Ep. #10, The Whirlwind Pace of AI with Taylor Dolezal

John McBride: Welcome back, everybody. I'm here again on the Open Source Ready podcast. Brian, how are you doing?

Brian Douglas: I am well, doing very well. It's a great Friday here, Valentine's Day, and I'm really lucky to be here with Taylor.

John: Yeah, we're here with Taylor from formally the CNCF, the head of ecosystem, leading a ton of initiatives there and now, the Chief of Staff at Merly AI.

We're very, very excited to have you here. Taylor, how are you doing?

Taylor Dolezal: Doing well. I've been checking my start date at Merly and I keep going, wait, has it been a week, two weeks, three years, what's going on?

So, I'm in some kind of time dilation as far as everything goes there, but it's been fun. It's been fun. Head of Product Engineering, Chief of Staff, dual rolling it.

I'm pretty sure there's something about that in the Dungeons and Dragons handbook.

Brian: If you have only one role at a startup, you're probably doing it wrong. So at least you made it official.

Taylor: You can't see it, but I'm wearing lots of hats today.

John: Excellent, excellent. So as I mentioned, you existed within, you know, this kind of amazing ethereal role at the CNCF as the head of ecosystem.

Can you just tell us what that was? And maybe I'll kickstart you on a bit of a retrospective on the CNCF, that whole ecosystem and get going from there.

Taylor: Yeah, absolutely. So at the CNCF, I worked as head of ecosystem. And what that meant was I needed to come up with a new elevator pitch every time I told somebody about what I did.

Different audiences too. My parents asking, "What exactly do you do?" And at the end of the day, it was working with the end user community.

So being aware of everything going on within the ecosystem, being aware of all the projects and different initiatives and things that were happening.

But we had other people to help out with that, like Chris Aniszczyk and Jeffrey Sica, known as Jeefy, Bob, and the rest of the team. There's several others that I'm not naming just because there's lots of people helping out on that front, but love them all.

I got to talk to folks like Intuit, Mercedes-Benz, Apple, and about 163 or so other members at varying levels, roles within the CNCF and the Linux Foundation.

And we talked about a whole bunch of things from strategy to project that it might behoove them to check out. It was an interesting role, because of the fact of that vendor neutrality, right?

We weren't making recommendations. It was just saying, okay, here's all of the LEGO pieces that you get to play with. And figuring out how to be supportive and a steward of that end user ecosystem.

There's no other open source foundation or effort that... I haven't seen any that I'm aware of. I've been looking, I had looked when I was in that role for like is there anyone to follow, or copy off of or iterate from.

And there really wasn't anything. So, that was one difficulty as far as the role, but kind of the fun thing too, 'cause you're the first through the wall to figure out what's going on with some of these efforts.

And everyone's looking for something different too, depending on the size of the organization, the vertical, et cetera. So, it was never boring i s kind of how I would try to summarize in TLDR, describe that.

But as you can see, I feel like I'm over five minutes with describing that. So this is a long elevator ride, but lots to do. Lots to do.

John: Yeah, that's great. I mean, you have such a unique view on, you know, just the open source ecosystem at large. How is the health of that ecosystem?

Taylor: It's good. It's well. It's Friday. 2024 was a very interesting year because we went through so many changes with projects that were forking, trying to figure out from the end user perspective what is the right path for us.

What should we care about? And the end users would typically have a day job or another focus, right?

So as much as they would want to be working within open source 24/7, you know, if they never had to sleep, or eat, or any other humanly need, they just couldn't because they're making cars, they're making tech software, they're working in manufacturing.

So it becomes an ask from them to the foundation, in this case the CNCF, like, hey, can you free up some time for us or some guidance or something.

Such that we're able to contribute with a little bit more clarity or faith, can you help read us in or get us from zero to 60 and much more quickly than we'd be able to on our own? So, that was a big thing.

Brian: I love to talk like really this-- double click and I love that term, just double click, because it's kind of dumb. But anyway.

Taylor: I heard it.

Brian: I wanted to double click into like 2024, like literally, are you speaking to OpenTofu? Is this the situation?

