Can AI save DevOps? with SystemInit's Adam Jacob

Hi, I'm Scott Hanssel and this is Hi, I'm Scott Hanssel and this is another episode of Hansel Minutes. Today another episode of Hansel Minutes. Today another episode of Hansel Minutes. Today I'm chatting with Adam Jacob. He's the I'm chatting with Adam Jacob. He's the I'm chatting with Adam Jacob. He's the CEO, co-founder, chairman of System CEO, co-founder, chairman of System CEO, co-founder, chairman of System Initiative and the co-founder of uh Chef Initiative and the co-founder of uh Chef Initiative and the co-founder of uh Chef where you were the original author of where you were the original author of where you were the original author of Chef and you were the CTO at Chef. Chef and you were the CTO at Chef. Chef and you were the CTO at Chef. >> That's true. >> That's true. >> That's true. You're like all about the DevOps. You You're like all about the DevOps. You You're like all about the DevOps. You were you always DevOpsy or were you a were you always DevOpsy or were you a were you always DevOpsy or were you a programmer who was like, you know, we programmer who was like, you know, we programmer who was like, you know, we need to build server? need to build server? need to build server? >> Yeah. No, I was always I was always the >> Yeah. No, I was always I was always the >> Yeah. No, I was always I was always the the ops side. Yeah. The history of that the ops side. Yeah. The history of that the ops side. Yeah. The history of that space is filled with people who hated space is filled with people who hated space is filled with people who hated the operations part, you know, and they the operations part, you know, and they the operations part, you know, and they were like, the operations thing was the were like, the operations thing was the were like, the operations thing was the worst and so I became a software worst and so I became a software worst and so I became a software developer so that I could make it so developer so that I could make it so developer so that I could make it so nobody ever had to suffer like I nobody ever had to suffer like I nobody ever had to suffer like I suffered. I'm the opposite. I love suffered. I'm the opposite. I love suffered. I'm the opposite. I love infrastructure. I've always loved infrastructure. I've always loved infrastructure. I've always loved infrastructure. I loved it when I ran infrastructure. I loved it when I ran infrastructure. I loved it when I ran bulletin boards when I was a kid. I love bulletin boards when I was a kid. I love bulletin boards when I was a kid. I love infrastructure now. Like yeah, infrastructure now. Like yeah, infrastructure now. Like yeah, infrastructure is my thing. infrastructure is my thing. infrastructure is my thing. >> Ah, then we are we are people of a >> Ah, then we are we are people of a >> Ah, then we are we are people of a certain age because I I ran a uh Wildcat certain age because I I ran a uh Wildcat certain age because I I ran a uh Wildcat BBS on OS2 in my parents' garage. BBS on OS2 in my parents' garage. BBS on OS2 in my parents' garage. >> Yes. >> Yes. >> Yes. >> Isn't that delightful? And then I just >> Isn't that delightful? And then I just >> Isn't that delightful? And then I just >> I too bought OS2 so that I could have >> I too bought OS2 so that I could have >> I too bought OS2 so that I could have multiple I had multiple modems in the PC multiple I had multiple modems in the PC multiple I had multiple modems in the PC and I needed to do that delightful and I needed to do that delightful and I needed to do that delightful multitasking.

multitasking. multitasking. >> Yeah. >> Yeah. >> Yeah. >> Yeah. >> Yeah. >> Yeah. >> Oh my goodness. two big 144 hazes or >> Oh my goodness. two big 144 hazes or >> Oh my goodness. two big 144 hazes or whatever and then ah the 192 V32 BIS. whatever and then ah the 192 V32 BIS. whatever and then ah the 192 V32 BIS. These are these are the times. These are these are the times. These are these are the times. >> So fun building bulletin boards like >> So fun building bulletin boards like >> So fun building bulletin boards like bulletin boards were the funnest and bulletin boards were the funnest and bulletin boards were the funnest and phytonet was the best. phytonet was the best. phytonet was the best. >> I was a phenet node as well. What a >> I was a phenet node as well. What a >> I was a phenet node as well. What a delight. delight. delight. >> Yeah, I was the official uh I was one of >> Yeah, I was the official uh I was one of >> Yeah, I was the official uh I was one of the seed nodes for the for the official the seed nodes for the for the official the seed nodes for the for the official rooech uh network which was like a like rooech uh network which was like a like rooech uh network which was like a like a rootech devoted phenonet network. a rootech devoted phenonet network. a rootech devoted phenonet network. While we're getting just nerdy about While we're getting just nerdy about While we're getting just nerdy about randomness. randomness. randomness. >> Yes. classic cartoon enough people don't >> Yes. classic cartoon enough people don't >> Yes. classic cartoon enough people don't appreciate. Robotech. appreciate. Robotech. appreciate. Robotech. >> Yeah. When I was a kid, I I I was the >> Yeah. When I was a kid, I I I was the >> Yeah. When I was a kid, I I I was the keeper of the official uh timeline. keeper of the official uh timeline. keeper of the official uh timeline. >> Seriously? Was it asy in a text file? >> Seriously? Was it asy in a text file? >> Seriously? Was it asy in a text file? >> It sure was. >> It sure was. >> It sure was. >> Is it up on text files.com? >> Is it up on text files.com? >> Is it up on text files.com? >> I have no idea. >> I have no idea. >> I have no idea. >> You should go see >> You should go see >> You should go see >> I've actually never looked. I should go. >> I've actually never looked. I should go. >> I've actually never looked. I should go. >> You got to go see if Jason Scott has the >> You got to go see if Jason Scott has the >> You got to go see if Jason Scott has the official rootech. official rootech. official rootech. That would be That would be That would be >> put that on the list of things to do. >> put that on the list of things to do. >> put that on the list of things to do. >> Yeah, that would that would please me >> Yeah, that would that would please me >> Yeah, that would that would please me deeply. deeply. deeply. >> That would be awesome. Yes. >> That would be awesome. Yes. >> That would be awesome. Yes. So um I in my software engineering life So um I in my software engineering life So um I in my software engineering life somewhere I want to say in the mid 90s somewhere I want to say in the mid 90s somewhere I want to say in the mid 90s you know we were still releasing stuff you know we were still releasing stuff you know we were still releasing stuff from you know a computer under Anna's from you know a computer under Anna's from you know a computer under Anna's desk you know like oh that's the build desk you know like oh that's the build desk you know like oh that's the build server and then at some point uh I was server and then at some point uh I was server and then at some point uh I was in C in like early 2000 we had builds in C in like early 2000 we had builds in C in like early 2000 we had builds server.net you know and we started server.net you know and we started server.net you know and we started making build.bat bat, you know, and making build.bat bat, you know, and making build.bat bat, you know, and there were make files and make files there were make files and make files there were make files and make files would run on a cron job, but when do you would run on a cron job, but when do you would run on a cron job, but when do you think like think like think like >> build servers became CI/CD and then >> build servers became CI/CD and then >> build servers became CI/CD and then CI/CD became DevOps?

CI/CD became DevOps? CI/CD became DevOps? >> Two, >> Two, >> Two, it's a tough call cuz it's blurry the it's a tough call cuz it's blurry the it's a tough call cuz it's blurry the lines. I would say uh 2005 lines. I would say uh 2005 lines. I would say uh 2005 is probably when that began to shift by is probably when that began to shift by is probably when that began to shift by the you know and maybe even earlier like the you know and maybe even earlier like the you know and maybe even earlier like we I think I'm bad at dates but I can we I think I'm bad at dates but I can we I think I'm bad at dates but I can tell you what happened from my point of tell you what happened from my point of tell you what happened from my point of view. I'm also bad at dates, but tell me view. I'm also bad at dates, but tell me view. I'm also bad at dates, but tell me what happened. what happened. what happened. >> Which was essentially, you know, around >> Which was essentially, you know, around >> Which was essentially, you know, around the late '9s, early 2000s, we were all the late '9s, early 2000s, we were all the late '9s, early 2000s, we were all trying to figure out like now that you trying to figure out like now that you trying to figure out like now that you could get a lot of people to come do could get a lot of people to come do could get a lot of people to come do stuff on the internet, right? So, the stuff on the internet, right? So, the stuff on the internet, right? So, the first problem was like get people on the first problem was like get people on the first problem was like get people on the internet. The second problem was they internet. The second problem was they internet. The second problem was they need things to do. And so, we started need things to do. And so, we started need things to do. And so, we started building stuff for them to do. And then building stuff for them to do. And then building stuff for them to do. And then when the stuff that they could do on the when the stuff that they could do on the when the stuff that they could do on the internet turned out to be cool, lots of internet turned out to be cool, lots of internet turned out to be cool, lots of people showed up all at once. And then people showed up all at once. And then people showed up all at once. And then we were like, "Oh my gosh, how do we we were like, "Oh my gosh, how do we we were like, "Oh my gosh, how do we deal with like all these people who come deal with like all these people who come deal with like all these people who come to use these things now?" Because we to use these things now?" Because we to use these things now?" Because we can't just like run our systems like can't just like run our systems like can't just like run our systems like they were computer labs and we were they were computer labs and we were they were computer labs and we were shipping you floppy discs. And all of us shipping you floppy discs. And all of us shipping you floppy discs. And all of us sort of independently figured out what sort of independently figured out what sort of independently figured out what would become, you know, CI, what would would become, you know, CI, what would would become, you know, CI, what would become some variation of CD, you know, become some variation of CD, you know, become some variation of CD, you know, like we had normalized build processes, like we had normalized build processes, like we had normalized build processes, we had, you know, configuration we had, you know, configuration we had, you know, configuration management systems. We figured out how management systems. We figured out how management systems. We figured out how to deploy software and we were all doing to deploy software and we were all doing to deploy software and we were all doing it in different ways. And so like I it in different ways. And so like I it in different ways. And so like I probably wrote my first configuration probably wrote my first configuration probably wrote my first configuration management system in 1998, management system in 1998, management system in 1998, right? and then you know the first right? and then you know the first right? and then you know the first deployment system shortly after that and deployment system shortly after that and deployment system shortly after that and then you know had written three or four then you know had written three or four then you know had written three or four by the time I got around to using puppet by the time I got around to using puppet by the time I got around to using puppet and then chef and you know so and I

