Well, I had the chance to be in London last week or the week before. I don't really remember for DevOps UK. I haven't been to a DevOps conference in a long time. I guess it was two weeks ago because they just posted the recording of the talk that I gave and they noted that it was given two weeks ago. So I'm going to assume I'm going to go with what they said there instead of researching myself. And well, first of all, you know, the DevOps people are nice. There's at least two types of conferences that I go to speak at. I don't know about you, Ben, but the first type is what you could call there's some real professionals and they get very excited about, you know, checking on you, making sure you show up. They always want to have lots of pre-arrangement stuff. They might want you to send your slides ahead of time. They seem very professional and that can be annoying, I guess. And then there's the people more like DevOps where they're just cool. They're just like, hey, that's great. Come talk here. Here's some emails to remind you that it happened. And then you kind of just walk in there and there's the AV person and, you know, you got to get your microphone. And then maybe someone from the conference might come up and be like, hey, how's it going? Everything good? And then you're like, yeah, it's great. And they're like, cool. Enjoy your talk. And they're super nice. They're chilled out. They know who the talent is. They know you've done it before. That's what it is. Like, oh, Coté, relax. He's got this covered. I was going to say they even had bottles of water up there, right? So they know what's going on. That's the sign of a good. And you don't want that that that sparkling water with gas water, because I don't know about you, but that is not refreshing when you want to clear your throat. You want you want some you want some liquid watery water, not not this manmade stuff.
2:01And what kind of sadist leaves that underneath the podium on the stand? Yeah, exactly. Just waiting to blow up. But I had a different question before we get on to conference food, which I'm sure you also partake partook in. How did you find the cycling experience difference between Amsterdam and London? Because I think London is more like Mad Max and the Thunderdome. Yeah. Whereas like Amsterdam is so chilled out and so relaxed. It's it's a this is this is this is a interesting, astute question here, Ben. Now, I'm going to give you a trick answer, which is I don't really when I go to London, I don't really cycle in London because I haven't won. I've done it once or twice, but I haven't really taken the time to research how those whole city bikes things work. I think you have to, like, return them to it's San San Santadar. How do you say that? Santander. Santander sponsors it now. So they're all over the place. I forget what it was back when I first moved over here. But I think you've got to go from one Santander docking station to the next. Just totally fine. Right. But but yeah, like when I look around, it looks like I mean, Mad Max is a pretty good description of it. Like I look around and I'm like, I think if I biked here, I'd get killed. Right. Yeah. Or even even just being a pedestrian. Oh, yes. Not not to mention the fact. Not to mention the fact that, like, you know, I am of the part of the world who drives on different sides of the road. So, like, I just I would get very confused about what was happening and probably, you know, I'd get hit by someone just because thankfully the parts of London that I'm always in, you know, they put on the street. Look left or look right as as is appropriate, which I find very handy. But I always feel bad because I have this habit that I learned at a long long age at a young age.
4:02And it's reinforced when you're cycling. Right. Because when you're cycling, even in a nice place like Amsterdam, I always have in my head. My goal is to not get killed and not kill anyone else. So, like, you just got to have complete situational awareness. And so I'm always looking all ways when I'm crossing the street. And I just know there's some localers who are like, stupid foreigner. You don't need to look that way. But, you know. Yeah, I think that's true. It's very noble of you not to want to damage anyone. I think. Thank you. That's very good. And I feel like the cycle lanes in London, it's a bit like the wildebeest scene in The Lion King. Trying to cross one of those things. Right. There's a 50-50 chance you won't come out of it alive. Yeah. I mean, I think. Do you remember the, like, Manhattan messenger, like, sort of, like, idea that existed in the 90s? Where, like, you know, in Manhattan there would be these, like, bike people on messengers with messenger bags, which since became very popular. And, you know, they would just be sort of, like, grungy, like, cyclers who would go around delivering pieces of paper between the firms. And, you know, I'm sure there was at least one movie made where they were some adventuring swashbucklers, you know, and some romance involved. But I feel like that might be what it would be like to bike in London. You would just be constantly, like, going between cars, hang on to some of them, like that guy in Back to the Future, skateboarding around. You know, it would just, every now and then you'd just, you'd, like, skid over the hood or, as you would say, bonnet of some car and be like, hey, I'm trying to drive here. But, I don't know. Yeah, I think you're right. I think there's a scene a little bit like that in Hackers as well. And that's from the 90s, isn't it? I reckon it might be in there. But, yeah, you know those folks, they have the fixies, they call them, don't they? Yes. No brakes. It's just, like, everything through the pedals. Yeah. You're just living the dream.
