Invisible Infrastructure, Autonomous Agents, and 10 Years of Cloud Native with CNCF's Chris Aniszczyk
Intro Music & Soundbyte / Promo Clip
“This whole notion of, you know, what I kind of call is like ephemeral evidence destruction is going to be a thing, right? Agents will do something, hide their tracks, like any well written potentially malware out there will try to do something. But agents are even smarter because they could do it on the fly. That’s kind of what’s on my mind. Agents can be naughty, and evidence can be destroyed very easily in this new world.”
CRob (00:24)
Welcome, welcome, welcome to What's in the SOSS, the OpenSSF podcast where I talk to developers, engineers, architects, executives about this amazing space we call open source security. Today, we have an incredible treat. We have a friend of the show, a dear partner of ours, Chris Aniszczyk from the Cloud Native Computing Foundation.
And we're here to talk a little bit about all the amazing things that are going on over there. So Chris, welcome to the show.
Chris Aniszczyk (00:54)
Awesome, glad to be here. It’s exciting
CRob (00:57)
So for those of our audience that may not be familiar with you, could you maybe give us a real quick TLDR of what your open source origin story is? How did you get into this space?
Chris Aniszczyk (01:08)
Yeah, sure. It's probably been a little over 25 years or so. I started out as a simple, um, Slackware user back in a day, if you remember…
CRob (01:18)
Nice.
Chris Aniszczyk (01:18)
..Good old Slackware. so it kind of played around with Slackware, uh, BSD and ports eventually got involved with Gentoo and became a contributor and maintainer there and maintain the Java and accessibility, uh, packages as part of Portage back in the day. And kind of, that's kind of how I got started with my You know, it's a Linux journey. I found the community so super helpful.
Like, you know, I remember vividly, you know, a memory where I couldn't get my my freaking, it was like ALSA sound drivers working and, I jump on IRC and, you know, just people all over the world randomly, like willing to help and pitch me out with like, with no, no expectation return, you know, that kind of like, you know, give freely, you know, help people other philosophy is something that I loved and kind of have been paying it forward since and obviously that pulled me into kind of open source contribution and the benefits there and kind of went all over the place after that.
You know, worked a bit on Eclipse back in the day for people that remember the developer IDE, wrote a lot of the tools involved with source control and also ran the plugin development team responsible for the plugin ecosystem. you know, there and had an open source startup around Eclipse developer tooling.
So I kind of all over, you know, the place and eventually, you know, after having a fun experience building out the open source efforts that an organization called Twitter, you know, at the time, which was really running and scaling workloads at a, at a crazy, you know, way using some kind of modern technology called something called containers, which didn't really exist at the time.
I had an opportunity to go build out the Cloud Native Computing Foundation at the Linux Foundation along with a sister organization called the Open Container Initiative, which is just as important in my opinion, which people don't talk about, but really standardized how containers are packaged, built and distributed across multiple operating systems and environments. So yeah, I kind of have like a random Linux user to developer, to contributor, to maintainer, to… open source startup founder to OSPO lead at Twitter to foundation, CTO / lead now. So been all over the place. Yeah.
CRob (03:37)
That's a pretty amazing journey. Yeah, that's amazing. And again, very deep roots in this community. It's amazing. Speaking of deep roots, you just touched on it, the CNCF. You all just passed a pretty major milestone, 10th anniversary, right?
Chris Aniszczyk (03:54)
Yeah, 10 years, which is probably like 40 years in open source here. It's a long time. It's a long time. So.
CRob (04:01)
It's crazy. like the last one, we did our fact checking and it might've changed. It changes daily. But at the time you all had 230 projects. You have over 300,000 contributors and over 190 countries around the world. So thinking back 10 years ago, it's 2015. We're in the way back machine.
Reflecting on kind of where you are today, thinking about today's ecosystem, what would have surprised 2015, Chris, as you're starting off on the CNCF journey?
Chris Aniszczyk (04:36)
2015, Chris was like, Kubernetes was not even accepted at the time. We're like trying to form our technical operating committee, which is our technical board of directors trying to get that all bootstrap and like praying that they would vote Kubernetes in. the Kubernetes project itself, you know, I think, you know, they did some interesting things that I think really set the course and kind of surprised me.
CRob (05:03)
Mm-hmm.
Chris Aniszczyk (05:04)
To this day, you know, one, they set really good kind of values from day one, right? They have this thing of like, a community over company, right? And kind of all about doing what's best interest for the project. And so there's kind of values like welcoming and inclusive. They kind of set these things very, very early. And I think at the time, I maybe have not put as much value on that.