and then chef and you know so and I and then chef and you know so and I wasn't alone and then velocity which was wasn't alone and then velocity which was wasn't alone and then velocity which was a conference that used to exist that Tim a conference that used to exist that Tim a conference that used to exist that Tim O'Reilly ran that sort of spun out of O'Reilly ran that sort of spun out of O'Reilly ran that sort of spun out of the web 2 expo for that was focused more the web 2 expo for that was focused more the web 2 expo for that was focused more on operations and uh and performance web on operations and uh and performance web on operations and uh and performance web performance um was the first time that a performance um was the first time that a performance um was the first time that a lot of those people who had built those lot of those people who had built those lot of those people who had built those systems all got in the same room at the systems all got in the same room at the systems all got in the same room at the same time. And once we all got in the same time. And once we all got in the same time. And once we all got in the same room at the same time and started same room at the same time and started same room at the same time and started talking to each other, that's when you talking to each other, that's when you talking to each other, that's when you started to see like vocabulary start to started to see like vocabulary start to started to see like vocabulary start to grow, right? Suddenly it was like, ha, grow, right? Suddenly it was like, ha, grow, right? Suddenly it was like, ha, that thing we're doing, that's that thing we're doing, that's that thing we're doing, that's continuous delivery. You know, that continuous delivery. You know, that continuous delivery. You know, that thing's progressive delivery. This thing thing's progressive delivery. This thing thing's progressive delivery. This thing is DevOps. And and it was because all of is DevOps. And and it was because all of is DevOps. And and it was because all of those people who were very desperate those people who were very desperate those people who were very desperate before had no idea each other existed. before had no idea each other existed. before had no idea each other existed. We had no way to know if we were good or We had no way to know if we were good or We had no way to know if we were good or bad at our jobs other than the site bad at our jobs other than the site bad at our jobs other than the site stayed up. You know, suddenly we knew stayed up. You know, suddenly we knew stayed up. You know, suddenly we knew each other and it started to be more each other and it started to be more each other and it started to be more professionalized. And you know to me professionalized. And you know to me professionalized. And you know to me that's sort of where that happened. When that's sort of where that happened. When that's sort of where that happened. When I started thinking about continuous I started thinking about continuous I started thinking about continuous deployment, I wrote a a paper how called deployment, I wrote a a paper how called deployment, I wrote a a paper how called >> word documents won't break the build. >> word documents won't break the build. >> word documents won't break the build. And you know the idea was that like some And you know the idea was that like some And you know the idea was that like some some person was like here's the some person was like here's the some person was like here's the playbook. Here's the recipe. It's in a playbook. Here's the recipe. It's in a playbook. Here's the recipe. It's in a word document like you want to learn how word document like you want to learn how word document like you want to learn how to build the software first to build the software first to build the software first >> get this word document and then someone >> get this word document and then someone >> get this word document and then someone would change the document and they would would change the document and they would would change the document and they would get mad because the build would not get mad because the build would not get mad because the build would not break because they had changed the break because they had changed the break because they had changed the policy. They changed the recipe.

policy. They changed the recipe. policy. They changed the recipe. >> Yeah, for sure. So you got this idea of >> Yeah, for sure. So you got this idea of >> Yeah, for sure. So you got this idea of infrastructure as code like pretty infrastructure as code like pretty infrastructure as code like pretty early. You figured out that word early. You figured out that word early. You figured out that word documents and markdown documents aren't documents and markdown documents aren't documents and markdown documents aren't easy to compile, easy to compile, easy to compile, >> right? And I mean I not only I did I >> right? And I mean I not only I did I >> right? And I mean I not only I did I mean I had that idea like there there mean I had that idea like there there mean I had that idea like there there were many people who had that idea. were many people who had that idea. were many people who had that idea. >> That's a light bulb. It was invented >> That's a light bulb. It was invented >> That's a light bulb. It was invented multiple times by multiple people. multiple times by multiple people. multiple times by multiple people. >> Yeah. a little. I mean, I think to me, >> Yeah. a little. I mean, I think to me, >> Yeah. a little. I mean, I think to me, we're all sort of standing on the we're all sort of standing on the we're all sort of standing on the shoulders of Mark Burgess on this front shoulders of Mark Burgess on this front shoulders of Mark Burgess on this front where like it was Mark. Mark was the guy where like it was Mark. Mark was the guy where like it was Mark. Mark was the guy who wrote CF Engine um came up with who wrote CF Engine um came up with who wrote CF Engine um came up with promise theory and you know Mark's a promise theory and you know Mark's a promise theory and you know Mark's a professor and a scientist and he's got professor and a scientist and he's got professor and a scientist and he's got you know he wrote down the physics of you know he wrote down the physics of you know he wrote down the physics of why these things that we do works and why these things that we do works and why these things that we do works and the rest of us are just sort of living the rest of us are just sort of living the rest of us are just sort of living in his universe I think you know to in his universe I think you know to in his universe I think you know to varying degrees of homage but like I varying degrees of homage but like I varying degrees of homage but like I think yeah the thing about think yeah the thing about think yeah the thing about infrastructure as code is that infrastructure as code is that infrastructure as code is that >> we started getting to this spot where we >> we started getting to this spot where we >> we started getting to this spot where we were like yeah as these systems grow in were like yeah as these systems grow in were like yeah as these systems grow in complexity obviously the thing we need complexity obviously the thing we need complexity obviously the thing we need to do is treat the infrastructure as if to do is treat the infrastructure as if to do is treat the infrastructure as if it was an application artifact, right? it was an application artifact, right? it was an application artifact, right? And just adapt all of the same And just adapt all of the same And just adapt all of the same mechanisms that we adopted for deploying mechanisms that we adopted for deploying mechanisms that we adopted for deploying applications into the infrastructure and applications into the infrastructure and applications into the infrastructure and that being able to write it as code that being able to write it as code that being able to write it as code would be good. And then we sort of fell would be good. And then we sort of fell would be good. And then we sort of fell into camps, you know, there's people who into camps, you know, there's people who into camps, you know, there's people who are like real programming language are like real programming language are like real programming language people and DSLs. That was more me, Chef, people and DSLs. That was more me, Chef, people and DSLs. That was more me, Chef, Lumi, Lumi, Lumi, >> uh the CDK, those things sort of fill >> uh the CDK, those things sort of fill >> uh the CDK, those things sort of fill that void. Then there was the like that void. Then there was the like that void. Then there was the like specialist language camp which is you specialist language camp which is you specialist language camp which is you know more puppet more terraform you know know more puppet more terraform you know know more puppet more terraform you know more anible maybe a little more more anible maybe a little more more anible maybe a little more kubernetes.

kubernetes. kubernetes. >> So Mark did this in like 93 so you're >> So Mark did this in like 93 so you're >> So Mark did this in like 93 so you're right like we sat on we've been sitting right like we sat on we've been sitting right like we sat on we've been sitting on top of the idea of like automating. on top of the idea of like automating. on top of the idea of like automating. >> Oh yeah and Mark Burgess has been >> Oh yeah and Mark Burgess has been >> Oh yeah and Mark Burgess has been writing down how to do that and how to writing down how to do that and how to writing down how to do that and how to think about it and why it will work at think about it and why it will work at think about it and why it will work at scale in incredible notebooks in perfect scale in incredible notebooks in perfect scale in incredible notebooks in perfect handwriting and then publishing handwriting and then publishing handwriting and then publishing scholarly books on how to do it since scholarly books on how to do it since scholarly books on how to do it since then. then. then. >> That's awesome. for 30 plus years. >> That's awesome. for 30 plus years. >> That's awesome. for 30 plus years. >> Yeah. And he's truly one of the most >> Yeah. And he's truly one of the most >> Yeah. And he's truly one of the most incredible people in computer science. incredible people in computer science. incredible people in computer science. >> That's amazing. >> That's amazing. >> That's amazing. So, one question I would have though is So, one question I would have though is So, one question I would have though is that it's the year of our Lord 2025. You that it's the year of our Lord 2025. You that it's the year of our Lord 2025. You know, I don't know about you, but I bump know, I don't know about you, but I bump know, I don't know about you, but I bump into customers all day long that don't into customers all day long that don't into customers all day long that don't even have a decent build server. Like, even have a decent build server. Like, even have a decent build server. Like, the idea that they could check something the idea that they could check something the idea that they could check something in in in >> and without fear go to lunch. >> and without fear go to lunch. >> and without fear go to lunch. >> Yeah. uh that it's just like there's >> Yeah. uh that it's just like there's >> Yeah. uh that it's just like there's there's people zipping up source code there's people zipping up source code there's people zipping up source code out there like out there like out there like >> we're doing all the crazy things. People >> we're doing all the crazy things. People >> we're doing all the crazy things. People think think think >> crazy things. >> crazy things. >> crazy things. >> Yeah, we like to pretend that we're not, >> Yeah, we like to pretend that we're not, >> Yeah, we like to pretend that we're not, but we are. And then even the solutions but we are. And then even the solutions but we are. And then even the solutions we came up with, you know, don't work we came up with, you know, don't work we came up with, you know, don't work super great for most people. Um and you super great for most people. Um and you super great for most people. Um and you know, the analogy I use for know, the analogy I use for know, the analogy I use for infrastructure is code is there's a infrastructure is code is there's a infrastructure is code is there's a video game I love called Dark Souls. You video game I love called Dark Souls. You video game I love called Dark Souls. You ever played Dark Souls?

