrace bought Arise for $915 million in order to expand their reach and observability. I think, you know, beyond the like the marketing justification, like Observability, I think, is an interesting spot where how that's been built over time, how we how we have brought it into the market, the shape of the problems that it's solved, like the in the changing software landscape in AI causes this massive shift in what you need from observability capabilities. I think it's different now it when you're running agents than when you had teams doing it, they were trying to debug things. And so this feels this feels like This feels like a the kind of thing that companies of the size of Dynatrace like ought to be doing right about now, which is like finding their finding, finding somebody who's solved some of those problems and making acquisitions and trying to sort of pull that together into a cohesive story. I'm curious to see if that cohesive story actually can happen. You know, every time what's happening is we're sucking those things in and we're putting them on the core of whatever it was we used to do before and then saying that what we do now is a new thing. That's always risky business. James Governor: Hold so Adam Jacob: but James. Professional analyst. Hit it. James Governor: Well, I I think the thing here is is is is for me having spoken a lot over the years about sort of progressive delivery, where we are thinking about limiting the blast radius, you know, absolutely having a closing the loop on our pipelines, being able to do all of the things like canarying and everything else. And when I mean I read a book about this, and then it was like when the AI When the LLMs came in and started crushing, I was like, is this relevant anymore? You know, is i is sh is all that stuff jettisoned? And I think what's really interesting is actually some of the patterns about the way we use models there there's a similarity of pattern. I mean, if you're an observability company and you are not thinking about evals and understanding the way the agents behave and understanding the way models behave, understanding behaviors, you want to do something like customer experience behavior. And this is super weird to me. The world we're now in a world where, you know, like so Dynatrace is buying a company where literally they are measuring like the sense of humor of an agent and how it interacts with the customer. That's sort of science fiction. But if you're not doing that, then you're not doing observability. Because it as as you know, and I always go back to this, you know, as Charity Major said, you know, it doesn't matter if you have five lines if your users aren't happy. Well, some of that happiness might be based on Your interaction with an agent and how the agent treats you and what that experience feels like and how consistent it is. So I think there are whole set of disciplines. And are integrations between companies and acquisitions hard? A hundred percent. that said, you know, from Dynatrace, I think they've got a new thing called Blue Box, which is about like you know, taking the developer out and just giving all the data to the agent and let it do its work. And so seeing this, I just thought it was an interesting. Adam Jacob: Yeah. James Governor: by the way, Diamond Trace a client. So I don't want to just be an analyst. I full disclosure. So I've tried to understand this from that perspective. But yeah, there is no doubt that and yeah, I mean, maybe he was just giving me some nice confirm confirmation bias, but talking to Thomas Domkey, you know, last year, and I asked him, and he said, You you can't use models without progressive delivery. And it is that notion of understand the user experience, be able to do retries if it breaks, be able to do rollbacks. How we deploy models and agents, all of the disciplines around observability, I think, are relevant. Now, is it possible to take your existing platform and each it together and find some OTEL and glom all that together and actually create a great user experience? We'll find out. But I don't think that you can do observability without and I mean another one. Yeah. You know, any any observability platform sufficiently advanced is a cost management platform. Back to tokenomics, what's the biggest problem right now from any enterprise's perspective about on models? understanding the cost of tokens. So Adam Jacob: Totally. James Governor: from that alone, buying Nick: And ROI on those tokens. Yeah. James Governor: a fucking observability platform for agents makes sense. Nick: I'll just add in case it's valuable two things. One, blue box super interesting. Mapping your live topology is the holy grail of observability stories, of being able Adam Jacob: Yeah. Nick: to see all of the pieces of of the stack and understand how they all fit together. And adding AI into that, you know, makes complete sense back to James' point. The other thing that I just found interesting is I found out about this acquisition from the Arise people and the way that they were positioning it. And it was very unclear until you actually clicked in that it was Dynatrace that was buying Arise. Instead, from the a Arise perspective, it sounded like they were merging with Dynatrace. It's very subtle, Adam Jacob: Yeah. Nick: you know, when you see messaging like