6:03Yeah. But I think what I need to do is learn how those rent-a-bikes work. And I just need to bite the bullet or bite the car and start biking around there. You'll probably bite the pavement. That's right. Yeah. But it would be nice. Because there are, like, you know, to be fair, right? Like, I walk around a lot when I'm in London. And it's actually more or less a lot calmer than certain intersections and areas are. So biking would be totally fine. It wouldn't be a huge deal. But, you know, also on that note, I picked up two bags of weird crisps, as you would say. How do you say crisps or crisps? How do you say that? What's the plural? Crisps. Crisps. You got to really just, like, crisps. See, I can't do it. I'm going to have to practice. A bag of crisps. Oh, okay. So there are kind of two S's there at the end. Anyways, I got one that was the – I'll put some links to the reviews that my daughters and I did. But we did one that was the – basically, it tastes like charcoal grilled steak. It was pretty amazing. Like, it was very genuine. And then we got another one that was the Herefordshire sausage and mustard flavored chips. For everyone else in the UK, that's Herefordshire. There you go. Yeah. Thank you. I only know how to speak English from the Lord of the Rings. Oh, you murdered that one. Murdered it. Tell me what's up with that part of the country. What do they do? A lot of sausage and mustard? Yeah. I mean, there's lots and lots of areas that are, you know, very big agricultural zones in the UK. Herefordshire would be one of them, I suppose. Herefordshire. There's lots of them, yeah. Okay. Huh. Well, now, if I were to tell you you were going to have Herefordshire sausage and mustard chips,
8:06would you be like, oh, I know that flavor. I know what that's going to taste like. Okay. Yeah, absolutely. Yeah. Yeah, definitely. For sure. I mean, I don't know if we do – do US, UK sort of do sausage in the same way? In the US, I think of a hot dog, right? Yeah. And in the UK, it's kind of artisanal almost. You know, there's all sorts you can get in sausages. Yeah, yeah. You can get all kinds of weird and wonderful things. Yeah. Well, you know, for our frequent sausage corner section of the podcast, I think – I would say in the States, there's only three types of sausages, one of which we don't consider a sausage. You've pointed it out already, the hot dog. I think if you were to do a survey, a little person on the street thing, no one would say a hot dog was a sausage, its own thing, but as we know, it is a sausage. And then you have – what would you call it? You have like – what is the equivalent over here? They have – I forget. I don't know what they call it here in the Netherlands, but it's this sausage that you can either buy off the shelf or in the cool section that's basically just like ground up stuff in a sausage that's maybe about an inch thick. And it usually comes in like a deep red tube and like, you know, you slice it up and it's like not that great, but it's kind of like a typical sausage you would have. And then you have like barbecue sausages and those are just their own things. And I guess, you know, there's Italian sausages as well. So you've got that. And I'm sure the nation of Italy is like, yes, which of the 50 different types is it? Can you be more specific? Yeah, but an Italian sausage basically just means usually like a spicy sausage made with pork is what you've got going on there. So I think if we've got any listeners left, if they can tell us, first of all, are there
10:08any sausages that are kind of protected, right? You get European protections, don't you, for different things, right? That's right. The Cornish pasty would be an example of that from the UK, right? It's protected. Are there any protected sausages? I'd love to know. And also, as a sort of plant-based diet person, if you've got any top tips on, you know, plant-based sausages, things I might not have tried yet, let me know. I'd love to hear them. So yeah, for our one listener who's left. Well, and that's sausage corner. Well, that's it. That's it for this sausage episode. Until next time. Well, I thought it was one of our main topics. We'll see if we have time for a second one. But I came across a survey this morning. It was from, you know, the US that said it was something like, you know, 50 or 60% of people have some interests or thoughts, know about chat GPT and AI and that kind of stuff. But only 14% of them had actually used it. And that kind of, like, added some numbers to this intuition I've been built. Intuition. This gut, this thought I've been building up over the past maybe 90 days or so. But that is, like, there's not actually that many people who have used this thing. And so, sort of, like, that's why it's easy for there to be all this, like, both hype and paranoia around it about, you know, it's going to take over the world or, like, threaten things or that it just, like, you know, equally that it does all these magical things. Like, I was at a dinner with a bunch of some executive types last week. That's why I was in London last week. And, of course, we had a conversation about AI stuff. And I think, you know, there was actually out of six of us, like, half of us had actually used it directly and used it frequently.