ever played Dark Souls? ever played Dark Souls? >> I have. And then Elder Ring and all the >> I have. And then Elder Ring and all the >> I have. And then Elder Ring and all the Souls like experiences. Souls like experiences. Souls like experiences. >> Elden Ring. Yep. More people have >> Elden Ring. Yep. More people have >> Elden Ring. Yep. More people have probably played Elder Ring than Dark probably played Elder Ring than Dark probably played Elder Ring than Dark Souls. And the thing about those games Souls. And the thing about those games Souls. And the thing about those games is that they're hard. They're like is that they're hard. They're like is that they're hard. They're like ridiculously difficult. And you spend a ridiculously difficult. And you spend a ridiculously difficult. And you spend a lot of time just dying, just getting lot of time just dying, just getting lot of time just dying, just getting punished, just worked, right? I remember punished, just worked, right? I remember punished, just worked, right? I remember in Dark Souls being stuck on this one in Dark Souls being stuck on this one in Dark Souls being stuck on this one like night. Wasn't even a boss, just a like night. Wasn't even a boss, just a like night. Wasn't even a boss, just a normal homie. And he would just he just normal homie. And he would just he just normal homie. And he would just he just murdered me for like a week, you know? murdered me for like a week, you know? murdered me for like a week, you know? And I felt as good when I figured out And I felt as good when I figured out And I felt as good when I figured out that pattern as I did when like my that pattern as I did when like my that pattern as I did when like my daughter was born, you know? I was like daughter was born, you know? I was like daughter was born, you know? I was like I had accomplished a task, you know? I had accomplished a task, you know? I had accomplished a task, you know? And infrastructure as code is kind of And infrastructure as code is kind of And infrastructure as code is kind of like that. Like if you're a person who's like that. Like if you're a person who's like that. Like if you're a person who's figured out all of the knobs and pieces figured out all of the knobs and pieces figured out all of the knobs and pieces and all of the way that everything fits and all of the way that everything fits and all of the way that everything fits together and you know how to put it all together and you know how to put it all together and you know how to put it all together just so and then when it all together just so and then when it all together just so and then when it all snaps together the machine runs, you snaps together the machine runs, you snaps together the machine runs, you know, and you're like the crazy guy with know, and you're like the crazy guy with know, and you're like the crazy guy with the yarn, you know, like and because it the yarn, you know, like and because it the yarn, you know, like and because it was so hard and because you had to learn was so hard and because you had to learn was so hard and because you had to learn all that stuff, you're also convinced it all that stuff, you're also convinced it all that stuff, you're also convinced it was good, was good, was good, >> right? you're also like and this was a >> right? you're also like and this was a >> right? you're also like and this was a fun time and like for most people it was fun time and like for most people it was fun time and like for most people it was not you know like yep there is a not you know like yep there is a not you know like yep there is a particular human who like oh that's the particular human who like oh that's the particular human who like oh that's the best uh and and incredible good for us best uh and and incredible good for us best uh and and incredible good for us and also the reason we're all still and also the reason we're all still and also the reason we're all still zipping stuff up and doing whatever zipping stuff up and doing whatever zipping stuff up and doing whatever works is because for most people that's works is because for most people that's works is because for most people that's not a fun time actually just getting not a fun time actually just getting not a fun time actually just getting punched in the face over and over and punched in the face over and over and punched in the face over and over and over again uh and just never getting any over again uh and just never getting any over again uh and just never getting any progress right progress right progress right >> I think of it as being somewhat Rube >> I think of it as being somewhat Rube >> I think of it as being somewhat Rube Goldbergian if you're familiar with the Goldbergian if you're familiar with the Goldbergian if you're familiar with the Rube Goldberg problems where it's like Rube Goldberg problems where it's like Rube Goldberg problems where it's like somebody you knocks over a domino and somebody you knocks over a domino and somebody you knocks over a domino and then that makes the bowling ball fall then that makes the bowling ball fall then that makes the bowling ball fall onto the yeah iron and I think that's

onto the yeah iron and I think that's onto the yeah iron and I think that's art and other people think that's art and other people think that's art and other people think that's insane. insane. insane. >> I think it is both. >> I think it is both. >> I think it is both. It's it is art and it is insane. Like It's it is art and it is insane. Like It's it is art and it is insane. Like when you think about it from the point when you think about it from the point when you think about it from the point of view of being able to design systems of view of being able to design systems of view of being able to design systems to solve our problems, it's kind of to solve our problems, it's kind of to solve our problems, it's kind of bananas. Like from that lens, it's like bananas. Like from that lens, it's like bananas. Like from that lens, it's like we should be able to do better than we should be able to do better than we should be able to do better than we're doing and we haven't been able to. we're doing and we haven't been able to. we're doing and we haven't been able to. And it's taken us a really long time to And it's taken us a really long time to And it's taken us a really long time to even come up with a concept of like what even come up with a concept of like what even come up with a concept of like what something else could even be. And you something else could even be. And you something else could even be. And you know the thing about the status quo that know the thing about the status quo that know the thing about the status quo that is amazing is that it works. Like you is amazing is that it works. Like you is amazing is that it works. Like you can do it and it is art and that is can do it and it is art and that is can do it and it is art and that is great. And also wow you know that it great. And also wow you know that it great. And also wow you know that it works is kind of incredible. And you works is kind of incredible. And you works is kind of incredible. And you know, part of what I've been doing the know, part of what I've been doing the know, part of what I've been doing the last 6 years is the R&D to be like, last 6 years is the R&D to be like, last 6 years is the R&D to be like, okay, if I don't love the outcomes, and okay, if I don't love the outcomes, and okay, if I don't love the outcomes, and I don't, you know, I watched a lot of I I don't, you know, I watched a lot of I I don't, you know, I watched a lot of I taught a lot of people how to do DevOps, taught a lot of people how to do DevOps, taught a lot of people how to do DevOps, and I taught them how to do and I taught them how to do and I taught them how to do configuration management, and I taught configuration management, and I taught configuration management, and I taught them how to move into the cloud, and I them how to move into the cloud, and I them how to move into the cloud, and I taught them how to do a lot of those taught them how to do a lot of those taught them how to do a lot of those things, and they're all better off for things, and they're all better off for things, and they're all better off for it, you know, like they tell me all the it, you know, like they tell me all the it, you know, like they tell me all the time that that they're pleased that that time that that they're pleased that that time that that they're pleased that that happened, that their careers were happened, that their careers were happened, that their careers were better, you know, like it's better if better, you know, like it's better if better, you know, like it's better if you have those things. It's better to you have those things. It's better to you have those things. It's better to have them than not. And then they have them than not. And then they have them than not. And then they proceed to tell me, all of them, how proceed to tell me, all of them, how proceed to tell me, all of them, how much they don't like it. They're like, much they don't like it. They're like, much they don't like it. They're like, "Ugh, and this happened today. It was "Ugh, and this happened today. It was "Ugh, and this happened today. It was awful. I hate how this thing works. I awful. I hate how this thing works. I awful. I hate how this thing works. I hate how this thing comes together."

hate how this thing comes together." hate how this thing comes together." Like, it's literally the worst part of Like, it's literally the worst part of Like, it's literally the worst part of the stack when it comes to like the the stack when it comes to like the the stack when it comes to like the actual experience of doing the work. Uh, actual experience of doing the work. Uh, actual experience of doing the work. Uh, it's hard to collaborate with other it's hard to collaborate with other it's hard to collaborate with other people. Uh, it's hard to remember all people. Uh, it's hard to remember all people. Uh, it's hard to remember all the pieces that have to fit together. the pieces that have to fit together. the pieces that have to fit together. You get one thing wrong, everything You get one thing wrong, everything You get one thing wrong, everything breaks, the feedback loops are too slow. breaks, the feedback loops are too slow. breaks, the feedback loops are too slow. There's like compliance regimes that There's like compliance regimes that There's like compliance regimes that you're supposed to comply to, but you you're supposed to comply to, but you you're supposed to comply to, but you don't know you're out of compliance till don't know you're out of compliance till don't know you're out of compliance till you get an audit that's six months after you get an audit that's six months after you get an audit that's six months after you made the change that broke the build you made the change that broke the build you made the change that broke the build and then you have to figure out which and then you have to figure out which and then you have to figure out which one did it and you can't figure out why one did it and you can't figure out why one did it and you can't figure out why because they have no audit trail, you because they have no audit trail, you because they have no audit trail, you know, like it's bad. know, like it's bad. know, like it's bad. >> And that's the stuff that I have been >> And that's the stuff that I have been >> And that's the stuff that I have been blessed enough to get to a spot in my blessed enough to get to a spot in my blessed enough to get to a spot in my career where I could like go think about career where I could like go think about career where I could like go think about that problem, you know, like how do you that problem, you know, like how do you that problem, you know, like how do you solve that solve that solve that >> and where that is such a big mess. I've >> and where that is such a big mess. I've >> and where that is such a big mess. I've got stuff that I wrote in like 99 that's got stuff that I wrote in like 99 that's got stuff that I wrote in like 99 that's still on the web and there's, you know, still on the web and there's, you know, still on the web and there's, you know, I talk about bit rot. I was trying to I talk about bit rot. I was trying to I talk about bit rot. I was trying to explain to someone who's 5 years into explain to someone who's 5 years into explain to someone who's 5 years into their career what bit rot feels like their career what bit rot feels like their career what bit rot feels like >> and it's like you got to have 25 >> and it's like you got to have 25 >> and it's like you got to have 25 30-year-old stuff 30-year-old stuff 30-year-old stuff >> and watch it break. So like I haven't >> and watch it break. So like I haven't >> and watch it break. So like I haven't updated my podcast site in like seven updated my podcast site in like seven updated my podcast site in like seven months and I had a small change and I months and I had a small change and I months and I had a small change and I was feeling quite clever. So, I get the was feeling quite clever. So, I get the was feeling quite clever. So, I get the GitHub mobile app on my phone and I do a GitHub mobile app on my phone and I do a GitHub mobile app on my phone and I do a PR directly in my phone. I think I'm so PR directly in my phone. I think I'm so PR directly in my phone. I think I'm so fancy. I'm doing a PR from my phone. And fancy. I'm doing a PR from my phone. And fancy. I'm doing a PR from my phone. And then I go home later, you know, then I go home later, you know, then I go home later, you know, expecting the thing to be live and I expecting the thing to be live and I expecting the thing to be live and I find out that some bits have rotted and find out that some bits have rotted and find out that some bits have rotted and it's like this Docker um base image is it's like this Docker um base image is it's like this Docker um base image is no longer appropriate. I need to use no longer appropriate. I need to use no longer appropriate. I need to use that one. And there's a dependabot pull that one. And there's a dependabot pull that one. And there's a dependabot pull request about it and I'd ignored it. And request about it and I'd ignored it. And request about it and I'd ignored it. And a certificate has expired and a personal a certificate has expired and a personal a certificate has expired and a personal access token that I because I haven't