that. But it makes you wonder like, you know, how well are they actually going to integrate these companies at the end of the day? so Adam Jacob: Yeah. Yeah. Nick: Anyway. Adam Jacob: another fun story that popped up here was was some folks basically at SoundSlice added a feature because ChatGPT hallucinated the feature. So, you know, the story here was SoundSlice takes photos of sheet music and then puts it on so that you can sort of follow along as you go. Chat GPT was telling people that they could import ASCII tab. and then hear the audio to play back, but that feature doesn't exist. and so people were disappointed in the experience. And so now they've actually added I think they've added that feature in now. Is that right, James? James Governor: Yeah, it's an amazing story. that that that yeah, I mean, w you know, what do you do if people keep reporting a bug? you you try and respond to that. And even if it is them, and and literally Chat GPT was creating screenshots of an ASTI-based music notation that that SoundSlice, which is a music notation platform, did not support. You're like, well, you know, what what do we do? We can't yeah, you know, we can't go and tell You know, open AI, you need to change the behavior of your model. So they did. They decided that actually it made sense for them to build the feature that these bugs, these these bug reports that were not actually about the platform. so so yeah, I mean, this is I mean, this is definitely we, you know, we're not in Kansas anymore. You know, we are this is a This is a whole different world in terms of product management. And I thought it was a fascinating, fascinating story. Nick: Yeah, we built Adam Jacob: Yeah. Nick: into Swamp the ability for it to file bugs as well as feature requests on your behalf as you're using the software. And so right in the context of doing the work, if you encounter an issue or you wish it worked a a little bit of a different way, then you know, that's part of being part of the community is submitting that feedback to us and then watching us close the loop on it very quickly. Adam Jacob: Yeah. And I think what's to me, what's interesting about this is that, you know, you can loop it into the observability conversation because as a product person, you need to have visibility into the fact that the agent thought that feature existed or that a user thought that feature should exist because it tells you where you got where you might have the shape just a little bit wrong in terms of what people would expect. And what's what's weird is that now you can go from nobody would have ever done that for a completed feature like ASCII tabulature, you know, like that's That's so beyond like, I expected the button to be in a different spot and I missed the click or whatever. Or the, you know, the CLI naming convention was weird and I struggled with it. but but it seems of a piece to me, you know, that like like it's the kind of data you want. It's the kind of information that if you had it, you could start making better decisions about how users would experience the product. And and I bet taking ASCII tab in for guitar players is so useful. You know, like that's like such a common thing. So like what a good it's a good feature. James Governor: That might be the case. And also, you know, I mean, if you want to find something that wants to build that freaking feature, Codex is ready to go. Because Adam Jacob: Yeah. James Governor: like OpenAI is like, I'm telling you, I'm telling you, build the thing. Just ask me. I'm ready to go. You know, it's like, yeah, yeah, just I'm I'm I'm ready Adam Jacob: I already designed it. Yeah, just tell me to rip. Yeah, that's a weird that's a weird angle too. James Governor: to go. Adam Jacob: Yeah. Are the robots gonna start pitching us features in our own products that they could then build for us? James Governor: A hundred percent. Adam Jacob: okay. lots of the conversation is about individual productivity. Slack now talking about sort of agentic coding being a multiplayer and figuring out like, hey, we're gonna launch a product where what you're gonna do is use Slack to sort of Collaborate with your peers and do and like write all the code here in Slack. I want nothing to do with any of this. in terms of the user experience, in terms of how I feel like all of this could actually exist in my life, like, like no part of me is like, you know what I want to do? I want to like use Slack code or Claude Tag and have it just like spin up a new channel per task. Like there's a story here that thinks everybody wants to see all this shit. I I've long been of the opinion that basically all of the documentation and idea that people go back and review like why did we make that decision and all that stuff is fucking fiction. That like outside of a very small number of very important industries, no one ever looks. No one cares why you made that decision four years ago. No one cares what your chat conversation was with the agent to build that feature. Not really. And if they do, the best place to find it will not be querying Slack history. So yeah, I looked at this and was just like, look. And man, I know some people are gonna love that thing. You know they will. You know that there's people who are like, This is what I wanted. I mean I'm sure that it can, but it's just