12:08But then, you know, the conversation we had with the other three people is they were just, it was just sort of like, oh, this thing is, like, potentially scary. And it could do all sorts of weird things and what's going to go on with it. And so, when I saw that 14% of people had actually used it, like, it kind of clarified in my head. I use it all the time, almost every day. And I think once you get around, at least you wouldn't using ChatGPT and even things like MidJourney, like, you use it after a while and you realize, like, oh, this thing is not a big deal or not as big of a deal and it's not magic. Right. Like, it's really just a tool that used properly is magical, but it's not like some gigantic big, I don't know, thing. Now, the thing with computers is every year they get better. So, maybe it would get better. But so far, and I think I'm copying this from some commentator, but my reflection on it has been that, like, there is, and I know there's people who very strictly like to say, instead of saying AI, they'll say LLMs or something generation, which I guess is technically accurate. But in ruining the joke that I'm about to make here is, like, whenever you use an AI, you realize that there's pretty much no intelligence to it. Like, there's no, like, like, I try to get it to write lots of text for me and it just, like, it kind of does it, but you can see that it's not really doing a great job. So, you have to go in and, like, help it out a lot. Right? It's great. It's great as, like, an entry-level person who would do a lot of the drudge work and the toil for you, but you really got to go in there and human it up, so to speak. Now, that said, I've used it to generate, like, Python scripts and stuff, and that, for my use, is pretty awesome. Like, it just, like, works and does stuff. However, when I was discussing that at that dinner I was at, someone did raise up the point that, like, yeah, but, like, can you have it, like, go in and, like, refactor things for
14:13you and evolve it and add new features? And, like, where are we as far as, like, you know, you could generate those initial scripts, but then, like, two years later, what's the deal with that? And, you know, I guess I haven't really tested out. Go into the script and modify it so that now you can work with Apple Pay. That would be fun. I should have the scripts and say, like, make this compatible with key passes or whatever. Maybe it would spit out, like, five pages of things I need to install, which would be cool. But I thought I would, with that opening monologue there, I know you've been looking at it from stuff that you putter around with. What have you been doing with the AI stuff? I've got loads of responses to what you've said, really. So, yeah. Let's make this interactive, as they say. I have been playing with it. Yeah, definitely. And, first of all, you're absolutely right. You know, the more that you play with it, you do start to notice the rough edges and the things that it can and can't do. And one way that you can definitely see that is what I've been doing recently is learning how to run it locally. So, just running it on regular desktop hardware, no sort of fancy graphics cards, all that sort of stuff, just laptop sort of hardware. And when you run it on that sort of hardware and you start to then be constrained by things like processing speed, memory, that sort of stuff, you have to use smaller, large language models, right? So, the models shrink in size so that you can fit them in the environment that you've got. So, a regular laptop, maybe a 16-gig laptop or whatever, it might be able to take maybe a 7-gig model, something like that. So, it's going to load it all into memory and then use that for its parsing. And then when you then chat with that model, its responses are much more dulled than they are when you're using something like ChatGPT online. So, you really, really notice then that actually, yeah, I can see that this is a system that
16:17is mimicking behavior. It doesn't actually have behavior. It just sort of mimics it, right? And the other thing I've sort of learned recently, I suppose, from reading around and listening to others in the space and what they have to say about it is that, you know, there are various sort of levels of dumbness that you can experience and there are ways that you can tune these models just through the setup of the model to sort of make it behave in radically different ways. A good example is setting the prompt to be, okay, whenever you respond to me, respond as a pirate and it will change all of its responses. So, it's like, argh lad, I think you should try this. It sounds very much like it's from the West Country, which is where I sort of live. So, that's kind of fascinating that you can, you know, you can sort of change these behaviors and the behavior itself is kind of dependent on the back-end model. And the other thing you said there was about coding, you know, wouldn't it be nice if you could get it to elaborate on your code for you? And there are several models that have sprung up that are tuned to do just that. So, an example would be the star coder model, I think. I hope I'm remembering that correctly. Folks, get in touch if I'm making a mess of this, you know, set us straight. But a star coder model, one of the things that that can do is take a piece of code that you've already written in the prompt and in that piece of code, you can sort of add a comment to say, like, fill out this section because it's not, you know, I want more. Like you said, maybe make it work with Apple Pay or make it do something that it's not currently doing. And it will go away and have a look at that and come back with code that fills out the middle part of that particular piece of code that you've got. So, the learning from that is that there's not just ChatGPT's GPT-4 model.