access token that I because I haven't access token that I because I haven't pushed I haven't built this thing. So, pushed I haven't built this thing. So, pushed I haven't built this thing. So, do I need to like build it every day? do I need to like build it every day? do I need to like build it every day? Like, it's rotted. It only took six Like, it's rotted. It only took six Like, it's rotted. It only took six months, months, months, >> right? >> right? >> right? >> Yeah. And then take that across the >> Yeah. And then take that across the >> Yeah. And then take that across the enterprise. enterprise. enterprise. >> Exactly. >> Exactly. >> Exactly. >> By what we mean by the enterprise is the >> By what we mean by the enterprise is the >> By what we mean by the enterprise is the stuff that runs the planet. Like the stuff that runs the planet. Like the stuff that runs the planet. Like the most important technology that runs the most important technology that runs the most important technology that runs the workloads that matter the most in your workloads that matter the most in your workloads that matter the most in your life, that are the least interesting in life, that are the least interesting in life, that are the least interesting in some ways, but that literally hold the some ways, but that literally hold the some ways, but that literally hold the world up. like there is loadbearing world up. like there is loadbearing world up. like there is loadbearing infrastructure that's like that you know infrastructure that's like that you know infrastructure that's like that you know like uh I was working at a multinational like uh I was working at a multinational like uh I was working at a multinational bank that no longer exists actually this bank that no longer exists actually this bank that no longer exists actually this is not why but they ran credit card is not why but they ran credit card is not why but they ran credit card transactions in Brazil and to do it transactions in Brazil and to do it transactions in Brazil and to do it there was like a Brazilian banking there was like a Brazilian banking there was like a Brazilian banking reconciliation law they had to comply reconciliation law they had to comply reconciliation law they had to comply with and they complied with this law by with and they complied with this law by with and they complied with this law by having a piece of software that was having a piece of software that was having a piece of software that was written in Deli that was deployed on a written in Deli that was deployed on a written in Deli that was deployed on a Windows machine that was literally under Windows machine that was literally under Windows machine that was literally under someone's desk someone's desk someone's desk >> they didn't know how to deploy it they >> they didn't know how to deploy it they >> they didn't know how to deploy it they didn't know where the source code was didn't know where the source code was didn't know where the source code was just needed that machine to stay turned just needed that machine to stay turned just needed that machine to stay turned on so that they didn't go to jail, you on so that they didn't go to jail, you on so that they didn't go to jail, you know, and they are not alone, you know, know, and they are not alone, you know, know, and they are not alone, you know, that exists in every single enterprise that exists in every single enterprise that exists in every single enterprise everywhere on the planet. We just don't everywhere on the planet. We just don't everywhere on the planet. We just don't look at it, you know. We just like to look at it, you know. We just like to look at it, you know. We just like to pretend that it doesn't.

pretend that it doesn't. pretend that it doesn't. >> We definitely do. I I was um if you >> We definitely do. I I was um if you >> We definitely do. I I was um if you remember when Crowd Strike happened in remember when Crowd Strike happened in remember when Crowd Strike happened in the entire world like blue screened. the entire world like blue screened. the entire world like blue screened. >> Yeah. >> Yeah. >> Yeah. >> I was in the Berlin airport >> I was in the Berlin airport >> I was in the Berlin airport >> just hanging out getting ready to get on >> just hanging out getting ready to get on >> just hanging out getting ready to get on a flight and the airport a flight and the airport a flight and the airport >> bluec screened, >> bluec screened, >> bluec screened, >> right? It's who and it was just the it >> right? It's who and it was just the it >> right? It's who and it was just the it was the most surreal thing and it's like was the most surreal thing and it's like was the most surreal thing and it's like all the little things you don't think all the little things you don't think all the little things you don't think about. A buddy of mine literally just about. A buddy of mine literally just about. A buddy of mine literally just before this this uh podcast texted me before this this uh podcast texted me before this this uh podcast texted me that something was wrong with his that something was wrong with his that something was wrong with his insulin pump and we started debugging insulin pump and we started debugging insulin pump and we started debugging the firmware on some open source the firmware on some open source the firmware on some open source software that we use for our open source software that we use for our open source software that we use for our open source artificial pancreases and the the uh artificial pancreases and the the uh artificial pancreases and the the uh company that makes the insulin pumps company that makes the insulin pumps company that makes the insulin pumps swapped out the Bluetooth chipset and swapped out the Bluetooth chipset and swapped out the Bluetooth chipset and like didn't tell anyone like didn't tell anyone like didn't tell anyone >> and there's like this this you know what >> and there's like this this you know what >> and there's like this this you know what I mean like and he just got a new batch I mean like and he just got a new batch I mean like and he just got a new batch in the mail. in the mail. in the mail. >> Right. Right. Right. And he could have >> Right. Right. Right. And he could have >> Right. Right. Right. And he could have just gotten on a plane and flown away just gotten on a plane and flown away just gotten on a plane and flown away and they like never would have noticed and they like never would have noticed and they like never would have noticed that that batch and it's only iPhone 16s that that batch and it's only iPhone 16s that that batch and it's only iPhone 16s plus this board plus Bluetooth plus iOS plus this board plus Bluetooth plus iOS plus this board plus Bluetooth plus iOS this and that insulin pump and then they this and that insulin pump and then they this and that insulin pump and then they no longer talk to each other. And he'd no longer talk to each other. And he'd no longer talk to each other. And he'd been using this for like years and now been using this for like years and now been using this for like years and now like this is like a full-on crisis.

like this is like a full-on crisis. like this is like a full-on crisis. >> Yeah. Now it's an actual emergency like >> Yeah. Now it's an actual emergency like >> Yeah. Now it's an actual emergency like a someone can die. a someone can die. a someone can die. >> Yeah. It's it's it's not awesome. So >> Yeah. It's it's it's not awesome. So >> Yeah. It's it's it's not awesome. So then here's the question. M you spent then here's the question. M you spent then here's the question. M you spent the first half of your career working on the first half of your career working on the first half of your career working on things like chef, working on CI/CD, things like chef, working on CI/CD, things like chef, working on CI/CD, making making making >> things deterministic. >> things deterministic. >> things deterministic. >> Yes. >> Yes. >> Yes. >> Everything is about determinism, >> Everything is about determinism, >> Everything is about determinism, reliability. CI/CD, reliability. CI/CD, reliability. CI/CD, >> the D should be for determinism, not >> the D should be for determinism, not >> the D should be for determinism, not deployment. deployment. deployment. >> Yeah. >> Yeah. >> Yeah. >> But your new thing system and it is >> But your new thing system and it is >> But your new thing system and it is automated infrastructure with AI. And automated infrastructure with AI. And automated infrastructure with AI. And AI, when I think AI, the first thing I AI, when I think AI, the first thing I AI, when I think AI, the first thing I think is nondeterminist. think is nondeterminist. think is nondeterminist. >> Net deterministic. Yeah. >> Net deterministic. Yeah. >> Net deterministic. Yeah. >> So that's insane. >> So that's insane. >> So that's insane. >> Well, it's interesting, right? Because >> Well, it's interesting, right? Because >> Well, it's interesting, right? Because what's interesting about AI to me, um, what's interesting about AI to me, um, what's interesting about AI to me, um, like there's lots of analogies for how like there's lots of analogies for how like there's lots of analogies for how to understand it, but I think a good way to understand it, but I think a good way to understand it, but I think a good way to understand it is that the current to understand it is that the current to understand it is that the current crop of of AI, you know, LLMs, crop of of AI, you know, LLMs, crop of of AI, you know, LLMs, generative AI, whatever you want to call generative AI, whatever you want to call generative AI, whatever you want to call them, them, them, >> what those machines mostly let us do >> what those machines mostly let us do >> what those machines mostly let us do that we couldn't do before is they let that we couldn't do before is they let that we couldn't do before is they let computers be creative in a way that computers be creative in a way that computers be creative in a way that computers could not be before. And, you computers could not be before. And, you computers could not be before. And, you know, they do that by building this know, they do that by building this know, they do that by building this giant map of all the data and putting it giant map of all the data and putting it giant map of all the data and putting it in relationship to other things and in relationship to other things and in relationship to other things and predicting the next token or whatever.

predicting the next token or whatever. predicting the next token or whatever. But the capability that that gives us is But the capability that that gives us is But the capability that that gives us is the ability for the machine to take the ability for the machine to take the ability for the machine to take create to do something creative that we create to do something creative that we create to do something creative that we didn't ask it directly to do. Right? And didn't ask it directly to do. Right? And didn't ask it directly to do. Right? And that's kind of amazing. That's why you that's kind of amazing. That's why you that's kind of amazing. That's why you can ask it to like, you know, a good can ask it to like, you know, a good can ask it to like, you know, a good example was I asked it to write me a example was I asked it to write me a example was I asked it to write me a Shakespearean sonnet about honeycomb the Shakespearean sonnet about honeycomb the Shakespearean sonnet about honeycomb the other day when I gave a talk at other day when I gave a talk at other day when I gave a talk at Honeycomb's observability conference. I Honeycomb's observability conference. I Honeycomb's observability conference. I think I'm the first person who probably think I'm the first person who probably think I'm the first person who probably asked the machine to do that. asked the machine to do that. asked the machine to do that. >> But there there it went and it wrote a >> But there there it went and it wrote a >> But there there it went and it wrote a perfectly reasonable sonnet. And it's perfectly reasonable sonnet. And it's perfectly reasonable sonnet. And it's the non-determinism that makes the the non-determinism that makes the the non-determinism that makes the sonnet possible. Right? If you tried to sonnet possible. Right? If you tried to sonnet possible. Right? If you tried to write a program to write sonnetss, it write a program to write sonnetss, it write a program to write sonnetss, it wouldn't write very good ones, right? wouldn't write very good ones, right? wouldn't write very good ones, right? They wouldn't be clever. They wouldn't They wouldn't be clever. They wouldn't They wouldn't be clever. They wouldn't be interesting. They wouldn't be able to be interesting. They wouldn't be able to be interesting. They wouldn't be able to use the data about the thing I want to use the data about the thing I want to use the data about the thing I want to talk about. Like, it's hard to write a talk about. Like, it's hard to write a talk about. Like, it's hard to write a deterministic sonic generator that feels deterministic sonic generator that feels deterministic sonic generator that feels natural. And yet, LM just sort of natural. And yet, LM just sort of natural. And yet, LM just sort of creatively figure out how to do it. But creatively figure out how to do it. But creatively figure out how to do it. But they're not very intelligent, right? they're not very intelligent, right? they're not very intelligent, right? They're just what they are is creative, They're just what they are is creative, They're just what they are is creative, which is a form of intelligence, which which is a form of intelligence, which which is a form of intelligence, which is incredible. But I wouldn't call them is incredible. But I wouldn't call them is incredible. But I wouldn't call them particularly smart, right? Like my cat particularly smart, right? Like my cat particularly smart, right? Like my cat is smart. She like sneaks out of the is smart. She like sneaks out of the is smart. She like sneaks out of the house because she is not allowed to house because she is not allowed to house because she is not allowed to leave the house. And then when she and leave the house. And then when she and leave the house. And then when she and she's always very noisy. I have a very she's always very noisy. I have a very she's always very noisy. I have a very loud cat. And when she sneaks out, she's loud cat. And when she sneaks out, she's loud cat. And when she sneaks out, she's silent because she knows if she makes a silent because she knows if she makes a silent because she knows if she makes a noise, I'm going to come get her and she noise, I'm going to come get her and she noise, I'm going to come get her and she has to go back inside. She doesn't want has to go back inside. She doesn't want has to go back inside. She doesn't want to, right? And like she'll spend the to, right? And like she'll spend the to, right? And like she'll spend the whole day outside in this vegetable whole day outside in this vegetable whole day outside in this vegetable garden out under my window never making garden out under my window never making garden out under my window never making a noise because if she makes a single a noise because if she makes a single a noise because if she makes a single sound, she knows I'm just going to walk sound, she knows I'm just going to walk sound, she knows I'm just going to walk right out and grab her. That cat's right out and grab her. That cat's right out and grab her. That cat's intelligent, smart, figured some things intelligent, smart, figured some things intelligent, smart, figured some things out. LM aren't. They're creative. So out. LM aren't. They're creative. So out. LM aren't. They're creative. So when it comes to infrastructure, one of when it comes to infrastructure, one of when it comes to infrastructure, one of the hardest things we have is that it's the hardest things we have is that it's the hardest things we have is that it's difficult to remember just like what all