boy, it's not the surface I want. Nick: Yeah, two two things can be true at once, right? Adam Jacob: When we were learning how to work in this way and building our own software factories, one of the great struggles was that I just kept wanting to hone the architecture for everyone because that's been my job for so long. And so I was just annoying the crap out of everyone by just like surfing over everyone's shoulders all the time, being like, Ooh, how are you building that? I don't know if that's right, you know? And like this is just it it gives me it gives me flashbacks to that moment where like if I could have just like piggybacked on all your Slack channels or whatever, ooh, I'd have done that. You know? I'd have Nick: Ha ha ha. Adam Jacob: been there. And like, it would have been awful. Awful. Nick: On one hand, though, the times where y you would ask us to like steer Claude in a certain way by typing something into our own Claude or whatever, like Adam Jacob: That's what I'm talking about. I could just I can feel it. Like people are gonna get lost here in this like terrible pit of horrific, just horrific behavior, you know? Nick: Okay, but let let's just reframe this as the need that I think is that we would all agree with, right? Which is right now adoption of AI among teams is completely ununiform. You have some people, it's very unevenly distributed. You have people who are like masters and building l l loops and graphs and really at the frontier, and then people who are still using LLMs as like a query engine or whatever, or glorified auto-complete. And You know, what this points to is that like having a centralized way to think about your business, where the work that I do compounds for the work that you do, as opposed to having to reinvent our own universes and build up our own memory files and our own claude.mds or whatever, you know, or agents.md. Like I think it points to that. Like, how do you how do you deliver outcomes through these things as opposed to like teaching everyone how to be AI maximalists? And so, you know. I think that that's the real the real thing that people are trying to get at here. But I think that, you know, seeing your cloud code session is not the right way to do that. Adam Jacob: James? James Governor: Well, I mean I've enjoyed you all sounding so much like old man shouts at cloud or shouts at AI. This is this is a a charming and good moment, on this podcast. I mean, I I think here's the thing, in the way that the throwaway, like, that's just for marketing pages. I mean, I would I would love to have Vercell's business, right? Like, you know, we're like, turns out marketing pages, you know, and shitty little websites, whatever, that That that you know, that's that that stuff is gold. And Nick: Absolutely. Adam Jacob: Totally. James Governor: you know, and and and I so I I I see that and I'm like and also a lot of what we've been doing has not been a team sport. A lot of the agent revolution and the AI revolution has been about individual productivity. so collaborating on things i i is is kind of valuable. Now, do we want to give four years? of history? Absolutely not. I was talking to an organization recently and they had been like, yeah, if we give if we give this model like 15 years of our Microsoft Teams history, it's gonna be good. That did not happen. That was fucked up. Recency recency is highly valuable in this building environment. But I I just think that that yeah like we're gonna have a broader range of people building things they are gonna be collaborating they're certainly gonna be collaborating in Slack. and yeah, we you know, we will have them working with agents that are, you know, from Anthropic or OpenAI or others. I think that this seems like the direction of travel. and and so is it you know, and we can be like, wait, that's gross. Why would you wanna why would you wanna think about it that way? But We've already accepted that not everything has human code review. That ship has sailed. That's done. That's a 2026 statement. And once we've done that, the question about how we do product management. I mean, Bill Joy from Sun Microsystems, and this is again aging myself, had this phrase which was innovation happens elsewhere. And I think now software development happens elsewhere. And you know, we we we've had you know, history of software development happens in the IDE or it happens in the editor. I'm a terminal person. Software development is in these environments. Software development is now going into other places. So you know it's it's not just Slack. If you look at where Atlassian is, like, if I mean we're all like I obviously software developers hate Jira or whatever, but everybody uses Jira. It is the foundation for a lot of product management. And product we all know product managers, software engineers, and it line of business employees, they're all munging up in weird kinds of ways. So I mean whether or not it's I'm s I think this sounds gross. I love it. It's I've got the my AI has zoomed in on me. I Adam Jacob: Yeah, it came to you. Yeah. James Governor: I'm gonna go back there. That was good. So whether Adam Jacob: That was good, yeah. James Governor: or not whether or not we think this is the best way or something that we like, it's gonna happen. And if you're Salesforce, I mean, yeah, I don't