But today I see how those early decisions of very much focusing on the culture of the project, making it welcome, inclusive and global led to the kind of success here where like, is so dominant across the world. Every major geography cloud provider kind of supports. I think a lot of that stretches to those original values that the project set.
So I kind of really...has at least shocked me of how like small decisions like that in the beginning could percolate throughout. And honestly, a lot of people try to compare Kubernetes and Linux together. And I do think there is some comparisons to kind of draw there where, I think in the early days, when I was using Linux, like you were back in the day, like it was a...
kind of more of a hobbyist thing, something you like to kind of play with, not serious business. And essentially what happened is like with Linux, a lot of people kind of got together and, you know, stretched in interesting ways. Like, hey, I want to put this in a phone. I want to put this in a car. I want to put this on a satellite, right? I want to go run this on servers, right? Same type of thing happened to Kubernetes where people, know, you know, originally Kubernetes started out to only really run on big data center servers, right? But now people have put Kubernetes in
cars and satellites.
And I think that culture of being open, know, inclusive, welcoming to crazy ideas and only stretch things, which Linux had Kubernetes also had really helped that project. And it kind of, you know, in, in 10 years, it, it's not a lot of time in, in, general, but like it went from almost no, like being nowhere to literally everything we interface and touch today. You call an Uber, you order something on Amazon, you do anything on it, book a flight, it's all touching Kubernetes and Linux for the most part underneath.
CRob (07:35)
I'm constantly amazed about the reach of our communities and the fact that they are so embedded in what we do every day and you don't even realize it. It's crazy.
Chris Aniszczyk (07:47)
Yeah, it's invisible infrastructure at this point, which is amazing. it's truly crazy to see how it feels like things just are moving faster. It's whether I'm getting older and my perception of time has changed. But I feel like things are definitely moving a lot faster than Linux back in 20 plus years ago days. But I think that's because we have things like GitHub, social coding, sharing is easier. The tools have changed, so I think that's probably part of it. Maybe you get a little bit older and your perception of time changes.
CRob (08:19)
Yeah, I keep seeing I have so much less of it available to me. It's a little scary sometimes. So back in January, there was a quote that talked about how Kubernetes is the operating system from AI. And I think that's very similar to how you noted the synergy between Linux and Kubernetes. And now we see Kubernetes and AI.
Are you comfortable with that framing or does it really kind of understate what's going on underneath the hood?
Chris Aniszczyk (08:53)
Yeah, so I think if you frame that as like Kubernetes is the operating system for AI infrastructure, then yeah, I absolutely 100 % agree with that, right? If you look at an operating system, what does an operating system do? It manages access to system resources and provides a set of common services for programs that run on an operating system. Kubernetes is essentially doing a very similar thing, right? It provides a, let's call it an API server that has a common lingua franca that works to allocate resources, whether you need CPUs, memory, GPUs now commonly for API resources.
And it's a very kind of a structured way. It has a very simple way to kind of declare things, a standardized API. It has a lovely little, you know, reconciliation loop that ensures that what you declare you want, you get.
Yeah, I consider that kind of like an operating system for a set of resources that are managed for your system. So I think that's definitely an apt description. And if you look at the reality, you know, all these modern AI systems that people use maybe on a daily basis, like ChatGPT, Gemini, they're all very much powered by Kubernetes underneath to help them help these systems, you know, scale, which if you kind of like keep peeling back the onion,
It's a bunch of like distributed Linux, you know, systems and services stitched together, right? And what is Kubernetes great at? That problem.
CRob (10:21)
Exactly. And I believe this came out in a survey you all did, and that same survey kind of flagged something interesting, where they talked about how cultural friction has overtaken technical debt as one of the top blockers. And what does that tell you about how your community has moved and matured?
Chris Aniszczyk (10:44)
I mean, we've always really been strong on solving hard technical distributed problems in CNCF. You know, I think what this problem, you know, you're kind of referring to as like, cultural friction over technical debt refers to something that, you know, has existed previously kind of in, if we kind of trace back time, there was a time before DevOps, there was a you know, problem where developers and operations folks, the folks that like rack servers and take care of all those and make sure things are up, they didn't really talk to each other, right?
CRob (11:23)
Yes, I remember.