difficult to remember just like what all difficult to remember just like what all the things are you want it to do. So if the things are you want it to do. So if the things are you want it to do. So if you listen to what people say, they can you listen to what people say, they can you listen to what people say, they can say things like I want to deploy this say things like I want to deploy this say things like I want to deploy this docker container to an ECS cluster in docker container to an ECS cluster in docker container to an ECS cluster in Amazon or I want to build a VPC and that Amazon or I want to build a VPC and that Amazon or I want to build a VPC and that sentence almost everyone can write. And sentence almost everyone can write. And sentence almost everyone can write. And the LLM can translate that pretty the LLM can translate that pretty the LLM can translate that pretty effectively into infrastructure if what effectively into infrastructure if what effectively into infrastructure if what you feed it is a deterministic layer you feed it is a deterministic layer you feed it is a deterministic layer underneath whose job is to take that underneath whose job is to take that underneath whose job is to take that information in in a structured way and information in in a structured way and information in in a structured way and provide feedback to the LLM about provide feedback to the LLM about provide feedback to the LLM about whether what it's proposing is good or whether what it's proposing is good or whether what it's proposing is good or bad. So if you tell it how to propose bad. So if you tell it how to propose bad. So if you tell it how to propose that infrastructure, if you tell it, that infrastructure, if you tell it, that infrastructure, if you tell it, hey, here's the specification and here's hey, here's the specification and here's hey, here's the specification and here's how you work with it and it maps one to how you work with it and it maps one to how you work with it and it maps one to one to the data you already have in the one to the data you already have in the one to the data you already have in the training set, then you can harness that training set, then you can harness that training set, then you can harness that creativity. you can put it with creativity. you can put it with creativity. you can put it with something that deterministically goes something that deterministically goes something that deterministically goes and proposes this infrastructure, and proposes this infrastructure, and proposes this infrastructure, evaluates it for quality, tells you if evaluates it for quality, tells you if evaluates it for quality, tells you if it's going to work or not work, and does it's going to work or not work, and does it's going to work or not work, and does that in immediate loop instead of a long that in immediate loop instead of a long that in immediate loop instead of a long one, right? So, it's not the end of a PR one, right? So, it's not the end of a PR one, right? So, it's not the end of a PR cycle. It's at the moment that you cycle. It's at the moment that you cycle. It's at the moment that you proposed a configuration change proposed a configuration change proposed a configuration change and suddenly the the whole thing works and suddenly the the whole thing works and suddenly the the whole thing works fantastically well. You can ask it fantastically well. You can ask it fantastically well. You can ask it questions. You know, we do things with questions. You know, we do things with questions. You know, we do things with it like uh migrate between it like uh migrate between it like uh migrate between architectures, like deploy some stuff to architectures, like deploy some stuff to architectures, like deploy some stuff to EC2 instances, then go look at the EC2 instances, then go look at the EC2 instances, then go look at the instances, figure out the workload instances, figure out the workload instances, figure out the workload that's running on them, then translate that's running on them, then translate that's running on them, then translate that into containerized deployments on a that into containerized deployments on a that into containerized deployments on a different architecture, and it just different architecture, and it just different architecture, and it just works. Um, and it's because there's this works. Um, and it's because there's this works. Um, and it's because there's this deterministic layer underneath it that deterministic layer underneath it that deterministic layer underneath it that system initiative provides that is this system initiative provides that is this system initiative provides that is this one:one model of how all the one:one model of how all the one:one model of how all the infrastructure functions. And then the

infrastructure functions. And then the infrastructure functions. And then the creativity is the LLM. creativity is the LLM. creativity is the LLM. Hey friends, this is Scott. I'm going to Hey friends, this is Scott. I'm going to Hey friends, this is Scott. I'm going to take a moment in the middle here and take a moment in the middle here and take a moment in the middle here and introduce you one of our new sponsors, introduce you one of our new sponsors, introduce you one of our new sponsors, Postman Flows. They let you orchestrate Postman Flows. They let you orchestrate Postman Flows. They let you orchestrate APIs, logic, and data visually. I've APIs, logic, and data visually. I've APIs, logic, and data visually. I've been chatting with Rodrik. Rodri, what been chatting with Rodrik. Rodri, what been chatting with Rodrik. Rodri, what has been the aha moment for you when you has been the aha moment for you when you has been the aha moment for you when you were building agents with Postman flows? were building agents with Postman flows? were building agents with Postman flows? >> The first thing that I've noticed is >> The first thing that I've noticed is >> The first thing that I've noticed is that when you connect an API to an AI, that when you connect an API to an AI, that when you connect an API to an AI, then it really transforms what you can then it really transforms what you can then it really transforms what you can do and the capabilities that the large do and the capabilities that the large do and the capabilities that the large language model can do for you. Um, so language model can do for you. Um, so language model can do for you. Um, so the aha moment for me was that very the aha moment for me was that very the aha moment for me was that very first connection and unlocking the first connection and unlocking the first connection and unlocking the ability of taking a AI which was ability of taking a AI which was ability of taking a AI which was essentially transactional with respect essentially transactional with respect essentially transactional with respect to text in text out and changing it into to text in text out and changing it into to text in text out and changing it into something that was outcome driven text something that was outcome driven text something that was outcome driven text in actions out and the way to do that in actions out and the way to do that in actions out and the way to do that was through APIs and using all of the was through APIs and using all of the was through APIs and using all of the Postman APIs that are on the network Postman APIs that are on the network Postman APIs that are on the network today over 100,000 plus was really the today over 100,000 plus was really the today over 100,000 plus was really the thing that I wanted to sort of just thing that I wanted to sort of just thing that I wanted to sort of just drive more and more of drive more and more of drive more and more of >> text in actions out actions uh >> text in actions out actions uh >> text in actions out actions uh text in actions out. I love it. Postman text in actions out. I love it. Postman text in actions out. I love it. Postman flows. It's a powerful tool for creating flows. It's a powerful tool for creating flows. It's a powerful tool for creating and running AI agents and API and running AI agents and API and running AI agents and API automations. You can check them out at automations. You can check them out at automations. You can check them out at postman.com/hancel postman.com/hancel postman.com/hancel minutes.

minutes. minutes. >> So, in the interest of discussion >> So, in the interest of discussion >> So, in the interest of discussion without like pushing back, but also not without like pushing back, but also not without like pushing back, but also not trying to like rat hole, trying to like rat hole, trying to like rat hole, >> you said creativity a lot and I kind of >> you said creativity a lot and I kind of >> you said creativity a lot and I kind of like had a reaction to that because I like had a reaction to that because I like had a reaction to that because I feel like feel like feel like >> randomness isn't creativity. Like next >> randomness isn't creativity. Like next >> randomness isn't creativity. Like next token prediction isn't creativity. It token prediction isn't creativity. It token prediction isn't creativity. It feels like it's Mad Libs. feels like it's Mad Libs. feels like it's Mad Libs. >> And Mad Libs, they're wacky because we >> And Mad Libs, they're wacky because we >> And Mad Libs, they're wacky because we were four and like, "Oh, purple were four and like, "Oh, purple were four and like, "Oh, purple underwear. Oh, green green eggs." So, underwear. Oh, green green eggs." So, underwear. Oh, green green eggs." So, like if I if I had a site go down and I like if I if I had a site go down and I like if I if I had a site go down and I had an LLM that was responsible for had an LLM that was responsible for had an LLM that was responsible for getting the site back up and and I was getting the site back up and and I was getting the site back up and and I was told, "Don't wake me up. It's going to told, "Don't wake me up. It's going to told, "Don't wake me up. It's going to iterate. It's going to, you know, it's iterate. It's going to, you know, it's iterate. It's going to, you know, it's going to going to going to >> kick the kick the EC2 instance. It's >> kick the kick the EC2 instance. It's >> kick the kick the EC2 instance. It's going to, you know, retry the build the going to, you know, retry the build the going to, you know, retry the build the Docker. It's going to do stuff. And just Docker. It's going to do stuff. And just Docker. It's going to do stuff. And just because it succeeded doesn't mean it was because it succeeded doesn't mean it was because it succeeded doesn't mean it was creative. It could have just been really creative. It could have just been really creative. It could have just been really really persistent. really persistent. really persistent. >> Yeah, I fair enough. Here's the thing >> Yeah, I fair enough. Here's the thing >> Yeah, I fair enough. Here's the thing about here's why I think creativity is about here's why I think creativity is about here's why I think creativity is actually the right metaphor. So of actually the right metaphor. So of actually the right metaphor. So of course you're right that like at its course you're right that like at its course you're right that like at its most brute force level it's Mad Libs and most brute force level it's Mad Libs and most brute force level it's Mad Libs and it's better than Mad Libs because it's better than Mad Libs because it's better than Mad Libs because when given the right set of context and when given the right set of context and when given the right set of context and given the ability to attach it to given the ability to attach it to given the ability to attach it to deterministic tooling what happens is deterministic tooling what happens is deterministic tooling what happens is you thin the layer of what the LLM needs you thin the layer of what the LLM needs you thin the layer of what the LLM needs to do.