18:25There are lots of models, hundreds of models, thousands of models. And they're all fine-tuned to work against perhaps subtly different use cases or perhaps created in the first place with different data, right? So, you can get ones that are open source that are based on completely open source data, ones that are closed source that are based on data sets and data that wasn't, you know, made available publicly and you can't test, for example, you know, what's in there. Some that are clean, some that are specific to coding problems or customer service problems or whatever it might be. There are literally hundreds of them. So, it's a really fascinating space to get into because you then start to really realize, yeah, actually, this is a tool and it's not necessarily something that you should be afraid of. It's just, in some cases, a time-saving measure. Right. And in other cases, it needs a lot of work before it be even remotely useful. So, you know, I'm not worried about it. I think it's just really exciting, the sort of things that it can do. Well, I think, you know, I haven't done stuff locally or changed different models, but I think that would be the next level, at least for me, of interestingness, right? And I mean, well, so first of all, like, how do you, like, is it difficult to make a model or do you just give it a bunch of stuff and it just figures it out? I haven't tried making my own model, right? But there are, the first thing that you need is, I think, tons and tons of data, right? So, you do need a large data set from somewhere. And usually that is, an example would be, there's a data set called the stack or the big stack, I think. And it's basically based on stack overflow sort of data. So, you imagine all the information that's going into stack overflow over years.
20:28You know, pulling that out and cleaning that so that it's removed all the sort of information that shouldn't be there, like personally identifiable information for GDPR or, you know, just removing all the sort of stuff that you don't need. That's the first sort of exercise. And I can imagine that's quite time consuming, I think. Yeah. But then you take all that stuff and run it through these chat model makers, you know, things. I think an example might be something like TensorFlow. Like I said, I've not really looked into it very much. The creation of the model is something that I've not dabbled with. But, you know, it can go through those steps to create you finally a sort of binary file that can then be parsed by these model parsers. But then it's not over. You can then also fine tune that again using things like language chains, where you could maybe take a generative model that's good at talking, right? It can actually create responses that look realistic and maybe add to that some of your own proprietary information, maybe technical documentation for a product, for example. It's one that comes up quite regularly. And then the model will then be able to answer questions about that product. Another example is Kubernetes. There's one called K8GPT, which you can run. It relies on a general model. So something like a GPT-4 relies on that general model, but then understands Kubernetes. And it acts like a bit like a Kubernetes operator would act. So, for example, you can give it your cluster. And if it spots that one of your apps is misbehaving in that cluster, it can start to recommend remedies and troubleshoot that problem so that you can make that problem go away. So there's all sorts of specialisms and ways that you can modify these general purpose models afterwards to make them more fine-tuned and useful.
22:34Yeah. Like, it seems like, I mean, I'm sure, well, I would hope people are already doing this, but it seems like, you know, in the spring community, they would, like, make their own, like, model for doing whatever, right? Like, you know, if anything, you know, you could have a first line of, like, reviewing contributions or, you know, just asking questions about how to use things or looking into the code base or just generating code. And because there's a lot of content about, there's documentation, there's all the different versions, there's all the source code for Spring. And then there's also stuff that uses it and conversation around it. So you've got, like, a huge body of knowledge you could start using to have a little Spring programmer at your side. I think so, yeah. I think, I think, I don't think I would say any of the models I've seen so far, any of this stuff I've experimented with so far has given me any confidence that a model will be able to make a decision about something. Like, which is the best way to go? I don't, I haven't seen much that does that very well. But I know there are lots of startups and lots of folks working on things like flow processing, you know, where, okay, I've got this administrative function that I need to do, like, I don't know, booking a vacation or whatever it might be. You know, within these parameters, here's sort of what I want, go away and sort it out. And it will then split that into multiple tasks that can then be broken down and then tackled by the AI and then sort of, you know, start to make suggestions or whatever. And I guess it's all based on weighting, you know, there's a probability weighting that this is probably what you want, so I'll do this. But they're not great at decision making, I wouldn't have thought. I haven't seen much of that anyway. Yeah, yeah, I went one day I need to make some, some, I don't know, local way of running it and just like feed it everything I've ever written, because I always want to be like, just write it like I wrote it.