Chris Aniszczyk (11:25)
They didn't have a, yeah, yeah, yeah. So, you know, and then, you know, there was a movement of like, hey, we could do better than this, right? We could, you know, maybe create technologies, have APIs, common tooling and language. And so hence the DevOps movement and culture was born. Similar thing is happening in AI land where, you know, obviously
You have, let's call it, know, data scientists, which are like proto AI develop, basically people like, you know, you know, working on algorithm, creating notebooks, generally, you know, folks that are away from how the infrastructure and how these things are actually run. And then you kind of have the AI infrastructure folks that sometimes are even spun up as brand new teams in organizations. You would not believe the amount of companies I've had talked to that have like a traditional, let's call it platform engineering Kubernetes team, and then like a parallel, like AI organization that was like spun up based on some edict from a CEO or CTO, right? And there's kind of this separation right now amongst these things. And there's kind of this issue of like, hey, how do we get these folks to kind of work together? Just like we kind of solved with the DevOps space. So I think you have a similar cultural issue here where the people that are actually developing the AI, let's call it
technology using
things, developing algorithms, data science are working way are not really talking to the folks underlying the infrastructure that they need to actually do those things. And so that's, think, where this cultural friction is coming from, in my opinion.
CRob (12:58)
Yeah, well, and I'll say that our AI/ML working group did a white paper last year talking about MLSecOps. And one of the biggest insightful findings there was the fact that there are so many non-engineers that are included broadly with AI development. So I absolutely, it makes total sense that we would have this cultural friction because we have a lot of people that aren't used to software engineering or infrastructure engineering participating in it's.
We're a little off. We're a little strange group of folks.
Chris Aniszczyk (13:30)
Yeah, absolutely. And know, for the folks that are in the weeds, you know, some of the tools that folks use to like train, you know, AI workloads are not in the traditional stack that maybe your platform engineering team has used here, right? There's a lot of folks that use, you know, like Slurm, which is a little bit old school for us to, you know, play with, you know, university lab setups a long time ago. That's a very common technology that was used for training things. And for inference, it's obviously a little bit different. But yeah, so it makes sense why this cultural friction exists and and like I'm super confident that the folks you know out there especially in CNCF and other Linux Foundation you know groups like PyTorch as they'll solve this problem like we're really good at this type of type of stuff
CRob (14:13)
Yeah, I totally agree. every technology has its rise, and then you get into kind of mainstream adoption. So I think we're kind of on that rising part of the curve right now, where everyone's learning and everything's very new and exciting.
Chris Aniszczyk (14:31)
Totally agree.
CRob (14:32)
Since you mentioned it, my little friend, artificial intelligence, there's both within your Foundation, my foundation, other permissions. There's a ton of tools in this space that can help things like also or in-toto. these types of things and like our, have some members like Kusari that work on developing these tools and sometimes they'll donate them so that like your foundation and my foundation, our projects get access to these things. But so from, have a lot of tools that can help.
From your perspective as the CTO and kind of the cheerleader for the community, how do you see maintainers actually adopting these tools or are we building a great toolbox that is kind of sitting off in the corner, dusty, unused?
Chris Aniszczyk (15:23)
Yeah, it's so, you know, we've always had a philosophy in CNCF to kind of give maintainers choice and, you know, give a broad swath of tools that they could choose from and like the best, you know, tools eventually kind of get used and rise to the top and allows us to kind of evolve like, you know, systems and tooling change, right? You know, back in the day.
You know, as an analogy here is like back in the day, a lot of people love Jenkins and then CircleCI came out and GitHub Actions. so like you have to kind of constantly, you know, adapt. And I think for a certain class of, um, you know, security tools, we've done a good job of, you know, adopting, least for like understanding provenance of software that's consumed by CNCF projects. We've done a pretty good job for the traditional, I always call it SCA and container scanning tools that we've widely adopted across the ecosystem.
For some of the newer, let's call it SBOM, you know, related tooling, you know, projects are a little bit slower than I would like. Like Kubernetes has been an early adopter of producing SPDX, you know, SBOM since kind of very early days. And that's great and kind of showed that it is possible to integrate it into your build and other things.
We also are working kind of a bit of a new initiative where we kind of prototype some tooling called SeeBOM, which we're actually working with OpenSSF on. If you go to seebom.cncf.io, you can check it out. it's kind of a combination of trying to suck in all the relevant SBOM related data that's generally produced and generated in different ways across project and try to give a visibility across the whole CNCF ecosystem, just so we could kind of like understand like what's missing, are the gaps, what can we improve?
And potentially encourage our projects to produce more useful data. to answer your original question is like, I think our role is to provide choice and options and kind of push maintainers in the best direction, but it's truly up to the maintainers themselves to sniff out what the best tools work. Cause you know, at least for certain SBOM related tooling, like, know, there's like the difference between like a guess bomb and like a real.
I don't say real SBOM, SBOM that's truly generated from build artifacts. And so as far as TIC, those are two different things. And one's a lot easier to implement than the others. And getting your maintainers to do the right thing to produce accurate SBOMs could be a little bit challenging. So our role, in my opinion, is like, hey, let's make this easy as possible and encourage maintainers to move in the right direction.