to do. to do. >> So the more you ask the LLM to do the >> So the more you ask the LLM to do the >> So the more you ask the LLM to do the more it's like mad lips more it's like mad lips more it's like mad lips >> right? So, and it's very paradoxical >> right? So, and it's very paradoxical >> right? So, and it's very paradoxical because the first thing everybody does because the first thing everybody does because the first thing everybody does when you approach AI is be like, "Well, when you approach AI is be like, "Well, when you approach AI is be like, "Well, let's just put AI everywhere. The let's just put AI everywhere. The let's just put AI everywhere. The magical intelligent robot's going to magical intelligent robot's going to magical intelligent robot's going to figure out how to do all this shoo for figure out how to do all this shoo for figure out how to do all this shoo for me." And, you know, now I don't have to me." And, you know, now I don't have to me." And, you know, now I don't have to get paged at night because it's going to get paged at night because it's going to get paged at night because it's going to wake me up because it's going to wake me up because it's going to wake me up because it's going to automatically fix all my infrastructure. automatically fix all my infrastructure. automatically fix all my infrastructure. And like, that's not going to work, And like, that's not going to work, And like, that's not going to work, right? Because the LLM is not going to right? Because the LLM is not going to right? Because the LLM is not going to do a very good job. And it's not because do a very good job. And it's not because do a very good job. And it's not because the LLM couldn't make good choices. It the LLM couldn't make good choices. It the LLM couldn't make good choices. It could. It's actually they're pretty good could. It's actually they're pretty good could. It's actually they're pretty good at making the right next choice. at making the right next choice. at making the right next choice. Sometimes they'll make the wrong one, Sometimes they'll make the wrong one, Sometimes they'll make the wrong one, but you have and you can correct them. but you have and you can correct them. but you have and you can correct them. But they're pretty good at making the But they're pretty good at making the But they're pretty good at making the right next choice in general if what you right next choice in general if what you right next choice in general if what you have are systems connected to them that have are systems connected to them that have are systems connected to them that are designed to give them the feedback are designed to give them the feedback are designed to give them the feedback they need to know if the choices they're they need to know if the choices they're they need to know if the choices they're making are good or bad. And that's where making are good or bad. And that's where making are good or bad. And that's where we fall down when we're thinking about a we fall down when we're thinking about a we fall down when we're thinking about a lot of the modern systems we build. lot of the modern systems we build. lot of the modern systems we build. They're just not built to actually work They're just not built to actually work They're just not built to actually work with how LLMs function as technology. with how LLMs function as technology. with how LLMs function as technology. Most people just rubbed an LLM all over Most people just rubbed an LLM all over Most people just rubbed an LLM all over what they already had, right? They're what they already had, right? They're what they already had, right? They're like, "Well, I guess I'll just rub an like, "Well, I guess I'll just rub an like, "Well, I guess I'll just rub an LLM all over my codebase and ask it to LLM all over my codebase and ask it to LLM all over my codebase and ask it to do wackiness." And you're like, "Yeah, do wackiness." And you're like, "Yeah, do wackiness." And you're like, "Yeah, woo, that doesn't go as good as you woo, that doesn't go as good as you woo, that doesn't go as good as you hope." You know, like suddenly code hope." You know, like suddenly code hope." You know, like suddenly code quality is awful and suddenly all those quality is awful and suddenly all those quality is awful and suddenly all those things don't work. Well, it's because of things don't work. Well, it's because of things don't work. Well, it's because of the loop that was actually kind of bad the loop that was actually kind of bad the loop that was actually kind of bad for people, funnily enough. Like, if you for people, funnily enough. Like, if you for people, funnily enough. Like, if you really think about the testing loop, really think about the testing loop, really think about the testing loop, like it's weird that you don't know that like it's weird that you don't know that like it's weird that you don't know that you formatted the code wrong till later.

you formatted the code wrong till later. you formatted the code wrong till later. You know, why don't I know when I when I You know, why don't I know when I when I You know, why don't I know when I when I finished writing the statement that it finished writing the statement that it finished writing the statement that it was wrongly formatted? you know, if I was wrongly formatted? you know, if I was wrongly formatted? you know, if I knew that'd be better and that would knew that'd be better and that would knew that'd be better and that would also make the LLM loop better and then also make the LLM loop better and then also make the LLM loop better and then it would stop having so much garbage. it would stop having so much garbage. it would stop having so much garbage. So, like yeah, when I think about it as So, like yeah, when I think about it as So, like yeah, when I think about it as creativity, it's not because it's not creativity, it's not because it's not creativity, it's not because it's not sort of going at randomness, but that sort of going at randomness, but that sort of going at randomness, but that that randomness is a big piece of what that randomness is a big piece of what that randomness is a big piece of what creativity is. A good definition of creativity is. A good definition of creativity is. A good definition of creativity is simply being able to creativity is simply being able to creativity is simply being able to create things that didn't exist before. create things that didn't exist before. create things that didn't exist before. And that's a thing that LLM are pretty And that's a thing that LLM are pretty And that's a thing that LLM are pretty good at doing. Now, there's limits on good at doing. Now, there's limits on good at doing. Now, there's limits on their creativity, right? So if you asked their creativity, right? So if you asked their creativity, right? So if you asked it to create something truly unique that it to create something truly unique that it to create something truly unique that has never been seen before, of course it has never been seen before, of course it has never been seen before, of course it can't do it because it has no model for can't do it because it has no model for can't do it because it has no model for what that thing would be, but it would what that thing would be, but it would what that thing would be, but it would give you an approximation of it. And give you an approximation of it. And give you an approximation of it. And that is a creative act. Um, a really that is a creative act. Um, a really that is a creative act. Um, a really good example of how to see this in good example of how to see this in good example of how to see this in motion is just ask an LLM to build you motion is just ask an LLM to build you motion is just ask an LLM to build you an image of a topographical map. Just an image of a topographical map. Just an image of a topographical map. Just say, "Give me a topical topographical say, "Give me a topical topographical say, "Give me a topical topographical map of Washington DC." And watch it map of Washington DC." And watch it map of Washington DC." And watch it produce a perfectly reasonable produce a perfectly reasonable produce a perfectly reasonable topographical map that looks nothing topographical map that looks nothing topographical map that looks nothing like Washington DC. Right? That's not like Washington DC. Right? That's not like Washington DC. Right? That's not randomness. That's not Mad Libs. Like, randomness. That's not Mad Libs. Like, randomness. That's not Mad Libs. Like, it's got roads. It's got the little it's got roads. It's got the little it's got roads. It's got the little markers on it. It's got mountains and markers on it. It's got mountains and markers on it. It's got mountains and rivers. It's got all the things you rivers. It's got all the things you rivers. It's got all the things you need. Uh they're just completely need. Uh they're just completely need. Uh they're just completely misplaced because there's no misplaced because there's no misplaced because there's no deterministic thing you attached it to deterministic thing you attached it to deterministic thing you attached it to that was like, "Hey, what's the that was like, "Hey, what's the that was like, "Hey, what's the topography of Washington DC?" Does that topography of Washington DC?" Does that topography of Washington DC?" Does that make sense?

make sense? make sense? >> Yeah. So, I had a conversation about >> Yeah. So, I had a conversation about >> Yeah. So, I had a conversation about this this morning with someone and uh this this morning with someone and uh this this morning with someone and uh they were making a knowledge base and they were making a knowledge base and they were making a knowledge base and someone on the call said, "Oh, that's someone on the call said, "Oh, that's someone on the call said, "Oh, that's just rag. That's just retrieval just rag. That's just retrieval just rag. That's just retrieval augmented generation. And you know, it's augmented generation. And you know, it's augmented generation. And you know, it's never a good idea to say just never a good idea to say just never a good idea to say just >> just, >> just, >> just, >> that's just a bad thing to say. >> that's just a bad thing to say. >> that's just a bad thing to say. >> But they said, well, no. So they said, >> But they said, well, no. So they said, >> But they said, well, no. So they said, no, there's a difference between rag and no, there's a difference between rag and no, there's a difference between rag and being actually grounded in facts. being actually grounded in facts. being actually grounded in facts. >> Totally. >> Totally. >> Totally. >> Uh because an example they said is like >> Uh because an example they said is like >> Uh because an example they said is like you have two you have two you have two >> PDFs and you're going to do rag over >> PDFs and you're going to do rag over >> PDFs and you're going to do rag over them and they one PDF uh says a thing them and they one PDF uh says a thing them and they one PDF uh says a thing like 2 plus two is four and the other like 2 plus two is four and the other like 2 plus two is four and the other PDF says 2 plus two is seven, right? PDF says 2 plus two is seven, right? PDF says 2 plus two is seven, right? Well, at if you just did rag, you would Well, at if you just did rag, you would Well, at if you just did rag, you would not deterministically get the right not deterministically get the right not deterministically get the right answer. But if you grounded it in facts, answer. But if you grounded it in facts, answer. But if you grounded it in facts, the better tools you give the LLM, the the better tools you give the LLM, the the better tools you give the LLM, the more likely you'll be able to reconcile more likely you'll be able to reconcile more likely you'll be able to reconcile things like that, right? Why do we care things like that, right? Why do we care things like that, right? Why do we care whether how well an LLM does math whether how well an LLM does math whether how well an LLM does math without a calculator? without a calculator? without a calculator? It doesn't even make sense that you'd It doesn't even make sense that you'd It doesn't even make sense that you'd ask it to. It's going to ask it to. It's going to ask it to. It's going to probabilistically be wrong some amount probabilistically be wrong some amount probabilistically be wrong some amount of the time. It's never going to get to of the time. It's never going to get to of the time. It's never going to get to 100% correctness. If it got to 100% 100% correctness. If it got to 100% 100% correctness. If it got to 100% correctness, it's not an LLM anymore. I correctness, it's not an LLM anymore. I correctness, it's not an LLM anymore. I don't know what it would be, but it don't know what it would be, but it don't know what it would be, but it wouldn't be an LLM. And so, like, you wouldn't be an LLM. And so, like, you wouldn't be an LLM. And so, like, you know, we got to think about building know, we got to think about building know, we got to think about building those systems and exposing them in a way those systems and exposing them in a way those systems and exposing them in a way that allows it to do the work. It turns that allows it to do the work. It turns that allows it to do the work. It turns out that that is harder than it sounds out that that is harder than it sounds out that that is harder than it sounds because in a lot of our domains, we because in a lot of our domains, we because in a lot of our domains, we haven't really designed them to be used haven't really designed them to be used haven't really designed them to be used in this way, right? They're highly they in this way, right? They're highly they in this way, right? They're highly they have to be highly parallelized. You have have to be highly parallelized. You have have to be highly parallelized. You have to have designs that mimic the data to have designs that mimic the data to have designs that mimic the data model that's in the training. So for