24:44Like, like, like, I've even tried some experiments to like give it my writing and say like, how would you describe this writing, because then you can like feed it back to it at any point. And I would assume also, if you run things locally, you can like, give it a lot more information. Like, that's the problem I always have with wanting to use it for writing nowadays is you can, you reach a maximum character or token or whatever, like you reach a maximum amount of text that you can give it. And then you can also see if you're like working iteratively on a chunk of text, you can almost detect when it starts to forget stuff from the beginning. Like, it's just kind of throwing off text that it can no longer fit into whatever it's, it's buffer. I don't know what people call it. Yeah, context. Yeah, they call it a context. The context, you know, kind of limited in size right at the present time. But again, I saw something just last week that was about a company that's figured out a much more efficient way to do context so that they can grow to larger in size. So I think the average context at the moment is somewhere between sort of one and 4,000 characters or something like that. So the context is kind of limited. You know, there's only so much you can do. This new one could easily handle 70,000 characters. So characters, was it chunks? I can't remember how they described it. But basically, it was enough to be able to take in an entire text of a book and then give you a pricey back about what was in it, you know, and tell you, answer questions about the story and do all that sort of stuff. Because the context is both the sort of the context that you're working in and the memory available to be able to go back and keep asking questions about the same thing. I don't know if you've ever noticed, but if you're using something like chat and GPT, the further you go, it kind of sometimes stops being useful anymore. It's like, oh, you're just repeating yourself or, you know. Yeah, it gets tired.
26:44Yeah, it gets sort of tired and kind of goes off the rails a little bit. And that's because the context has started to lose the initial part of the conversation and it's gone away. And it's focusing on just maybe the last three or four sort of questions and responses. So, yeah, it's then lost the context of what you were working on and it can kind of go off the rails a little bit. Yeah. Yeah. I always wish I could ask it about itself to be like, do you still remember what we were talking about? Like, you know, things like that, but maybe, yeah. Well, the answer would be yes, because the model is designed to be convincing in its answers. That's one of the things that you notice about these is that they are purposefully designed to create convincing text, even when it's completely wrong. Or, you know, yeah, no, I'm absolutely fine. I think you'll find that you've got the problem. That's why it'll be perfect at being a programmer. It'll work out very well. Absolutely. Yeah. But it's a brilliant, brilliant thing. It's a great area. And the stuff every week I see something new that I think, oh, that's clever. That's a really good, you know, use case. I could see that being influential in some way. So, yeah. Yeah, yeah. Yeah, it'll be interesting to see what if the more options you get if you run things locally. Maybe one day I can just download something and run it instead of having to know what to do. And I'll check that out. I can show you after the show. It's like it's ridiculously easy. Oh, maybe maybe today is tomorrow. No, tomorrow is today. Well, the other thing I wanted to talk about just briefly, like I think since last time we talked, I, you know, we had our we have our state of Kubernetes survey, which which I've made a lot of stuff out of. And there's two more posts that I wrote about, I think if I remember the one we talked about last time was about like the multi cloud, like management, like people using multiple things.
28:52And I've since published one about like the benefits to developers and the business side and also the tools that people use. And I think, you know, the fun thing about the tools thing is being a vendor, of course, we're interested in tools people would pay for, which is which is a fun set of responses to get. And if if, you know, when you look at the tools people are willing to pay for, it's more or less what you would expect, like essentially doing, you know, people like to say monitoring nowadays, but like keeping up with the the runtime status if things are going well with with your old Kubernetes stuff. And people are always interested in paying for security tools and things like that, of course, which which is fine. But, you know, the other thing I wanted to check in because I also like have encountered some some sentiment from people recently. In fact, there was a thread just yesterday in a Slack channel that I'm in. I was looking for some some recommendations of people who would talk about like, you know, what what developers should know about Kubernetes and then true. Technical, technical expert fashion, this this conversation quickly devolved into questioning the premise of the question. And at that point, I was like, oh, looks like it's time for me to go make dinner. Yeah, y'all have fun with this conversation. But, you know, this is this is like the perennial question of of Kubernetes things is how how directly are you expecting developers to access this thing? And, you know, I think I think we contemplate this often. We don't really know for for a little while. But, you know, the one thing that at least I think we'll see that I I've I've been thinking I'm pretty confident in the following statement that unless you have some, you know, I guess, pun intended edge case. Probably you're just going to use like whatever managed Kubernetes stuff that the big cloud vendors have out there. And maybe you'll you'll have a reason to like have to run it on your own.