Because as...open source stewards, we should do our best in that department, especially with regulation that's coming up that producing these things is gonna be necessary. But I think it's a challenge. I would love to hear your thoughts of getting maintainers always to adopt the latest and greatest security things is always a challenge, right? Because they're always busy on like, I'm busy dealing with features and stuff and bug fixes right like get out of my way like stop bothering me so I'd love love to hear your thoughts on this we we try our best we try to make it easy we give people choice make sure it's free through great partners like you know Kusari and Michael over there but we can't force things essentially.
CRob (18:44)
Correct. and that's the blessing in the curse of open source is you have a infinite spectrum of choices, which is great, but you have an infinite spectrum of choices to choose from that makes it hard sometimes. Yeah, no, I totally agree. And I had the opportunity to get to see the Seebom work back at KubeCon. Really excited to see that moving forward. That's a pretty great piece of work.
Chris Aniszczyk (19:13)
Yeah, we're happy to collaborate with OpenSF and all y'all and I'm trying to prove that hopefully make it more accessible and useful for not only CNCF, but many, many other other folks that are going to come across the same problem.
CRob (19:23)
Exactly. Thinking about it, we touched on it a bit. There was a recent blog that you all had that talked about, it was arguing that AI sandboxing is currently having its Kubernetes moment and that AI labs are kind of converging on structural isolation while traditional security still kind of bets on policy and other kind of more traditional tools at the boundaries.
So do you buy that framing that they are going through this moment and they've all kind of agreed on this isolation? Or are there changes we need to make to Kubernetes to enable that?
Chris Aniszczyk (20:02)
I think we're still in the early days of AI in general. Yes, there's a lot of usage, there's a lot of innovation happening. I don't think there's any necessarily 100 % standardization on one way to do sandboxing. mean, this is a similar problem that Kubernetes in the early days, if you're just running Kubernetes on your own trusted infrastructure, that's great.
Right. That's a fairly trusted, you know, boundary, your control, pretty much everything. Once you start hosting third party workloads, right. And you don't know what's happening on what you're hosting, like a traditional cloud. Then that becomes a little bit tough. So what did people do? Right. They invented things like micro VMs and kind of stuffed containers to run inside of them to sure if a container escape happened, you'd be trapped. Right. And, and, you know, a micro VM. So I think there is, you know,
AI and agents definitely change, I think, what traditional security deals with. The attack surface is not, the boundary has changed, Like there's like these things, like agents are covered, right? Any network connection tool they use, like I even saw cases where agents were taking existing project we haven't seen, like Falco, like classic runtime security tool that
Chris Aniszczyk (21:09.325)
you know, generally for certain syscall usage, certain path usage, usually like a static check, right? Well, agents are smart. If you block it, maybe they're gonna go rename or symlink something in a way where they could break out of the tool. So you have to be, you definitely have to take a different approach, right? And sandboxing is just part of that story. So I think we're not in a way where there's like, a clear way to kind of, you know, do things.
I think we're all working together to kind of figure this out. There's a variety of open source projects, both inside the Linux Foundation, CNCF and outside. They're trying to kind of come up with ideas on how to do sandboxing. You know, know, properly for our workloads. And I do think some of the, you know, frontier labs and AI labs obviously are hitting this problem first. Right. And so they're kind of trying to figure it out, but I don't think we're there yet. Like I'm even seeing like Unikernels having a second coming with this, you know, whole, you know, untrusted AI workload thing, but we'll, we'll, we'll see.
So my, you know, to your kind of question is like, is there a Kubernetes moment? Like, I don't think so yet. Cause there's no like dominant, you know, agreed upon a way yet. There's just a lot of innovation and approaches. And I think that blog post, I think it was probably came from a community member or one of our members that was probably trying to push, you know, a certain viewpoint, which, we allow. love, you know, diversity points and CNCF. So yeah, that's my view.
I think it's too early to tell. And I think there's some innovation that is going to have to happen both inside Kubernetes and, and, and, know, outside there's some new agent sandboxing CRD work that's happening within Kubernetes that's being done by Google and others to kind of provide a primitive at the core Kubernetes to run agents, but that's still early, early work.
CRob (23:19)
Yeah, and I talk about this a little bit with my friend Mike Bursell over at the Confidential Computing Consortium.
Chris Aniszczyk (23:23)
Yep. Yep. Whole other layer he works on. yeah. Yep. Yep. Yep.
CRob (23:28)
Whole other layer of stuff going on there. So you talked about our friend agents, and they are increasingly doing a ton of additional work, pulling down code, generating code, executing code inside of clusters. What's a threat model that the community isn't really talking about enough with these kind of autonomous agents?