Taylor: Yeah, how'd you know? Like yeah, OpenTofu, Valkey.

Brian: Yeah, so you're like, oh boy, yeah, there's a couple of them. So you're steering the ship.

You're interfacing to end users who don't have time to read a bunch of blog posts again or Hacker News.

Like how are you interfacing with that? And like, what was it like being on the ground for that?

Taylor: I had a surprising amount of people reach out and say like, "What do I do?"

And I'm like, Well, I can tell you your options, but it's you, you make the choice, right? It's you choose, it's a choose your own adventure book.

And so, I can't tell you, you know, like I'm not the one that gets to flip the page back if it's like, you know, page 34, like, oh, sorry, you fell in a trap hole, you know?

Like I can't help out with that sadly. But I always try to give my best educated guess on some of those things. But there was a lot of confusion, right?

Because looking at the license, there's a lot of talk about what is the role of a license, right? We know the letter of the law, but what is the spirit of the law?

What does the spirit of open source look like? And how do we adapt things to our workflows as end users, right?

So, I got put into that headspace quite a bit, especially with all of the AI things that were coming out, seeing budgets shift from what was cloud native, or cloud, or that kind of focus to AI focus and what does this actually mean.

I don't know. Dealing with all of these different kinds of approaches to what that looked like. So, it was a little bit confusing and dizzying with how fast all of that came to me, to us at the foundation, and just as everyone was experiencing it within the community.

I haven't seen anything like that since, I dunno, since like the beginning of open source and hearing things of like .net switching over to not being closed source anymore, right? It's a bit of a paradigm shift and just a lot of conversation.

So, a lot of uncertainty. But when we take a look at AI and open source, we see a lot of similarities there with like this fear of engaging with it, this misunderstanding of it, and then over time starting to figure out like, oh, this is how I hold it, this is how I wield this, and how I can be effective with it.

So, I'm not too concerned frankly, just because I've seen this before with DevOps, open source, right? All of these other kinds of similar things.

They aren't the same, but they follow similar patterns. So, it was dizzying, but despite that dizziness, that's just part of the job, right?

The only constant is change. So, being ready for that and trying to take a look at some history to get that clarity and certainty is the best thing that we can do.

John: That's amazing, yeah, and I think the shift that even from an end user perspective, I've seen with Tofu and, you know, Terraform, and some of those big licensing changes or even--

You know, we could talk about how Amazon brought back open search into the open search foundation and licensing changes all over the place.

What has been your perspective on, I guess the business side of that? Like have you seen end user companies wanting to reduce their risk by adopting some of these open source technologies?

Risk may be being legal risk or like capital risk. You know, where it's like, oh no, now we got to use this OpenTofu thing. And like we're spending all this engineering time actually doing that. Like has that perspective changed for companies or OSPOs?

Taylor: Of the larger companies, it was a lot of let's wait and see what happens, right? There is this benefit to waiting.

I think that what Eddie Knight, I remember talk... It was either a talk that he gave. He works at Sonatype, for this that don't know him.

He gave a really good talk at a open source summit. I think it was about two years ago. It might have been three years ago, where he was sharing data from what happens when you upgrade your packages immediately, right?

It's actually riskier. We've been told that this is a good practice, but it's actually not when we look at the data, because in a dot zero release there can be bugs and that can cause far greater risks than we know.

So in a similar case, that's kind of the guidance that makes the most sense for larger organizations or more risk averse organizations.

The more risk tolerant ones, they can jump in on that front or kind of have like a plan as far as how to mitigate those kinds of risks.

But that's anecdotally what I saw from a lot of organizations was like, we're just going to step back and wait to see how this pans out.

The news of IBM acquiring HashiCorp also, they're like, okay, I'm going to wait for that to close then see. So people were starting to get some plans together.

Whereas others, we had a couple end users that were a little bit more prominent in sharing that like, yep, we've shifted over to OpenTofu.

This is a license that we can trust. We've made peace with and that we understand internally as far as our culture goes.

So same thing with Valkey too and several others that we're shifting over within the course of the past couple of years.