know. So, Christ, another they're a client as well. I I have plenty of vendor clients. But no, when I looked at that, it was just this it it's not so much the team thing that I thought was interesting. I was always a no a low code skeptic. I always thought that was bullshit, but we're not in that environment anymore. It turns out that you know I I can just go build me a little app to do this and I can do that. And part of that Adam Jacob: Yes. Yes. James Governor: will be a conversation with my colleagues, and that conversation is likely to happen on Slack. And as I've already said, I mean begin, which is a little bot on Slack, that's not complicated, but it's making my company more productive. Yeah, and think you know, we see this. Adam Jacob: Yeah. And I think, you know, we see this now. Like our our co-founder Mir, like he works in Claude on our business all of the time, right? We like have an application that is our business model that like pulls in all the data and does all that stuff. And, you know, that part of why that's working for him is that we built the software factory that he's using to build those complex pieces of software. I think, you know, it's totally possible that those that those sorts of interfaces get vended back to people. you know, wherever they want to find them. I think that makes perfect sense. When I think about this from a software engineering point of view, as the person who has to build the factory or who's going to build a complex piece of software, there's definitely going to be a manager who's like, we do all that and stuff. All that stuff happens in Slack now for visibility. And it's going to just like, that just makes me want to sink below the ground. You know, I just want to dig a hole and stay there and and pretend that that's not my life. Okay. So let's talk about software factories. so I gave a talk last week in Seattle on building software factories. I walked through how to build them, why they work. We've got more and more folks every day sort of doing this in the swamp discord. The conversations about it are really interesting because this is a moment where lots of people are figuring out how to do this effectively. I think right now we're we're in the era where a lot of folks who are trying to do this are trying and failing. Some of us are succeeding. I think those who are succeeding are having a really good time. Those who are failing obviously are not. I think one of the things that has been coming back to me since we gave that talk was sort of how critical the approach we take to black box testing really is, that we have like the full product essentially instrumented for acceptance testing so that it makes it safe for us to run a factory to, you know, refactor that code base to add new features to like make sure that that stuff stage stays working over time and and that we can move at the volume that we do. and so I think that that change in acceptance testing w I think is a really important piece of it where now that you can write software at this rate of speed, now it also enables us to rethink how we would have tested it, right? Because to James's conversation on progressive delivery, you would never have built an acceptance testing suite as complete as the one that we use for Swamp in a world where you had humans writing the code because they would have killed you. You know, they would have absolutely rioted. There would have been, it would have been madness if you had put that in as a requirement for every feature to ship, you know? But when the robots are doing it, this is no big deal. And so suddenly your ability to have that comprehensive test suite really goes up and our ability to do that delivery really, really alters. Paul, like as the as the as the person of it, speak on the topic of of of UAT. James Governor: Hmm. Adam Jacob: Yeah. I think James, you are a software factory skeptic. I think was the was the phrase you used for me the other day. James Governor: yeah, I mean, software factory is not a new idea. we've had like multiple generations of of this. It began in like the 70s. some of the Japanese services companies got very excited about service software factories. The DOD loves software factories. we've we've we've had multiple eras this, and I think anytime like it's trivial. To like build new shit. It needs to be a factory. so we we we see this again again. There was a time, my god, Microsoft got very excited about software factories for a while. I think my skepticism or my initially I was like, LLMs and AIs, that's just automation. And increasingly I'm of the opinion that I that was blase of me and wrong. And that we do need to distinguish between deterministic automation and LLM-based automation. And the thing about factories is not that everything is done, you know, you know, like intent creation. The thing about factories is that we're able to understand the costs of things, we're able to have control of things, manage things, and and factories are by definite definition deterministic. And so what I think is interesting is at the moment you've got like factory AI, you know, everyone's like, it's a software factory. But we we're we're all as an industry doing really stupid shit with LLMs. You know, and if we're business consultants, you know, I think it was Ernst and Young, they were like, What