Chris Aniszczyk (23:53)
So, I mean, one way I look at this is, you know, I would say agents are smart. It's just not like a static binary you run and kind of does something, right? I think this whole notion of, you know, what I kind of call is like ephemeral evidence destruction is going to be a thing, right? Agents will do something, be naughty, hide their, you know, their tracks, like any well, you know written potentially malware out there will try to do something. But agents are even smarter because they could do it on the fly.
CRob (24:27)
Like my kids.
Chris Aniszczyk (24:28)
Yeah, so what do you need? How do you track your kids? Potentially maybe have a camera in the house to see what naughtiness is happening, or programs on phones. But this whole notion of what does a camera do? It observes. You basically need independent strong observability across all layers of this stack to A, provide understanding, evidence, and ensure that you have everything you need to kind of deduce what is actually happening.
And I don't think we've fully figured that out yet for the agent layer in a way that guarantees some security there. So that's something at least I think about and maybe I'm thinking about it. Often these days is because we have a project in CNCF called OpenTelemetry, which recently is going through its graduation process. And that's all about providing a standard API and way to do observability across many, many different aspects of your stack, including, and more importantly, agent-related workloads. So that's kind of on my mind. Agents can be naughty, and evidence can be destroyed very easily in this new world.
CRob (25:44)
Right, well, and also we've seen many examples where the agent is given explicit instructions, but the drive to complete the task is stronger. So it will absolutely find ways to work around that.
Chris Aniszczyk (25:57)
Yeah. Yeah.
CRob (25:58)
It's an interesting set of months for us all as we kind of think through these things and collaborate on solutions.
Chris Aniszczyk (26:03)
Yeah, yeah, totally.
CRob (26:08)
And a recent study said that basically only 7% of organizations today are deploying AI models daily, which again, in your cloud native space is, it's very strange. That doesn't happen. We're deploying daily, hourly sometimes. And we know that the infrastructure absolutely can handle that. Kubernetes and the associated technologies absolutely can handle this.
From your perspective, what's holding these organizations back from deploying more frequently? And is it your problem to solve or is it somebody else's?
Chris Aniszczyk (26:41)
Yeah, yeah, you know, I think there's a lot of facets here where I think, like I mentioned earlier, I think we're still very early in kind of the AI adoption phase. Some of us who are exposed to working with leading edge companies all over the world, we tend to kind of live somewhat in the future when you're kind of more traditional, industrial you know, or caterpillar stuff. There's a whole lot of organizations that are a little bit slow to adopt.
There's also regulated industries, right? That, you know, sounds great on paper, maybe to have an agent, you know, deployed that maybe helps with fraud or financial data. And then you realize that these things could, you know, potentially cause issues and make things up. So I think there's, there's still a lot to solve in terms of, you know, some level of determinism, at least around the security and servility aspects of you know, AI models, there's also a cost issue.
These things necessarily aren't cheap. You know, if you kind of look at some of the data, you know, people's AI bills essentially have gone through the roof, right? It's actually one of the, I have a peer or, you know, we both have a peer, J.R. Storment, who runs the FinOps Foundation. And I remember we talking to J.R. handful of years ago and the big focus at that time was like, Cloud bills, right? Like we need to get our cloud bills down. It was a huge effort to like standardize cloud billing and build this whole FinOps movement. And it's been very successful. But now instead of cloud bills, it's kind of AI, you know, usage, whether it's Anthropic or GPU, OpenAI type, you know, things. so cost is a big, you know, thing here because companies may be willing to spend a bit of money for a lot of gain in business value and speed and so on, but maybe that's not fully fleshed out yet for traditional enterprises to go to.
I think like, you know, I think I always view CNCF roles like provide the primitives and tools to ensure that you could go run models anywhere, whether it's locally or on the cloud of your choice and provide that management layer to ensure that you could easily do that in a secure and observable way. And we're working towards providing that. And I think in the next probably let's call it year or so is that their local models are becoming a thing, right? People are running models locally on their computer. They're also running models on cheaper GPU providers and so on.
There's going to be this kind of world where back in the day, people, they moved the cloud, they used one cloud and then they're like, crap. this cloud's expensive, I need to go use another one and provide some competition. And then, crap, I need to run this locally because it's actually cheaper. So we're getting to that model where AI is like, everyone's using the dominant one or two providers of their models. And then they're like, wait a we could run this locally, or we need to run this locally due to regulatory needs. So it's going through that essentially evolution that early cloud computing did to kind of run things both, no one really even blinks anymore that you could go run
Kubernetes and containers across three public clouds and your own private cloud. That's just works. We don't necessarily have that yet for AML.