John: Yeah, you did briefly mention AI there for a second, and I do definitely want to talk about that, but I'm curious your perspective on Kubernetes specifically and AI.

The last few KubeCons I've gone to, it definitely seems that Kubernetes is almost positioning itself as the defacto platform to build a AI ML, like a dev platform on top of GPUs or, you know, whatever nodes or the compute that underlie all that.

I've even built a small GPU platform on top of Kubernetes, and it was amazing. Is that the direction you see Kubernetes going or being another sort of a facet to this, you know, computing ecosystem within CNCF?

Taylor: Definitely, I think there's a want to kind of capture that at the CNCF too, right?

And I fully back Kelsey Hightower's take on Kubernetes over time fade into the background, much like Linux has and I'm like plus one. 1,000%.

I know that's over a 100% for the scientists that's out there, but that's a 100% what I agree on. And I think that I've been startup life, so sleep is not a luxury as much I'd say anymore.

John: Oh, we get it. Yeah, we get it.

Taylor: Even with CNCF and everything too, like the community is always awake somewhere. So, you know, it's five o'clock somewhere, the community is always awake somewhere.

And there were some things I was taking a look at last night talking about the difference between what it takes to train models, right? And it's very similar with people, right?

It's if you take these 10,000 hours and learn how to play the piano, tennis, chess, whatever it might be, that's a lot of time.

And then when you move over to actually doing it, the inference side takes a couple hours, right?

Even though Monopoly game feels long, it's much longer to learn how to play that game rather than to actually play it.

So I think that that's an important distinction that I'm going to keep my eye on. Is like what does a separation of both of those kinds of workloads look like?

I've seen a lot of Kubernetes used for data processing, event-based workflows, those kinds of things when it comes to the training side. And that feels like a much better match with Kubernetes and everything else.

I think when it comes to inference, that's the one that surprises me. That it feels very much, again, like the DevOps movement kind of thing happening, where the development teams are kind of rolling whatever makes sense for them.

Sometimes it's Kubernetes, sometimes it's not, sometimes it's a different runtime altogether or like Slurm, SUNK, there's all these like onomatopoeic words that people are choosing for product names.

And on that one, in my heart I think that Kubernetes is the right move there because it's the same primitives. It will allow for easier interoperability between the training and the inference and tying all these things together. It's a common language that we all know much like Linux is, right?

Do we need a different type of Linux to be able to run this? I would argue no. And even the inference side of GPUs and how GPUs integrate with Kubernetes, that's still... It's a solved problem in some aspects, but it's still rough around the edges.

At least that's how it feels to me. That's anecdotally that's what I've heard. As you can see behind me, I don't have a fleet of each one hundreds, and I'm not testing this out personally, so I can't speak to actual experience, but it's really interesting.

And seeing platforms like Groq with a q, and how fast they're able to provide this kind of inference, I would love to see more data around that and kind of like, what do those workloads actually look like?

And I'd love to hear from more people around what are the GPU based issues that they're having? Are they using things like Ollama to test locally and just running things off of the CPU?

Can we make things better there? There's the DeepSeek conversation. Obviously I could talk about this for hours, but we don't have that time.

John: Well, that's what we're here for yeah.

It is interesting 'cause yeah, even at a small scale, when Brian and I were at OpenSauced, building our very small Kubernetes GPU ish platform, you know, just driver issues with like getting the right version pf the right drivers onto the nodes to be able to, you know, actually run VLLM to then do the inference while, you know, eventually being able to figure it out.

I'm excited for more turnkey solutions on Kubernetes or maybe looks like a turnkey API down the road, like a Groq or something in the future.

You mentioned something briefly about, you know, kind of the rise of DevOps being a whole, you know, a whole movement and a whole thing.

Part of me feels like, you know, with all that context, we're on the precipice of this like new movement and I know some people have coined it MLOps, or kind of the merging of like, you know, data practices and data things and like platform engineering and also, you know, sort of the AI engineering product side of it.

You know, there's huge gaps between that where, you know, data, people don't always want to care about what's going on on the actual infrastructure.