30:56But I think a lot of this consternating and discussion and and and stuff that it's complex, like a lot of it is like removed if you don't actually have to like run and manage the Kubernetes distro. And or let me phrase that another way. I would hope it was because if that's not included in people saying Kubernetes is complex and. It's a lot worse if people don't take that into account. I think there's a chance that there are two schools of thought on this as a as a as a developer. You know, I've been around developers a lot. I've been in development teams a lot. And I think it sort of boils down to to developers too broad a title. Right. It's just really, really broad. And I think if you're the sort of developer who really likes fiddling about with Kubernetes and getting involved in the weeds of how that's all operating and making, you know, making all that stuff work, then you should go and join the platform engineering team because you'll be happy like every day. That's right. You'll be skipping into work because you'll have that stuff to do in abundance. But if you're the sort of developer who perhaps prefers to be working with end users and gathering requirements and making, you know, end users dreams come true, if you know what I mean, that's probably not going to happen by by moving the levers on Kubernetes. So you're probably more in the sort of user satisfaction space, if you like. And I think there's a sort of a rough sort of separation there. I think there's definitely developers who enjoy doing both. But if your primary sort of, you know, enjoyment comes from digging down into the technology and being in the weeds, you're probably a platform engineer and you'd probably really enjoy that space. So you should probably, you know, try and spend some, try and get more time in that space if you're not in it already.
32:56Yeah. No, I think that distinction is key. And I try to be disciplined about saying application developers versus whatever. But it is, it can be very difficult to tell in surveys and commentary if it's sort of like infrastructure or platform developers or engineers, I guess, or more application developers. I think that's a fun distinction is if you were to say engineer, that perhaps means lower down the stack than when you say developer, which always feels to me a little more like, you know, that's your, comes to work in like flip flops and, you know, Ironman t-shirts and things like that. I think your application developers, whereas, you know, the engineers, I think they might wear pants is, you know, and longer, more formal dress, probably what they show up in. Maybe it also comes down to whether or not you're willing to be on call. Because if you're dipping about in the sort of Kubernetes space, you know, and people are relying on you for those reliability numbers, you know, the five nines and all that, then it's more likely than not, sooner or later, you're going to get a call in two in the morning, you know, to try and start something out. But that doesn't tend to happen in the kind of user satisfaction space, you know, there's no user suddenly has a bright idea at two in the morning and really must, you know, have a brainstorm around a whiteboard. Yeah, yeah, maybe, maybe the application developers creed is no nines here. Just. They have a t-shirt with one of those big red signs that's crossed out that has a nine. Oh, everyone's welcome. But, you know, you, you have your, you know what your own preferences are. If you're okay with being woken up at two in the morning, then, you know, again, probably platform engineers a good space. Now, now speaking, speaking of, you know, the, the ongoing trend of like, you know, the bumpy waters of, of the hype cycle, which, which seems to accelerate more.