CRob (30:08)
And I remember back in the day when Kubernetes was first kind of coming into its own, I worked at Red Hat and…
Chris Aniszczyk (30:13)
Yeah. Yeah. Yeah.
CRob (30:16)
…We had the controversial opinion that people should be able to run their cloud workloads in any cloud they want.
Chris Aniszczyk (30:23)
Yeah.
CRob (30:24)
And that was kind of a scary thing to say, but again, I'm glad that that idea has kind of taken hold, let people have their work be portable and move it where you want it.
Chris Aniszczyk (30:34)
Yeah, absolutely.
CRob (30:38)
Thinking about neutrality, you all have had a great long track record of kind of ensuring that things are vendor neutral. And as we see, there's just a handful of frontier model people and a handful of AI hardware people, and we're going to inevitably start to see some consolidation there. How do you think about protecting the neutrality and practice, not just align in your charter. How do you actively make sure that that's something that's kept in the community's focus?
Chris Aniszczyk (31:11)
Yeah, I mean, so, you know, definitely top of my mind. And honestly, one of the original reasons I came to, you know, build out CNCF was to really provide people, you know, choice, right. And I think if you look at the early days of CNCF, there was actually a lot of concern that, you know, Kubernetes may not be, you know, potentially vendor neutral because people will go fork it and kind of make it work so well on, let's say, a large cloud provider that people will just use that version, right? And then it's very difficult to go port off of that. So with that in mind is we did something, I think looking back now was really smart as we came up with such as something called the Kubernetes Conformance Program, right? Which was a way to basically have a TCK
you know, test toolkit that would go run against the Kubernetes API of your choice of cloud provider, you know, private, you know, or OpenShift, whatever product out there. And would basically go test basic, you know, compatibility that like, Hey, Kubernetes running on OpenShift or Kubernetes running Amazon, Kubernetes running on Google meets this bar, basic bar of interop. and
You know, at the time we're a little bit worried that, just people really care and, know, but it turned out to be really extremely successful. Like there is now a hundred probably or so products, projects out there that are conformant to that baseline. And, know, it necessarily doesn't mean that like you could magically go take you like one workload from the cloud and move it to.
another, because obviously you're using third party services on these clouds, right? That bit's not portable, but the basics are. And there are so many stories that I've heard from large providers out there that like, oh, we were maybe an Amazon shop and we moved everything to Azure. And just the fact that we use Kubernetes made that whole process a hell of a lot easier than it could have been. And so with the AI world is, at the time we were doing the Kubernetes conformance program, let's say you had like three, four big clouds, right? That people were concerned about and then eventually that ballooned, right?
And there's a cloud providers all over the In the AI world, you have something a little bit similar now where you have these kind of, I call them Neo clouds, but I recently learned that they don't like that. Like Neo clouds actually don't like that term. They prefer to be called AI native clouds. I learned, but yeah, folks like CoreWeave, Nebius, Lambda, there's a lot of these kind of GPU, they're essentially GPU first. cloud providers that will go run inference and training workloads for folks. And the setup of these folks, it's all different. Like it's insane. Like, you know, to be able to sometimes, you know, it's almost like going back in time where, you know, I won't name the provider, but basically you remember back in a day, you'd have to open up like a Jira ticket or a Bugzilla ticket, whatever the hell it was to get like someone to go give you server access. It's kind of that bad almost. It's some of these.
CRob (34:20)
Oh goodness.
Chris Aniszczyk (34:21)
GPU first providers. And so what we're doing at least with CNCF and this really new launch Kubernetes AI conformance program is to kind of bring those same principles of programmatic infrastructure, some basic level of interop for these AI native clouds out there and basically ensure that Kubernetes itself, which is mostly running on these clouds anyway, it's just the way things are kind of provisioned or done in a very, very different way, but kind of bringing that same level of interop and expectation that most organizations and end users out in the world expect. So that's kind of how we're going about of like, if we build a Kubernetes AI conformance program as strong as our original conformance program, we will guarantee and protect that vendor neutrality and practice, in my opinion.
CRob (35:10)
That’s awesome. I applaud you all for doing that. That's great.
Chris Aniszczyk (35:15)
Yeah, we only have about like a dozen or so AI certified solutions right now, but it's growing actively. And my hope is by the end of this year, it'll hopefully triple and continue to grow. A lot of these companies are funny enough new to how to really do like programmatic cloud native style infrastructure, oddly enough. And we're doing our best to kind of help them and onboard them into our community.
CRob (35:42)
That's cool. That'll be very valuable, I think, for the community once that's in place. As we wind down here, past Chris, what advice do you have for future Chris? Let's imagine it's 10 years from now and you're putting out your CNCF 20th anniversary report. What do you hope that's so obvious about Cloud Native then that it's not even debated?