And product people really probably don't want to care about how data built the actual models, but you know, they sort of kind of are bridged between all that.

Is that a movement? Is that happening? Is that solved? Like where are we with all that?

Taylor: There was a shocking moment that I had while at the CNCF, 2023 it might've been, DockerCon. I live in Los Angeles. They, they had a DockerCon here.

And despite containers, Docker kind of the relational history with Kubernetes and all of these container workflows, CNCF, et cetera, I walked around this expo and I saw maybe two or three people that I knew, and I was like, why?

Why is this I, where are all my friends, right? From the community, from the systems engineering side? And the answer obviously was they weren't there.

And so it was all of these developers that were, this is like in the beginning kind of throes of AI and like the gen AI stack, I think was part of one of the DockerCon keynotes that was there, got rag style applications were just kind of coming onto the scene.

And so I was like, this is really interesting. That like, again, no one I know is here. Typically, there's I'm not seeing the overlap that I'm used to seeing.

And so that was what was sparking that initial thing. Like, okay, this feels a little bit like the DevOps movement.

You have your developers over here, your ops over here. And then as lately as I got invited to the DevDays event by OpenAI.

And they released obviously some really cool things, but it wasn't spectacular in the sense of the technological interactions that were there. It was things like web sockets, you know, real time AI options and APIs.

I'm like, yeah, we've had this for a while, web sockets, WebRTC, all of these technologies.

And so, I'm seeing a lot of duplication and to me that signals that oh people aren't talking to one another. And that concerns me a little bit. And that's the DevOps feel again that I'm kind of sensing within the community.

So, I think that's a great opportunity for us to be able to talk and say, yo, you know, you could use containers and these other things.

We have X at home, right? We have food at home and it's actually good. It's cool this time. So I think that there's a lot of opportunity for that.

The devs are moving incredibly fast within the AI space, but I think that, again, we can all benefit from having systems, platform people kind of get together and collaborate, coordinate, and try to figure out some things that make it easier for us all to deploy.

Even going to AI meetups and things in San Francisco, Los Angeles and Paris, France, when KubeCon was out there, I just kept seeing this thing happen again and again and again.

And typically there are these really cool demos that are being shown, but it's all like, hey, check this out. I have this on my laptop, hold on, let me pip install requirements.

Okay, cool. You know, it's nothing that you can actually pick up and take with you. There's not a lot of interoperability past someone's machine.

And again, we've already done this with works on my machine, right? Like, no, let's get it to the clouds. Let's figure this out. It will take time.

I'm assuming heavily that it's just, we just need to spend more time on it and kind of have that patience.

But that's hard to do when there's a new technology and you're in the throes of the hype cycle, right? No one wants to be like, hey, let's slow down and figure this out for real. Like, ah, get out of here.

Look at how cool this is, right? Yeah, I get it. I've been there, done that. You know, I get it.

Brian: Yeah, I just spent some time at an office in Hayes Valley, one of these up and coming tech companies or sorry, AI companies that took a lot of money.

And just like to start out some ideas, like super excited about their tech and be like, hey, you should think about this.

I'm like, yeah, we have no time for that. I was like, yeah, which is what I'm asking them to think about is like, Hey, I'm normal dev, just want to use your product.

It'd be really good to have a better developer experience. And they're like, yeah, we're working with enterprises and they're going to figure it out.

And they're spending lots of money for us to walk into the door and teach them how to use this. We don't have time for like the actual on the ground dev stuff.

So like, I could feel where the worlds are kind of like conjoining eventually, but there's a huge gap right now.

So like anybody looking to start a startup or improve developer experience with MLOps and et cetera, like it seems like an exciting time.

I'm not sure how long this will last, but it's exciting right now.

John: That's been a lot of my experience, you know, day to day building some of these things with, you know, some existing libraries like LangChain for example, you know, great stuff.

And they were so early on the scene, but almost to their detriment where like they have to break the API so frequently that it's kind of a mess.

But then like, you try to search for things and you get old versions of documentation that are popping up with SEO and you're just like, I can barely build with this tool because they're almost moving too fast with the library and all the things.

