Transcript
Hello and welcome to this very special episode of I Owe You an Explanation. I am Dan Cerulli, Vice President and General Manager of Cloud Native at Nutanix. And you might notice that the audio is a little bit different than you're used to. That's because we are joining you here from the floor of .NEXT. And we'll be talking about a lot of the stuff that's going on here. It's loud, it's high energy, and it's a perfect place to talk about how things actually get built. Our goal, as always, is to cut through the noise and deliver a signal you can actually use. And my guest today is someone I had the pleasure of working very closely with back during our days at Google, Kelsey Hightower. At the time, he was a developer advocate and worked his way up to distinguished engineer. And he's one of the people who really helped increase the prominence of many of the projects that we talk about in open source. Some people call him the explainer in chief. Today, we're going to be cutting through the noise and explaining all about open source. Kelsey, welcome to the show. Happy to be here. Now, before we get into the meat of the conversation, we like to do something we call the before times. And the before times is we'll take a topic and something that's changed our lives, sometimes professionally, sometimes personally. And I want to talk about video games. I want to talk about before you could game online, before you could get a CD-ROM with a video game. How did we play video games back then, Kelsey? NES in the 80s. I remember there was this thing where you would get a Nintendo cartridge and you were blowing it as if that would make the game play better because it started to glitch out. So back then, we would put the cartridge in, you push down. Some games you could save, you can hold the power button and the reset button at the same time and flash the save state. But yeah, NES, two buttons on the controller, all you needed to do what you had to do. It's funny because I can go back a generation before. Like I'm a 2600 kid. We had the original Atari 2600, one button on the joystick, no save game state ever. And one fun thing was that, of course, you only had the games you owned, right? And if you're going to a buddy's house and he didn't have the game, you would bring your cartridges with you, right? You would like take your cartridges, you'd go to your friend's house. Like, here we go. Oh, you got Missile Command. Awesome. It was pretty cool. I remember that was the first time I realized that some technology just doesn't work. I remember buying the Power Glove. Oh, yeah. And the Power Glove, you watch the commercials, like, you can play Mike Tyson's Punch-Out like a real boxer. And I'm looking at this like, this is going to change everything. Once I get that Power Glove, I can finally beat that last stage Mike Tyson. I remember putting it on and this thing is just, and you read the instructions, you got to turn it and use the controller that's built into the glove. I'm like, this is completely pointless. And that might be my first experience with over-engineering. And marketing. Oh, 100%. Right? The lies of marketing. 100%. Sorry to my marketing friends. There might be some here. All right. Well, that's the before times. Now, let's come back to the present. So, we got to know each other in your time at Google. You got into open source before then. But I think that while open source has been around really since the 90s. My first boss was a prominent member of the Open Computing Foundation, a big contributor to Berkeley DB. But open source really took off. I think probably Linux is the first thing that got really widespread. And then as some of the projects you got involved in, open source really dramatically changed the way enterprises work. Why do you think that in a world where I grew up, which is the code you write is the biggest secret your company has, how do we change to a model where people are willing to take that code and put it out for anyone to use? I think for me, it was all about access. So, I started my career right after high school in 1999. And at that point, most companies are running AIX, big IBM mainframes, Sun Microsystems. You couldn't even afford this kind of thing to have in a home lab. I don't think anyone I knew had a home lab. No, no, no. And I remember going to a place like Micro Center and buying one of those Linux magazines. And on the back of it was like 800 free pieces of software. This thing called GNU Linux, because it's not just Linux. It's like you got to have that GNU user land. And I'm asking myself, why would anyone write software for free and just give it away? And I remember in that time period, I'm reading some of the history around people like Richard Stallman that thought software shouldn't be trapped in type of enterprise licenses. And that created access for me. So, for me, getting access to things like Linux and open source implementations, FreeBSD, it brought people like me in who otherwise I think would have been kept out because of the cost. And when I got in, that became the way I learned about the IT industry through open source. And my entire generation, that's what I think we brought to the workspace. Super interesting to think about that. It's like the democratization of software. And I hadn't really thought about that. Yeah, but absolutely. And you're right. I mean, there are certainly those people like Richard Stallman who thought software should always be free. Free not just in the