model that's in the training. So for model that's in the training. So for infrastructure if I build an abstraction infrastructure if I build an abstraction infrastructure if I build an abstraction layer between like AWS or Azure and my layer between like AWS or Azure and my layer between like AWS or Azure and my tooling then I have to train the LLM on tooling then I have to train the LLM on tooling then I have to train the LLM on what the abstraction layer is and how what the abstraction layer is and how what the abstraction layer is and how the map works and it gets it wrong if I the map works and it gets it wrong if I the map works and it gets it wrong if I do it onetoone right if I the more I can do it onetoone right if I the more I can do it onetoone right if I the more I can make it just exactly what you say in make it just exactly what you say in make it just exactly what you say in your own documentation it gets it right your own documentation it gets it right your own documentation it gets it right almost all the time almost all the time almost all the time >> and you know but we didn't design those >> and you know but we didn't design those >> and you know but we didn't design those systems that way we started building systems that way we started building systems that way we started building abstractions where we were like oh it'd abstractions where we were like oh it'd abstractions where we were like oh it'd be much simpler if I could interact be much simpler if I could interact be much simpler if I could interact acts, you know, generically over all acts, you know, generically over all acts, you know, generically over all compute, you know, and it turns out compute, you know, and it turns out compute, you know, and it turns out those things are terrible. I need those things are terrible. I need those things are terrible. I need simulators and review because you can't simulators and review because you can't simulators and review because you can't trust them because they're going to get trust them because they're going to get trust them because they're going to get it wrong, you know, like all this stuff it wrong, you know, like all this stuff it wrong, you know, like all this stuff you need that suddenly, you know, you need that suddenly, you know, you need that suddenly, you know, there's some analog in the way we did it there's some analog in the way we did it there's some analog in the way we did it as people when we were building those as people when we were building those as people when we were building those systems originally, but they don't map systems originally, but they don't map systems originally, but they don't map onetoone anymore. onetoone anymore. onetoone anymore. >> Mhm. >> Mhm. >> Mhm. >> And you have to rebuild how you and >> And you have to rebuild how you and >> And you have to rebuild how you and rethink how all that stuff works in rethink how all that stuff works in rethink how all that stuff works in order to make a system that actually order to make a system that actually order to make a system that actually works well with AI. So we've gone from I works well with AI. So we've gone from I works well with AI. So we've gone from I had my first build system was in you had my first build system was in you had my first build system was in you know a B you know a make file and then know a B you know a make file and then know a B you know a make file and then it was XML and then it was JSON and then it was XML and then it was JSON and then it was XML and then it was JSON and then it was YAML and now it feels like uh and it was YAML and now it feels like uh and it was YAML and now it feels like uh and this is a an oversimplification but it's this is a an oversimplification but it's this is a an oversimplification but it's like people are writing markdown with like people are writing markdown with like people are writing markdown with emoji and calling it like their build emoji and calling it like their build emoji and calling it like their build script but you're pretty close to doing script but you're pretty close to doing script but you're pretty close to doing that.

that. that. >> You can get pretty close to doing that >> You can get pretty close to doing that >> You can get pretty close to doing that if you have to think of it as layers of if you have to think of it as layers of if you have to think of it as layers of translation. So, you know, you shouldn't translation. So, you know, you shouldn't translation. So, you know, you shouldn't write your build script in a markdown write your build script in a markdown write your build script in a markdown file and just run it and be like, LLM, file and just run it and be like, LLM, file and just run it and be like, LLM, build my software and pray, you know, build my software and pray, you know, build my software and pray, you know, and just keep saying like, never do this and just keep saying like, never do this and just keep saying like, never do this wrong, you know, please don't make a wrong, you know, please don't make a wrong, you know, please don't make a mistake, LLM, you know, like that's not mistake, LLM, you know, like that's not mistake, LLM, you know, like that's not going to work. It's going to make the going to work. It's going to make the going to work. It's going to make the mistake. Your system won't build. That mistake. Your system won't build. That mistake. Your system won't build. That will be frustrating. will be frustrating. will be frustrating. >> But could you teach the LLM that there's >> But could you teach the LLM that there's >> But could you teach the LLM that there's a tool that's called build system, and a tool that's called build system, and a tool that's called build system, and it knows how to call the build system to it knows how to call the build system to it knows how to call the build system to create an application artifact? create an application artifact? create an application artifact? Absolutely, you can. and you can pass Absolutely, you can. and you can pass Absolutely, you can. and you can pass that off and it's going to run that make that off and it's going to run that make that off and it's going to run that make file for you and it's going to do the file for you and it's going to do the file for you and it's going to do the deterministic thing. No one needs a deterministic thing. No one needs a deterministic thing. No one needs a non-deterministic build process. non-deterministic build process. non-deterministic build process. >> That's ridiculous. >> That's ridiculous. >> That's ridiculous. >> There we go. >> There we go. >> There we go. >> But what you do need is a translation >> But what you do need is a translation >> But what you do need is a translation from plain English which people can say. from plain English which people can say. from plain English which people can say. How how many people can write the make How how many people can write the make How how many people can write the make file? Not very many. How many people can file? Not very many. How many people can file? Not very many. How many people can say please build the software please say please build the software please say please build the software please please? Everyone. please? Everyone. please? Everyone. >> Yes. Right. So you're making a pros >> Yes. Right. So you're making a pros >> Yes. Right. So you're making a pros compiler.

compiler. compiler. >> Yes. >> Yes. >> Yes. >> And it's a DSL because the DSL is pros. >> And it's a DSL because the DSL is pros. >> And it's a DSL because the DSL is pros. >> Yes. Yes. And when you think about >> Yes. Yes. And when you think about >> Yes. Yes. And when you think about infrastructure, that's the kind of infrastructure, that's the kind of infrastructure, that's the kind of semantic layer that we've always wanted. semantic layer that we've always wanted. semantic layer that we've always wanted. What you've always wanted was to be able What you've always wanted was to be able What you've always wanted was to be able to go have a co-orker say, "Hey, we got to go have a co-orker say, "Hey, we got to go have a co-orker say, "Hey, we got to redeploy that software right now and to redeploy that software right now and to redeploy that software right now and be able to take that sentence, hit be able to take that sentence, hit be able to take that sentence, hit enter, and have the system figure out enter, and have the system figure out enter, and have the system figure out what he meant." All this stuff we've what he meant." All this stuff we've what he meant." All this stuff we've been talking about about how systems been talking about about how systems been talking about about how systems should be declarative and what we want, should be declarative and what we want, should be declarative and what we want, like what we've always wanted was that like what we've always wanted was that like what we've always wanted was that semantic layer of intent, that ability semantic layer of intent, that ability semantic layer of intent, that ability for somebody to say, "Hey, my intent was for somebody to say, "Hey, my intent was for somebody to say, "Hey, my intent was this." And the LLM is great at taking this." And the LLM is great at taking this." And the LLM is great at taking that intent, adding in the in the data that intent, adding in the in the data that intent, adding in the in the data it has about the system, what it has about the system, what it has about the system, what applications are deployed, what's the applications are deployed, what's the applications are deployed, what's the infrastructure look like, and it will infrastructure look like, and it will infrastructure look like, and it will use tools to go answer those questions use tools to go answer those questions use tools to go answer those questions and then it will figure out how to and then it will figure out how to and then it will figure out how to deploy the app, right? And that is deploy the app, right? And that is deploy the app, right? And that is great, right? That's so helpful because great, right? That's so helpful because great, right? That's so helpful because tons of us can write that single tons of us can write that single tons of us can write that single sentence. Tons of us can do that. And sentence. Tons of us can do that. And sentence. Tons of us can do that. And most of us could also debug it when it most of us could also debug it when it most of us could also debug it when it breaks because when you have a mostly breaks because when you have a mostly breaks because when you have a mostly working system, it's not so hard to see working system, it's not so hard to see working system, it's not so hard to see where the things fall apart. And that where the things fall apart. And that where the things fall apart. And that loop also gets better, right? You can loop also gets better, right? You can loop also gets better, right? You can ask the LLM to do it. Like it'll go fix ask the LLM to do it. Like it'll go fix ask the LLM to do it. Like it'll go fix production outages for you. It'll find, production outages for you. It'll find, production outages for you. It'll find, you know, that are annoying. Like we had you know, that are annoying. Like we had you know, that are annoying. Like we had an outage that was caused by a a an outage that was caused by a a an outage that was caused by a a configuration parameter that was configuration parameter that was configuration parameter that was basically off by one in how the in the basically off by one in how the in the basically off by one in how the in the way that they had written it down. And way that they had written it down. And way that they had written it down. And the LLM found it in a couple minutes of the LLM found it in a couple minutes of the LLM found it in a couple minutes of just digging around being like, "Oh, just digging around being like, "Oh, just digging around being like, "Oh, well, here's the error message. Oh, why well, here's the error message. Oh, why well, here's the error message. Oh, why would that be?" And then it went and would that be?" And then it went and would that be?" And then it went and like found the bash script that used