So yeah, I also feel that kind of convergence happening. I'm like excited for things to kind of stabilize maybe, or just like become a bit more standardized, so that, you know, not every week, I'm trying to like, you know, reinvent the world with, you know, new stuff, right?

Taylor: It's like we got chain of thought, we got program synthesis, we got this, we got that. It's like, what is go... yeah, it's just like I was trying to ship, but now I can't. You broke all my stuff. It's true. It's really true.

John: It's like the ultimate nerd snipe. I think the industry is just like look at this, look at that, constantly something new.

Brian: Yeah, I mean we had that with... so we had DeepSeek a couple weeks ago, and now we have like I can't keep up with how many OpenAI announcements there are between the 03, and then not the 03, and then the GPT5.

It's like, cool, we're going to get something new from OpenAI. I don't know what it is, but I'll wait until someone distills this into a blog post I can read.

Taylor: Anthropic came up with something last night. you know, I know, I was like, just work speed as far as, they're like, oh, we have like a fusion model where it's like part chain of thought, part mixture of experts, like all these things, you know, finding out MOE isn't your friend's name, it's mixture of experts and like new acronym.

What's, oh my gosh. So it's hard to keep track all this.

John: Yeah, my very, I guess hot take maybe is that, you know, of course these companies who have the spotlight on them want the spotlight to continue and, you know, need that momentum to keep going and going and going, especially if they're going to raise, what was it, like half a trillion dollars to go build new data center somewhere.

Like they're going to need the spotlight to raise that kind of money. You know, Taylor, you're now at an AI company. What has been your perspective on that?

Like do you need that sort of constant churn of like new things happening? Or what is it like existing in the market right now at a startup in the AI space?

Taylor:

It is truly a whirlwind when it comes to trying to think through how to develop product on this front, and then what kinds of bets are you making in the construction of the AI stuff that you work on.

So, what I mean by that I know is very abstract, very generic, but you see big companies reach for AI, and that's the product, and that I disagree with personally.

I think that it's an ingredient, right? It's like salt or eggs or lettuce or chives, right? It's part of it. It's not the product. It's a product for the builders that are making something else.

So, that's where I think a lot of companies have gone wrong, where again, they'll go for like, hey, we can summarize this.

You know, it's like, oh, okay, that's a good use case. I'm not arguing against that, but that's something everyone has.

It's not necessarily special, as well as I feel like a lot of early adopters threw away most of their stack and forgot like, forgot the meaning of the holidays, right?

It's like, why are we doing this in the first place? And you don't have to throw away all your web dev experience, your systems knowledge. None of that goes away. It's just what happens if you add a database, a cache, some kind of new primitive has emerged. So this is a new material.

You don't build the same way with wood that you do with concrete, right? So when you're fusing these things together, you can play to all of their strengths.

And it's going to happen, that's fine. But that's part of the learning process is making a mistake and learning from it.

And some folks have, some folks haven't, some folks are still saying like, ope, AI is the thing. And I disagree heavily with that. I think that it's figuring out how to incorporate that.

Anecdotally, I've heard a lot of folks trying to figure out what their AI strategy is, and a lot of it is just sign the deal, make the big press release announcement, sign the deal, get access to the GPUs. And then unfortunately, there are several folks that I know of that aren't using them at all or minimally, like 2% utilization.

John: Oh wow.

Taylor: They have them. They're being powered. They're ready to go, but they're not being used.

And so because it's scarce right now, and that's what people are really buying, is just the opportunity to be able to potentially use the thing.

I'm kind of concerned about that too, when it comes to even taking a look at DeepSeek and these other things.

I do think that it's a win for open source and it's proof that this can be made better. I am highly optimistic, even in a cautiously optimistic way about LLMs becoming able to run on commodity hardware and CPUs, right?

I think that it will get down to that, but when you think about it, there's really no incentive for NVIDIA or these cloud companies to prioritize that, right?

Because that's more products being sold, that's more cloud compute being bought. So, why would I optimize that? I'm just going to give you more ways to use it.

And then, the onboarding types of frameworks, right? That fiscally makes a lot of sense, but is that the right thing to do? I also disagree with that too.