why is wait, what? Our token budget? And it turned out that people were using LLMs to parse PDFs. And it's like, that's not a good use of an LLM, right? Like, and in software, we're doing a lot of the same things. We're like, I mean. We'll do math. You know, hum you know, you know, get the get treat the LLM like like a fucking relational database. And it's like, it is not that thing. And something is and by the way, I've sworn at least three times on this podcast, and no one else has. And I I'm feeling very Adam Jacob: It's fine. Fuck yeah, we have. That's not true at all. And and James Governor: okay, well Adam Jacob: I would say, like, look, I would I'll you know, regardless of swearing, I'll just I'll give you a like, fuck yeah, James, because look, that like you're we are not a client of Red Monk, but like that is like we agree with you. Like that pitch that says basically, look, there's there's what LLMs can do, and that's amazing. And then there's what deterministic automation can do for you, and they're two different things, and what you need is both. And and James Governor: You need both. The software factory go keep going, sorry. Adam Jacob: And so, like for us, when we talk about software factories, like the vast majority of that work is actually transitioning work out of LLMs. It's using less LLMs. James Governor: Out of L L Yeah. Take the shit the that we absolutely. Adam Jacob: and and that transition, you're using LLMs in the places where what I need is their intelligence, where what I need is their creativity, which is a shockingly small piece of the puzzle. You know, I don't need an I don't need the LLM reinventing from first principles every time how to run my tests. I know how to run I don't need it parsing whether or not they succeed or fail. I know how to get them to succeed or fail. I know I James Governor: Is he? Adam Jacob: can measure it. I know what it is. It's easy enough to do. And like, so, you know, that software factory process, that's why it's working so well. You know, that's why Paul is shipping like 30 times a day on his own. That's why we're not having regressions. That's why we can build these crazy features because what's happening is we're you we're not using the LLM as like the world's most expensive for loop. And James Governor: I by the way, I am really enjoying this podcast. Adam Jacob apparently knows something about automation. That was that was yeah, that was quite a good Yeah, what like a weird But but I Adam Jacob: Yeah, that's true. That's my secret. Yeah, that's my secret thing. James Governor: do I worry about the industry. We're so carried away that we want to use LLMs for everything and it and some of the software factory rubric, it it's not about let as you say, what we want to do is is take what we've learned and built with LLMs and then drive that into repeatable deterministic automation and Adam Jacob: Yes. James Governor: that's where value is gonna lie. That's where tokenonomics really token token token economics should be about like using fewer tokens. And that's about automation is cheap, right? Like traditional automation is cheap. We should do as much of that as we possibly can. And I think that's where the software factory value really really can come into being. And I just so much Okay, there we go. Adam Jacob: Well, look, your lips to God's ears, because that's SwampClub.com. You heard it from James. Okay, so I think the other piece I will say for for on factories, John Watson, who works for us, had helped one of his one of his good friends who is building an application, set up a software factory with Swamp. And they have all the data about that factory and all the things that it's done over the over a couple of months. and he went and did a bunch of interesting analysis on how the factory was performing. And and some of the stuff they found was like 65% of the issues. So everything they push through the factory they call an issue are merge ready after one round of review. So basically 65% of what goes in basically gets gets it right roughly on the first crack, right? because if it needed more than one, it would be two rounds. 29% total of all the things that merge. have some warnings attached to them. So structurally, like, you know, if you have an agent that's like reviewing it for accessibility or whatever and it found an issue, but it's not like a huge issue, it's just like a nice to have, then it'll flag that as a warning instead of a critical issue. which means that trying to think about tuning your LLMs for exact precision, like it should be a hundred percent right every time, is probably a mistake. Because the other thing they found, and James Governor: Mm-hmm. Adam Jacob: we see this in our own factories. But it was interesting to have it sort of confirmed by someone else is that after round three or four, you're basically just oscillating between arbitrary requirements. So James Governor: Yeah. Adam Jacob: like it could go this way, it could go that way, rather than actually converging on something good. And you're just basically burning tokens for nothing. and over the course of like the 372 issues they had worked in this data set, there were 1,463 code review rounds. And only 338 plan rounds. So the vast majority of the time, what was happening was if it was going to get something wrong, it