35:03It seems to repeat itself. We, we go, you know, kind of like a bumpy waves, you kind of slap your boat along. Like what's, what's your take on the recent, uh, microservices versus monolith commentary been? I think, you know, just to summarize some, some, uh, Amazon prime video streaming people, they wrote this blog post that they moved from a serverless back to a monolith application. And this created a whole big hoopla on both sides of like, Oh, you know, serverless and microservices were always ridiculous. And then as often happens, the sort of, uh, champion is the wrong word, but the, the, the rat, the, the level-headed semi boosters of microservices came in and they're like, well, we've always said it depends that, you know, it's not like, it's not a hundred percent thing that you use that. And, you know, here's, here's links and footnotes and citations to us saying it's always depends. And, uh, you know, I always feel like there's a third part left out, which is, let's call them the, uh, the just like consumer developer who, who's like, yeah, I thought that was what we were supposed to all be doing at this point. So that between these three forces, it's always like a, uh, a little, a little, uh, tri-cornered spider diagram of, of weirdness that's happening. But what's, what's your, uh, what's, what's your summary is that, what have we learned from the most recent brouhaha of, uh, monoliths versus microservices and serverless? Um, I think, you know, personally, I was over it years ago, like the whole argument between these two, you know, types. If you look into stuff that, um, Ollie does, for example, on the spring team, sorry, Ollie, your surnames escaped me for a second, but, um, Ollie's doing micro lifts, for example, where you might, you might design something in an, imagine it in a, in a monolithic fashion for deployment, but it's constructed in such a way that you can then break those pieces apart. And start to use them as microservices if you want to, but there's flexibility in the design to allow you to either coalesce them together or separate them apart at will, as you wish using things like event-driven architecture, for example.
37:18So, you know, it's interesting to see all that stuff going on. Um, I was never totally a fan of microservices for everything because it tend, I mean, particularly with rest, for example, it tended to introduce a lot of architectural issues that you then had to solve elsewhere, um, you know, with other patterns. So, um, I'll think of a good example of that and put it in the show notes, but, um, but there are, there are, there are several sort of issues with, with that, um, which you then start to compensate for by having lots of other things all around you. Like, um, uh, you know, where you've got lots of, um, tools inside the gateway to compensate for other issues that you might have, like services disappearing and coming back because the network's not very reliable. So all sorts of issues around that, particularly with rest and I, I spent a lot of time doing event-driven architecture instead where you're, you're, you're messaging rather than using HTTP and then some of those problems go away. Right. So I'm over it. It's like, yeah, do, do what makes sense. Don't do what is popular. Yeah. Yeah. Yeah. I mean, it, it, it seems like, like, like, uh, my, my, my feel of it. Again, I haven't programmed since 2005, so what do I know? But like, you know, my, my program was always like, yeah, it's a good idea to think, make things like modular and, uh, have kind of clear boundaries between them. And, uh, you should do that. And it, and if you need for it to be like a network, network components that talk with each other, then do that as well. And if you need to like deal with unreliable things and shifting stuff around, then you should add that in as well. But like, beyond that, I, I don't know if there's that much new going on. I mean, other than just redoing it.
39:18And then you get to the point where, uh, uh, you know, I, I wrote this up recently in my newsletter that it's, it's bad to like bad. It's not preferred to, uh, kind of talk about the benefits of something, to market something based on the, uh, the sharp bag of knives theory, which is like, you know, and I think all of us have been guilty of this kind of like talking about stuff to sort of be like, oh, microservices are awesome. But they're really complicated. So you need to like, like, you know, the classic thing from, I think, uh, even like a Martin Fowler blog was like the sign of like, you know, you need to be this tall to ride the ride. To get all these great benefits, you need to deal with the complexity because it's very dangerous and weird. And, you know, that, that kind of conversation about me always strikes me as like, well, sounds like this isn't good for me to use. Then I'll just go back to what I was doing. Like, yeah, I don't know. I don't know what you think, but I feel like I've noticed a lot less tub thumping about serverless in the last sort of year or so as well. And this was also included in this blog from Amazon, wasn't it? They were moving away from their own serverless technology to sort of get better performance by not using it anymore. So I don't know. Serverless is perhaps another one of those things that I don't know if I've noticed as many people caring as deeply about it recently as perhaps a year ago or two years ago. Yeah. Yeah. Well, we'll have to, we'll have to check in. And after, after, you know, in a couple of months after another sausage corner, we'll check in on if, if Kubernetes should be used directly by, by developers. And if serverless is finally going to like bust out of the, the kind of market share concern that it has at the moment. And by concern, I don't mean worrying. I just mean activity. Isn't that a meaning of the word concern? Like a going concern, like a going business that you have? Probably definition number four.
41:18Not really sure. Well, speaking of definition number four, I think the fourth thing we need to go over is that you've been listening to Tanzu Talk or watching it. And if you want to get the show notes for this episode, you can go to TanzuTalk.com and find them for, for this episode. And, or if you're watching the video, you can just look below in the description and I'll put show notes there once I publish it after the streaming here. And with that, we'll see everyone next time. Bye-bye.