Chris Aniszczyk (36:10)
Two things that I want to bring up. like one is when CNCF started and I kind of got involved in building out CNCF, there used to be, it was kind of funny, there was this undercurrent in industry that like, you know what, just choose AWS, right? Like just choose one vendor, they got it all, you don't need to use anything else.
It's kind of like back in the day, if you remember, there used to, you know, I worked at IBM back in the day and there used be a saying like, you don't get fired for, you know, choosing IBM or buying IBM, right?
CRob (36:45)
I’ve heard that.
Chris Aniszczyk (36:47)
That..it was like this funny meme that was somewhat true because, you know you had guaranteed support, you know, you would have someone to call, totally get it. you know, what I want is like, I want it to be so obvious, you know, that I think folks like, like choice and competition are critical and this is what end users expect, right?
And I think, you you wanna choose what's best, you know, for you and there's some standard level of interop. So I hope Cloud Native helps bring that, you know, to a bunch of other ecosystems. Like a lot of end user companies I talk to now, like big banks and others, they actually love CNCF because they feel like it has brought competition and choice for them versus that they don't have to depend on one vendor that either doesn't innovate fast enough or raises prices in some ways and kind of holds them hostage.
So open source and CNCF and the cloud native movement have brought choice and competition. So end users now could choose like, where do they want to run Kubernetes? Also for a lot of them now, observability is critical. So OpenTelemetry has basically allowed them to be like, maybe I don't need to just
CRob (37:30.562)
Mm-hmm.
Yes.
Chris Aniszczyk (37:52.065)
have all my stuff on one observability vendor, I could actually have some competition and baseline choice with OpenTelemetry powering a lot of that. that is something that, you know, I, you know, at least I, I don't think people will debate that anymore. It's there, there won't be this like saying of like, you just buy IBM or, you know, Amazon or whatever, right? Like you, like choice and competition and open source first things are, are, are just, are just there.
The other thing that, you know, kind of as important to me at least is, you know, we grew up working for vendors, right? You know, your IBMs, your Intel's, your Red Hats built amazing software, you know, and hardware, right? But I think the market change at least within CNCF that I've seen and actually now recently in other communities is innovation doesn't necessarily only have to come from vendors, right? If you look at CNCF, Prometheus, as born out of SoundCloud, you know, a listening music service, right? And it's completely changed the world of observability.
Envoy came from Lyft, you know, ride hailing essentially, and basically open source and shared how they did, you know, reverse proxies and essentially what became the service, service mesh industry at scale. That's from an end user company that was willing to share those lessons Backstage, came from Spotify, right?
And pretty much change how people do, you know, developer portals or IDPs is kind of industry's term. And I think, you know, the thing that I hope, you know, truly that people will debate is like, if you're an end user company, if you're a bank, you know, a listening service, automotive company, like sharing what you've done in the open and open sourcing it just becomes like a standard practice. Like, and that's something that we've lived, especially for us that like worked at Red Hat, we're like, open source first kind of methodology.
I think you're going to see that become more, more common from, from end users. And that's what I hope truly happens. And no one's just going to debate it. Companies are like, yeah, we're going to go share this because, know, maybe, you know, maybe I'm capital one in a bank and I'm going to go share cloud custodian because I just needed a tool to freaking figure out how to do policy and governance for, for my cloud things.
And it becomes a know, standard that a bunch of other companies come in and help and pitch in. So I think the culture of end users being more open source friendly and forward is something that we just won't like to be. It's just standards. It just won't be vendors. Innovation truly won't come from vendors only.
CRob (40:42)
I love that vision and I look forward to talking to future Chris about that someday, 10 years from now.
Chris Aniszczyk (40:48)
Yeah, I truly hope some of the coolest technology, I mean, some of the coolest technology we're using now today has come from end users. It's not just from vendors. It's not from just Google's or Intel or Amazon's. It's all over the world. Truly, truly amazing stuff.
CRob (40:01)
Absolutely agree. But more importantly, let's move on to the rapid fire part of our session. I've got a handful of crazy questions. Just give me the first thing that rolls off your head.
Favorite vacation location?
Chris Aniszczyk (41:19)
Sure.
Oh, I would say I literally just got back from Mexico. So any place in Mexico along the coast, western side, eastern side, I just love the beach and jungle vibes that Mexico provides on top of amazing food. I would say best food country in the world by far, in my opinion.
CRob (41:44)
Our mutual friend, Puerco, endorses that answer.