was going to get it wrong in code review, not in planning. So they were much better at sort of making a plan that adheres to all the criteria than they were making code that adheres to its own criteria over time, which also makes sense and sort of that we have seen sort of in that process. Keeb, you know, when you think about that being getting to a spot where we can start to put some numbers behind. how these factories are actually working and how to build them. you know, do you think that this shape of of how we think about factories and how we think about measuring them will carry us further? Do you know what I mean? Like if we start having this conversation more broadly, like what's the impact of being able to actually know the numbers? Nick: What's what's wonderful is that, you know, the community that exists inside of Swamp Club is is trying to answer this question for themselves and are, you know, like are cross-pollinating the ideas on how to do this and build this effectively. And so we're seeing this happen in real time as people are taking their existing SDLC and trying to incorporate it, incorporate it in the age of AI. And what they're saying is like, There's a massive fold increase in their ability to to to create and release software. And now they're thinking about how do I how do I do this for feature requests in addition to bugs? You know, because bugs are one sort of flow through it and feature requests are the other. And then yeah, how do you get how do you how do you measure the productivity increase and the impact on the business? I think is yeah, we're kind of just like all in the turmoil of of figuring that all out together as an industry right now. Adam Jacob: Yeah. And I I I think there's a moment where, you know, James coming in and having said he was a skeptic, I'm sort of straw manning him. But like I think it's an interesting, it's an interesting thing because the most common response I have gotten when we talk about how Swamp builds software is no, you don't. Like literally the most common response I've Nick: Ha ha ha Adam Jacob: gotten is it doesn't actually work like that, or you're lying, or you don't, or no one uses Swamp, or none of these things are true, or you don't have real customers, or you know, like those are the those are that tends to be the conversation about what happens. And I think some of that is because you can't it we're only just now getting to the spot where I can show you numbers, where I can say, here's what's actually been happening and here's how it's actually performing. And I do have that observability and I can see how that those things come together. And so I Nick: Yeah. Adam Jacob: think part of part of that question of like how do we build efficient software factories? How do we actually help people solve this problem? I think it I think it's it's incumbent on everyone who's doing it to actually move to the next phase where it's like, okay. Nick: Well that Adam Jacob: Let's actually talk about how we're doing it. Let's talk about what those numbers are. James, are you seeing other people talk about those numbers in that way yet? Or is it still a lot of like software factories? They'll change the world. You know, sign up for linear. James Governor: I would just say in general, that what I have found interesting is yeah, I think we're behind the game on measurement. But if we look at adoption patterns, so for example, I'm here in London, w you know, and and there was you know, it it it took European companies quite a while to get on like the cloud native sort of bandwagon or whatever. What I found very interesting is that European companies are actually so far ahead or so close Adam Jacob: Yes. James Governor: And and enterprises, you know, talking to banks, financial services companies, telcos, manufacturing companies, and they are fully on board with I don't I'm not hearing much no you don't. I'm hearing a lot of yeah, okay, give give us more. How do we how more details? How do we have Adam Jacob: Give us more details. Yeah. James Governor: Well you know, we're in the the context and governance stage right now that will make us even more productive. I'm not I'm I'm not seeing the usual levels of pushback that I I've seen in almost any wave of technology in my career. So you you need to see Adam Jacob: And I think I'm starting to see it l ebb. I think maybe I spend too much time on Twitter and LinkedIn. Nick: Absolutely. Yeah. James Governor: I mean, you know, maybe I don't I don't know, maybe it's maybe Europe is maybe maybe you need to come and get some European customers out of Adam Jacob: My experience is absolutely that Europe is actually kind of ahead, in in very strange ways, for historical purposes, but Nick: Yeah. Just the shit posters are just moving the goalposts, you know, that's Adam Jacob: Totally. Nick: what I'm saying. It's like, yeah, okay. So first no, everyone's reading the source code still, or else you're like delivering slop or whatever. And then it's like, it's only for small features that you can do this. It's not for real products features or bug fixes or whatever or refactors. And now it's Okay, no one no one has delivered any value at all. And if they have, it's a immature product in a greenfield and you know, there's no meaningful users and it hasn't been live for six months