economic sense, but free in the don't fence it in. Let it be out there. Let people change it. So, you started using Linux that way. You started learning about the whole ecosystem around it. When did you start contributing? And how did that come about? I went to my first tech conference in, man, probably 2006-ish. And I started to learn Python at work. I'm getting to learning how to write code. Oh, you're doing it at the time. Working in financial services. We were moving things from mainframe, COBOL, packed decimal, fixed length fields. And I remember Python was really good for data science work. We had really good libraries. And I was like, look, I can probably automate some of this just using Python. And so, I remember going to my first Python meetup. And then I went to my first PyCon in Atlanta. And my experience with open source until that time was you download it for free. It's the thing that costs zero. And seems to be something that other people use. So, why not use it too? But when I got to PyCon, people like Guido van Rossum is there. These are the people who created the language. All the people supporting all the tools, utilities, package management, PyPy, all the infrastructure. They were all at the conference. And that's when I understood the sense of community. There's a part of this where they're showing the newcomers like me how to contribute. My first set of contributions were to VirtualIn and some of the package management tools back then. I ran into a bug at work. And it turns out you can just fix it. And so, I made my first set of contributions. Joined the community. We had a hackathon after the conference. They stayed an extra two days. And you would just sit in this room with all these core contributors. And they would try to help you become one. Speaking of the before time, so this is before GitHub, right? This is probably before. So, where was that source code? I think you used to push it to Mercurial Repository somewhere. There was an endpoint. You made all your changes. You pushed them. There was a web page where people can come and see releases and download things. So, yeah, this is before we had like the universal agreed upon place to go. You joined that community, and they kind of told you where the repository was. Right. Interesting. Interesting. So, we met after you. And I know you went through some other stops. But you got into Google, right? And while you were at Google, or no, you got into Kubernetes before you got into Google. What order did that happen? I was at CoreOS. We were working on Google's infrastructure for everyone else. We took the RAF white paper, combined it with what they learned from Chubby, and we produced things like etcd. Of course, CoreOS, this operating system designed around containers. And so, at CoreOS, I'm like, as a person who was leaving configuration management and DevOps, I was an engineer at Puppet Labs. And I saw the transition to Golang, the Go programming language, and the transition to containers. So, this is like 2012-ish. So, now I'm at CoreOS, and now we're like, hey, we're going to build Google's infrastructure for everyone else. Right. And so, we're building this Docker thing, and we started to have this own orchestrator called Fleet. Right. And then one day, like the rest of the world, I look up, Google's announcing Kubernetes to the world in a Docker kind of like, well, what's that? And they gave us early access to the repository. And so, I downloaded all the bits. There's no documentation. Of course not. It assumed you were probably going to run on Google Cloud. Of course. You were to kube up that SH script. You ran it. It did everything. But I had to get it running on CoreOS, on bare metal, or VMs. So, I spent that whole night just learning the repository, fixing knick-knack bugs here and there, writing up my findings. And as Google kind of announced the project, you know, hey, there's a press release. There's this thing called Kubernetes. We're releasing a tutorial on how to run Kubernetes on CoreOS. And so, for me, I guess I was one of the first people to introduce hands-on Kubernetes to the world. Right. Yeah. So, I'm making my contributions. I'm doing this for about a year and a half at CoreOS. And I make a decision to go to NASA. Google swoops in, make an enticing offer. Then I find myself at Google around 2015. So, at Google, you, of course, you were a contributor, but you somehow transitioned to another role too, right? As I said, I'm kind of the explainer-in-chief. And I want to talk about adoption of open source because we have seen these things. We saw this with Linux before become effectively the de facto standard in the data center, right? There are still people who use other operating systems, but it's mostly Linux. And then we saw there were other container orchestrators. You talked about Fleet. There was Docker Swarm. There was Mesos. Kubernetes became the standard. Tell me about the role that you don't have. It can be you, but in general, the community plays in taking open source and kind of explaining it to the world and saying, hey, here's how you do this. Here's why you want to use it. I think people don't realize is that any community has culture. And so, even though we think about big brands setting the tone, it actually doesn't really work that way. I remember the first time I saw like a RunDMC video and they're wearing Adidas. Adidas was not on my wish list at all, but you're watching RunDMC walk down the street with Adidas on. And then the