like found the bash script that used like found the bash script that used that variable, saw that the variable was that variable, saw that the variable was that variable, saw that the variable was named one thing, looked at the named one thing, looked at the named one thing, looked at the configuration parameter on the other configuration parameter on the other configuration parameter on the other side, was like, "Oh, that one's wrong. side, was like, "Oh, that one's wrong. side, was like, "Oh, that one's wrong. It's off by one character, and then told It's off by one character, and then told It's off by one character, and then told us what the problem was." That was sick. us what the problem was." That was sick. us what the problem was." That was sick. Like, we'd have found it, you know, like Like, we'd have found it, you know, like Like, we'd have found it, you know, like we wrote the infrastructure. we wrote the infrastructure. we wrote the infrastructure. >> You would have found it. It just would >> You would have found it. It just would >> You would have found it. It just would have taken longer. have taken longer. have taken longer. >> Well, yeah, it would have taken longer. >> Well, yeah, it would have taken longer. >> Well, yeah, it would have taken longer. And we found it by saying, "This thing's And we found it by saying, "This thing's And we found it by saying, "This thing's not working. and here's the error not working. and here's the error not working. and here's the error message blah blah you know we had message blah blah you know we had message blah blah you know we had somebody the other day who drew a somebody the other day who drew a somebody the other day who drew a picture of the architecture that they picture of the architecture that they picture of the architecture that they had deployed but had not built any had deployed but had not built any had deployed but had not built any automation around and then said this is automation around and then said this is automation around and then said this is what I want put in system initiative so what I want put in system initiative so what I want put in system initiative so I can automate it and we said great send I can automate it and we said great send I can automate it and we said great send that picture to the LM and tell it to that picture to the LM and tell it to that picture to the LM and tell it to discover the infrastructure and he did discover the infrastructure and he did discover the infrastructure and he did and it did and it did and it did >> you know and like that is incredible >> you know and like that is incredible >> you know and like that is incredible right and that's because we put it right and that's because we put it right and that's because we put it against a deterministic system under the against a deterministic system under the against a deterministic system under the hood that knows how to discover hood that knows how to discover hood that knows how to discover infrastructure. It knows how to look at infrastructure. It knows how to look at infrastructure. It knows how to look at classes of things and find them all. You classes of things and find them all. You classes of things and find them all. You know, know, know, >> it's a rich tool calling library. >> it's a rich tool calling library. >> it's a rich tool calling library. >> Yeah. Yes. So, so here's a question >> Yeah. Yes. So, so here's a question >> Yeah. Yes. So, so here's a question then. This might be a little then. This might be a little then. This might be a little philosophical. Let me see if I can get philosophical. Let me see if I can get philosophical. Let me see if I can get this out correctly.

this out correctly. this out correctly. >> If one makes a DSL, a domain specific >> If one makes a DSL, a domain specific >> If one makes a DSL, a domain specific language and they compile that into an language and they compile that into an language and they compile that into an intermediate language and then they run intermediate language and then they run intermediate language and then they run that through an a runtime. That's a that through an a runtime. That's a that through an a runtime. That's a that's a pattern that has existed for that's a pattern that has existed for that's a pattern that has existed for years. years. years. >> Yeah. >> Yeah. >> Yeah. >> The DSL though is always the >> The DSL though is always the >> The DSL though is always the authoritative source. So if you do a authoritative source. So if you do a authoritative source. So if you do a clean and delete everything, you go and clean and delete everything, you go and clean and delete everything, you go and you point at the original DSL and you you point at the original DSL and you you point at the original DSL and you say, I can always re I can always say, I can always re I can always say, I can always re I can always recreate everything. recreate everything. recreate everything. >> But if you write a bunch of pros >> But if you write a bunch of pros >> But if you write a bunch of pros >> and generate that intermediate thing, >> and generate that intermediate thing, >> and generate that intermediate thing, that intermediate thing now is that intermediate thing now is that intermediate thing now is deterministic by definition. deterministic by definition. deterministic by definition. >> Yes. >> Yes. >> Yes. >> But the pros is >> But the pros is >> But the pros is >> um is not. But if I deleted the >> um is not. But if I deleted the >> um is not. But if I deleted the intermediate thing, I could never intermediate thing, I could never intermediate thing, I could never recreate it exactly. I would get 99% or recreate it exactly. I would get 99% or recreate it exactly. I would get 99% or you know, whatever the temperature is, you know, whatever the temperature is, you know, whatever the temperature is, right? I'd get a nondeterministic right? I'd get a nondeterministic right? I'd get a nondeterministic version. version. version. >> Yeah, you get some kind of >> Yeah, you get some kind of >> Yeah, you get some kind of approximation. approximation. approximation. >> Right. So arguably now the exist the >> Right. So arguably now the exist the >> Right. So arguably now the exist the difference between prosbased LLM DSLs is difference between prosbased LLM DSLs is difference between prosbased LLM DSLs is that the intermediate thing that is now that the intermediate thing that is now that the intermediate thing that is now deterministic could be fed back into the deterministic could be fed back into the deterministic could be fed back into the system so that it might be even more system so that it might be even more system so that it might be even more deterministic. deterministic. deterministic. >> That's absolutely what happens like now >> That's absolutely what happens like now >> That's absolutely what happens like now and now you have designed system and now you have designed system and now you have designed system initiative. So what you do is I'm initiative. So what you do is I'm initiative. So what you do is I'm interviewing Will for you now.

interviewing Will for you now. interviewing Will for you now. >> You are you're doing a great job. Yeah. >> You are you're doing a great job. Yeah. >> You are you're doing a great job. Yeah. No. No. It moves from being source code, No. No. It moves from being source code, No. No. It moves from being source code, right, to being data. right, to being data. right, to being data. >> Yes. >> Yes. >> Yes. >> And once what you're doing is saying, I >> And once what you're doing is saying, I >> And once what you're doing is saying, I have this rich data model. And what I have this rich data model. And what I have this rich data model. And what I can do now is program the data model can do now is program the data model can do now is program the data model >> instead of storing it in text files and >> instead of storing it in text files and >> instead of storing it in text files and then compiling it and then having it then compiling it and then having it then compiling it and then having it with infrastructure. Instead, it's with infrastructure. Instead, it's with infrastructure. Instead, it's like, nope, I'm going to write a text like, nope, I'm going to write a text like, nope, I'm going to write a text file, right? Instead, what I'm going to file, right? Instead, what I'm going to file, right? Instead, what I'm going to do is send it to System Initiative. do is send it to System Initiative. do is send it to System Initiative. System Initiative updates a data model, System Initiative updates a data model, System Initiative updates a data model, and then that data model can go and then that data model can go and then that data model can go birectionally out into the world. So it birectionally out into the world. So it birectionally out into the world. So it can like you can use the data model to can like you can use the data model to can like you can use the data model to change real world resources, you can use change real world resources, you can use change real world resources, you can use it to generate templates, you can use it it to generate templates, you can use it it to generate templates, you can use it to uh to duplicate things. So like if to uh to duplicate things. So like if to uh to duplicate things. So like if you have an application and you want to you have an application and you want to you have an application and you want to deploy it from one region to another, deploy it from one region to another, deploy it from one region to another, you know, you're just like, hey, I have you know, you're just like, hey, I have you know, you're just like, hey, I have it running in in California. I want it it running in in California. I want it it running in in California. I want it to run on the east coast. All you have to run on the east coast. All you have to run on the east coast. All you have to do in system initiative is ask the to do in system initiative is ask the to do in system initiative is ask the system to make a duplicate of it and system to make a duplicate of it and system to make a duplicate of it and then set the region to the east coast then set the region to the east coast then set the region to the east coast and then you can hit apply and it will and then you can hit apply and it will and then you can hit apply and it will just redeploy it because it'll just just redeploy it because it'll just just redeploy it because it'll just rebuild it from all the same data rebuild it from all the same data rebuild it from all the same data because you have this structured because you have this structured because you have this structured understanding of exactly what the understanding of exactly what the understanding of exactly what the infrastructure is, its relationships to infrastructure is, its relationships to infrastructure is, its relationships to other things, what makes it valid or other things, what makes it valid or other things, what makes it valid or invalid, its documentation, your own invalid, its documentation, your own invalid, its documentation, your own corporate policy, all that stuff just corporate policy, all that stuff just corporate policy, all that stuff just goes into the data model and then you goes into the data model and then you goes into the data model and then you use that model to feed the loop both to use that model to feed the loop both to use that model to feed the loop both to people and to LLMs.

people and to LLMs. people and to LLMs. Fantastic. Oh my goodness, this has been Fantastic. Oh my goodness, this has been Fantastic. Oh my goodness, this has been delightful. Uh, I'm learning a lot. Uh, delightful. Uh, I'm learning a lot. Uh, delightful. Uh, I'm learning a lot. Uh, I need to go and explore the source I need to go and explore the source I need to go and explore the source code. I need to check out the SDKs, the code. I need to check out the SDKs, the code. I need to check out the SDKs, the APIs. I can find all of this starting at APIs. I can find all of this starting at APIs. I can find all of this starting at systeminit.com. systeminit.com. systeminit.com. >> There's a whole rich documentation at >> There's a whole rich documentation at >> There's a whole rich documentation at docs.systeminit.com >> and uh it's all uh it's all open source >> and uh it's all uh it's all open source and it's very good, very uh friendly and it's very good, very uh friendly and it's very good, very uh friendly pricing, you know, for certainly for pricing, you know, for certainly for pricing, you know, for certainly for hobbyists and people like me. I could hobbyists and people like me. I could hobbyists and people like me. I could get started for for for nothing, but get started for for for nothing, but get started for for for nothing, but even even the hourly per resource is is even even the hourly per resource is is even even the hourly per resource is is quite reasonable. quite reasonable. quite reasonable. >> Hey, thanks. Yeah, I'm I mean you don't >> Hey, thanks. Yeah, I'm I mean you don't >> Hey, thanks. Yeah, I'm I mean you don't have to run any Yeah, the free tier is have to run any Yeah, the free tier is have to run any Yeah, the free tier is 100 resources um which is more than 100 resources um which is more than 100 resources um which is more than enough to run a small production enough to run a small production enough to run a small production infrastructure. infrastructure. infrastructure. >> Fantastic. Well, thank you so much Adam >> Fantastic. Well, thank you so much Adam >> Fantastic. Well, thank you so much Adam Jacob for chatting with me today. Jacob for chatting with me today. Jacob for chatting with me today. >> Yeah, thanks for talking with me. >> Yeah, thanks for talking with me. >> Yeah, thanks for talking with me. >> All right, this has been another episode >> All right, this has been another episode >> All right, this has been another episode of Hansel Minutes and we'll see you of Hansel Minutes and we'll see you of Hansel Minutes and we'll see you again next week. again next week. again next week. [Music]