on a team or whatever. You know, like the it keeps shifting Adam Jacob: Yeah. That's what's happening. Nick: and it keeps getting more refined and it's so stupid. It's like anyway, yeah. Adam Jacob: Yeah, I think that's exactly what's happening, Keep. All right. Well, that's the news for this week. James, it's time for five questions. You ready? James Governor: God, I've that means I have to remember what we've spoken about in the last hour. I you know. go go ahead. Yeah. Adam Jacob: Of the s you're gonna get there. All right. So of the stories we just covered, which one are you gonna be thinking about later today? James Governor: What am I gonna be thinking about later today? I reckon that of the conversation we've had, you know, I I think it's just this conversation that we're having. I I mean, it's not the news conversation, but I'm always about like where are we in terms of company readiness to change. Adam Jacob: Yeah. James Governor: And yeah, I it this is something I'm just spending a huge amount of time on. you know, thinking about in my research, in my work with clients, it's just as say it's that fundamental thing about like what you just said. Like wait, what? Like just humans are out of that loop in some regards. They that that our role is not to read every piece of software that the robot writes. And that is a a fundamental and interesting change. And that that's one that my brain keeps Worrying around thinking about Adam Jacob: For sure. All right, which one of these stories is Okay. what's a common belief in our industry that you just think is wrong? James Governor: A common belief in our industry that I just think is wrong. Honestly, I can't. Every day is reassess your priors, but throw out your priors day. I think common beliefs, I mean, if if if an a common belief that is just wrong is people in our industry that think that that LLMs don't work. That I'm like, Adam Jacob: Totally. James Governor: can we just can we just stop because it is Adam Jacob: Yeah. James Governor: so f annoying? Like, on a daily basis there are good, useful things you can do with models and if you're like none of it works, I mean there are other criticisms that are valid, but Adam Jacob: Yes. Totally. James Governor: the idea that they don't work, that is just stupid. Adam Jacob: I agreed. What's the one thing you wish you could just teach everyone? James Governor: Be kind to each other just be kind to each other. Adam Jacob: what's the best and worst moment you've had working with AI agents? James Governor: God, I should have read the fifth question and in preparation. I mean, some of the early, some of the early experiences were pretty bad. you know. yeah, slop, and and there was a lot of it. so working with I You know, I I think it's I think it's when my daughter caught me, you know, using using a model to generate an image, and she was so mad at me. she was really she was like, Dad, I've told you about that. If you want an image, come and give me a spec and I will make it for you. You are not you are not allowed to like take the jobs from artists. So that was Adam Jacob: Yeah. James Governor: that was bad. So that was my worst moment and and maybe that was and also, yeah, slop, slop, so much slop. Adam Jacob: So much slap. James Governor: And it's interesting and it here's the thing. You can use models to to to answer questions, to generate code, that's all true, but my goodness, the the the the look and feel, the gross look and feel of so many of the images that are generated. The the recent fuss about like, you know, like Yeah, if you you know, people using them to create flyers for events and stuff. Th that it gives me the ick. I I some worst moments is seeing people that don't have good taste using models and and using and just t shoving slop down my throat. And worst I mean, that's I may have done the same thing twice. Worst moment getting caught by my daughter, she was pissed. Adam Jacob: Yeah, makes sense. All right. 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 summaries over at Swampdash Club.com/slash podcast. You can sign up to get it delivered straight into your inbox every Monday morning. If you're trying to figure out how to manage production infrastructure and applications or scale the use of AI within your organization, Swamp is here to help. Visit swampdash club.com to get started using AI agents safely and effectively. By building repeatable systems you can trust and verify, like, say, software factories. I'm Adam Jacob. Keeb. Nick: neck. James Governor: I'm James Governor from Red Monk. Adam Jacob: And James, as our special guest, you get to take us out. What do you want to plug today? James Governor: I don't know. I I I us. Yeah, my company is amazing. you know, I talk a lot, but I have very smart colleagues. And if you're interested in software development and the implic implication of software development in terms of agents, we we are a great research firm that will help you understand the decisions people are making, why they're making them and increasing I guess the decisions agents are making. So yeah, come and work with RedMunk. Adam Jacob: You should absolutely come and work with Red Monk. They are, in fact, the best in the industry at that work. thank you very much. We appreciate for listening. We'll see you all next Monday. Bye. you