thing that's on your Christmas list is like, I need a pair of Adidas. I got to have it. Adidas. There you go. Now we owe royalties. It was less than 10 seconds. And I think Kubernetes became a cultural element. When you start explaining this thing, what it can do, how you're doing it, that became a way to represent all the things we learned bundled into this piece of software we call Kubernetes. So that was just a name. But in the early days, it didn't have all these features. It didn't encapsulate everyone's experience about how to do distributed systems. And so when I was on the road, especially before joining Google, you would build something. But the other part of that was to show people what you built. That's the equivalent of like making a new sports coat and then walking around with it. If you don't walk around with it, no one's ever going to see it. All those keynotes, all those workshops, you were spreading the culture way faster than it would have ever done organically. In many ways, those keynotes were the music videos. I'm coming on stage saying, hey, all my puppet friends doing configuration management, here's another way to look at it. And I will put my proverbial shoes on, and they would say, yo, what are those? Is it the shoes that's making you good at this? And so you would show other people how they would actually use these tools to go. And I think that brought other people to the product. The other thing that we did, KubeCon, two of us, Patrick Riley, Joseph Jacks, were in another conference, and we were like, hey, Kubernetes deserves to have a community just like PyCon, just like the Golang community. We want something as secure. So three people. I wasn't working at Google yet. These aren't the big companies. There's no CNCF yet. The Links Foundation is not quite involved. And we built something for the community, and that legacy carries on. That's pretty amazing. I didn't know that aspect of it. That's pretty cool. So I think you're right. I think one of the beautiful things about open source, one of the powerful things about open source, is that community. And you know as well as I do, I have talked to countless people who do think like, hey, it was Kelsey's keynotes that brought me in, maybe his YouTube videos. So I get why that community gets people wanting to use it. You can play with it. You can tinker with it. What do you think made it so really big companies would feel, banks, insurance companies, tire companies, would feel comfortable using this thing in open source? If there's not a big company, you know, it used to be you would buy software from somebody, and you were paying them money, and they were saying, I guarantee you it's going to work. Why do these companies feel safe using open source when it isn't just a big company? It's not just open source. It's not just open source. I think some people think about like this free software, why would anyone install it and offer it as a service? That's not the thing that does it. Trends become products. If the wedding industry sees someone getting married in lime green wedding dresses, and that becomes a thing that's on the front page of every magazine, I guarantee you, every bridal shop in the world will carry lime green wedding dresses. And it's that component. When Kubernetes got popular, it got popular because people were using it. Hey, I had all these problems before Kubernetes. Here's how I'm solving them. And so if you're a big cloud vendor, you're like, hey, that's a Google thing. They can run that. And Google alone can't make everyone else do this. Something else has to happen. And then that culture spreads. You know, one of the co-founders, Brandon Burns, goes to Microsoft, and Microsoft is hedging their bets. Hey, we're going to do Docker Swarm. We're going to do Mesos, the place you used to hang out and do your thing. And we're going to probably bet on Kubernetes. And at some point, you just see the culture. No. We see the signs on the wall. We're going to go all in on Kubernetes. And if you're Amazon, you're looking at KubeCon go from 500 people to 10,000 people. And that's not enough. Then the customer says, hey, I just went to KubeCon, and I'm seeing how people are actually solving problems. I tried the open source version, and I would like a managed service from you. And I think that's the thing that completes the puzzle. So I think it was like it was a ground-up movement for sure. And vendors like Nutanix, of course, you look at all of this, and you say, hey, wait a minute. If this is what people are really doing, and since we don't even have a product, they're doing it themselves. This is probably going to stick. Can we make it easier for them? Right, right. So you mentioned the CNCF before, the Cloud Native Computing Foundation. One last thing, I think, what role does that organization play? So we know Kubernetes, it's an open source project, right? What is the CNCF, what role did it play in helping this whole environment to grow and this ecosystem to succeed? I think they handled the sustainability part. Because the truth is a lot of this open source work is built by volunteers, practitioners that will eventually move on. And as those core maintainers cycle out to something else, some of them, like Brendan Burns is like a senior VP at Microsoft at this point. I'm not sure Kubernetes is the number one thing on his mind, even if he still cares about the project. Right. Tim Hawkin, on the other hand. Tim Hawkin, he's still all in. Chopping