Chris Aniszczyk (41:48)
Absolutely. It's awesome. So I think Mexico underrated, even though I know Mexico City is popular these days, there's a whole swath of the country that I love to visit. Like I've driven coast to coast from like Western side to Eastern side across all these beautiful cities like Campeche and Merida. I just have a special, Mexico has a special place in my heart.
CRob (42:09)
That's awesome. What's your favorite beverage?
Chris Aniszczyk (42:13)
Unfortunately, Mexican Coke, not like it's for non alcoholic Mexican Coke because it just hits different love the natural cane sugar it is just a great product. It's probably why they're around for you know, over 100 years good stuff. In terms of alcoholic beverage, I like a little you know, you know, bourbon, you know here.
CRob (42:39)
Yum.
Chris Aniszczyk (42:40)
You know ~ here and there.
CRob (42:)
I think I know the next one hot or spicy food or I'm sorry spicy or mild food.
Chris Aniszczyk
Spicy. Yeah spicy by far like why why why live another way you need a little…
CRob (42:58)
Right, love that.
Chris Aniszczyk (43:04)
Yeah, you know, I love a good salsa matcha little habanero sauce.
CRob (43:10)
Now I'm hungry. I have to go eat. last but not least, Star Trek or Star Wars?
Chris Aniszczyk (43:19)
Oh, Star Wars. Sorry. Yeah, yeah, yeah. No, yeah.
CRob (43:22)
Nice. There are no wrong answers. I love both, but it's not nice to get somebody's personal view on it. Yeah.
Chris Aniszczyk (43:30)
Yeah, no, I find the let's call it the landscape and it just I just find there's a lot more depth. But whatever we all have our different try to say it in a nice way. But yeah. Of course, of course.
CRob (43:51)
Right. Well, we, in open source, we love us a good nerd fight occasionally. Well, and in closing, Chris, kind of for our audience, do you have any, you know, call to action or any closing thoughts you want to share with them?
Chris Aniszczyk (43:59)
Yeah, you know, I would say two things. You know, when I spent a lot of my time mentoring new contributors to open source, right. And so, I mean, and I do it mostly because in my early days, like people helped me, you know, just like the randos on, you know, IRC channel a long time ago helping me on some things. Like I feel like I love to mentor folks. We have great mentoring programs in CNCF and contributing resources. If you go to contribute.cncf.io lots of great materials, you get stipends to contribute to CNCF project. And I truly think it's one of the best ways to get involved.
And I truly think like people are concerned about agents replacing humans, developers, like nothing will ever replace, you know, thoughtful maintainers that are working to steward open source projects. You're always gonna have a bit of a human element there. And I still think it's a great opportunity for folks to build a career in open source.
So, contribute.cncf.io, learn how to contribute, we're always happy to help. And then finally, back to your first point of this being a bit of a global phenomenon, CNCF, we do five KubeCons across the world now in North America, Europe, Japan, India, and China. So please check our events out, truly amazing events that are really now the largest open source events in the world. So we'd love to have folks there. Face-to-face interaction cannot be replaced by anything. And it's funny to see, it's like a…you know, during that fun COVID times, like, people are just gonna like virtual conferences, virtual meetings. And it turned out to be like, humans love to spend time together discussing hard ideas, building bridges, you know, compromising on solutions, like that, that, that will never be replaced. And I think we'll only increase in value in our kind of future agentic role that we're willing, you know, building. that human to human element is very key.
CRob (46:03)
I agree and sometimes you get to do that with a tasty bourbon.
Chris Aniszczyk (46:08)
Yeah, absolutely. Yeah.
CRob (46:09)
So again, as we wrap, thank you to Chris Aniszczyk. Thank you to the CNCF community, some amazing projects and kind of, I love the culture and just the consistency there. That's you guys have done a phenomenal job. And with that, we'll call that a wrap. I wanna wish everybody to be cyber safe and sound and happy open sourcing out there. Thank you, Chris.
Chris Aniszczyk (46:31)
Thank you.
Clips from this episode
You May Also Like

Hell Money
Hell Money

Machine Learning: News on AI, OpenAI, ChatGPT, Artificial Intelligence, AI Models
Machine Learning: News on AI, OpenAI, ChatGPT, Artificial Intelligence, AI Models

ITSPmagazine
ITSPmagazine, Sean Martin, Marco Ciappelli

The Paul Barron Crypto Show
paulbarronnetwork.com

The New Stack Podcast
The New Stack

MacVoices
Chuck Joiner

Bitcoin Magazine Podcast
BTC Inc

Jack Hopkins
Jack Hopkins

Day Two DevOps
Packet Pushers

The Fat Pipe - Most Popular Packet Pushers Pods
Packet Pushers
Community discussion