wood, carrying water. But the project is much bigger than even the core people can ever handle. And so I think the CNCF does a good job of saying, hey, let's figure out a governance model, number one. Let's figure out the legal pieces around IP. So if you are a Nutanix and you want to carry that Kubernetes logo, you don't have to worry about getting sued for IP infringement. So once you take care of the legal component, then it kind of gives a little air cover to start to contribute. The other thing I think they do well is like give these projects airtime. Right. So people think, oh, KubeCon is just a conference. It's like, no, it's a reunion. A lot of maintainers that work apart online only get to see each other in person once or twice a year. And so during those reunions, I think a lot of people don't realize what happens. Those maintainer summits, those SIG meetings where we try to define, hey, how do we handle this new workload? AI is now on the scene. Can Kubernetes evolve to handle that particular thing? So lots of discussion happens in person. And then as a project maintainer myself at one point, being on stage to show the world your work, getting feedback from people that had a good experience with your work is very motivating. And so the fact that the Linux Foundation can afford to have venues this big to house 5, 10, 15,000 people at a time, and the number of people that is the first time being on site, 30% of the attendees always is their first time. Always, yeah, yeah, yeah. And so you just continue growing that community, and they have the ability to do this across the world, not just the United States, not just one country in Europe. They make this thing global. So without those foundations, I don't know if we would have this ability. No, I think you're right. I think it made it safe to contribute, made it safe to consume, and that revolutionized the industry. All right. Well, thank you very much. I think we learned a lot about open source. I hope everybody else did too. We're going to close with a segment we call Signal or Noise. I'm going to give you a phrase, a phrase we're hearing a lot of. And you tell me, is this signal, is this something we need to pay attention, or is this just noise? Are you ready? Your phrase is, the future is agentic. Noise. Noise. Agentic. Coming out hot against agentic. Tell me why it's noise, because I think this is very interesting. So when people say agentic, you ask them what do they mean. Most people have a hard time explaining it. And then you ask them what they want to do with agentic. What is the promise? What are they hoping for? So now we start to get closer to the signal. I want to automate a process that either I ignored before and just did it manually, or I lack the tools to envision how to do it. And let's maybe have people with a use case. If I asked you to be able to support any handwritten receipts and then turn that into a database entry, that's really hard to do with just a Bash script. Right. That's really hard to do with any off-the-shelf programming language. Now, some of you are really wise. Well, Kelsey, you could use a library that can parse handwriting and then turn that into entries. And then you end up with a bunch of if statements. And so now there's a bit of a new paradigm we have. We have this new component, that generative AI component where we have a model that can take handwriting or letter recognition to another level. And then it's programmable. That's like the nuance. So we have this new, let's call it library or system call, if you will. Right. So you take that system call and you think about the process. We want to take handwritten receipts and put them in a database. If you have a receipt you already understand, you'd never need a generative AI to do that. Right. You could have taken that receipt, parsed that once, turned that into a library, and you could have been putting receipts in a database for the last 10, 20 years. Now, the concept of using a new library or framework to parse even an unstructured receipt and put it in a database, well, that's a net new component. Right. But the rest of the pipeline looks exactly the same. So when people say a genetic AI, I'm like, stop it. You will be wasting tokens if you were to try to infer how to put a receipt into the database. Once you have the structure, okay, maybe you do a little bit of inference for that component. Right. But once you have that, now you can just use classic Python or the Oracle database driver and just insert it into the database. And I call that zero-token architecture. And the reason why I call it that is because I'm telling people that a lot of what people think about in terms of agentic and the value of automating any task, you were doing that for a very long time with your con jobs and your bash scripts, and even Kubernetes to a degree represents a model of infrastructure that we've given people the tools to automate. So I think it gets overblown a little bit to describe workflow automation as a term I think people are looking for. I think it's an interesting and really, really fantastic viewpoint. I can already tell you that zero-token architecture is Signal, and we're going to hear a lot of it. So I'd like to thank you for spending time with us today. I would like to thank you for joining us. As always, I have to say, please subscribe, so please do subscribe. We've got a bunch of episodes coming. Really excited about what's here. Kelsey, thanks for being here. Thanks for having me.