I think that there's smarter ways that we can think about how to design a better product and make it useful, like you said, with the developer experience, actually think about the long-term. I think that there's far too much short-term focus on all of these things, and it's okay to slow down.

I know how it feels. I think it's worth being intentional, rational, and trying to think about how we can be iterative and adaptive when it comes to these technologies rather than like we solved it.

It's like, no, it's again, the only constants change. How do we set up a framework such that we can continue this growth?

John: Yeah, that's a great perspective. I once heard somebody compare NVIDIA to Apple, where, you know, Apple essentially has captured, at least with MacBooks, you know, the entire stack from hardware to software now.

You know, you're not taking macOS and running it on a Raspberry PI, and they have no reason to enable that, because they want you to go buy the $3,000 laptop.

Maybe the same with, you know, CUDA software running on NVIDIA GPUs, right?

Taylor: I think it's incredibly interesting. Recently, as I'm sure many people out there might have done too, like, oh, who is Dr. Jensen? What's going on?

And kind of going all the way back within NVIDIA's history, I didn't know that they were about to go bankrupt with everything going on with like, what was it?

Like the Sega Dreamcast, and like NVIDIA, and like all of... There's some cool lore there for anyone curious. And you see luck really play out, right?

Luck really does have a big handle in deciding things. And even if you're fully qualified and ready, that doesn't always constitute, that's like 2/10 or a fifth of the equation.

The rest of those 4/5 is luck and simulations that they've run and specific things, I've taken a lot of look at that too, just out of curiosity and to see blockchain come along, to see gaming come along, to see AI come along, like NVIDIA knows very well how to lean into the market at the right time.

So, kudos to them on being able to... "CUDA" on them to-

John: There it is, there it is folks.

Taylor: Sorry it took me time to warm up.

John: Yeah, yeah. A classic Taylor pun, I love it.

Taylor: They know how to lean into the market. And I think that that is a really cool story, right? I've rarely seen people be able to execute like that, you know, especially with AI, especially with blockchain.

Whether it works or not, doesn't matter. They've been able to move product and that's awesome.

Brian: Yeah, I mean, it's funny you bring luck. I was actually chatting with, it was like a weird flex, but I was chatting with the CEO of HackerOne, who was also the former CEO of MySQL.

Taylor: Oh, wow.

Brian: And talking about he had worked at HP on something that was competitive to Kubernetes. Well, it wasn't really competitive. I forgot what it was called.

But everyone had their Kubernetes, their orchestration software, and it got to the point where everyone's just like looking over like, oh, where did this thing come from? I guess we're using that now.

But also talking to Brendan Burns at another conversation, it was calculated, like they spent a lot of time with the community. They built up adoption. They talked to people inside of GitHub issues.

And that's what was the difference of like the community, the ecosystem makes a ton of sense. And like for what you were doing before at the CNCF.

It's luck, but it's also like, oh yeah, Taylor can go like text somebody who has impact and whatever other coordinated effort to be like, oh yeah, let's go think about what the story's going to like pan out for this AI journey within Kubernetes.

And sometimes, it's the right place, right time all the time. But it's just because she happened to be working late nights and actively in Paris at a random meetup to listen to some random CEO.

Taylor: Completely, if you take a look back at Kubernetes and kind of how the CNCF got its start, that was because Chris and Dan, as co-founders of that foundation, they poured a lot of time into traveling, going to meetups, right?

There was this... I think the terms like the creative surface area, right? They like expanded as much possibility and hope and this potential that existed.

And that's what really attracted developers, engineers, and got people saying like, oh, this is cool. And since Kubernetes, the community has driven a lot of that, but it hasn't been this centralized effort for things I'd say is like OpenTelemetry or Argo right?

there are people that lean into those and that's why I would posit that they're so special and magical is because they have a driver behind them.

And that's seeing a lot of projects come into the sandbox state for those unaware, there's sandbox, incubating, and graduated projects at the CNCF, sandbox is every level of that is worth celebrating, 'cause it takes a lot of due diligence to even get to sandbox, right?

You have to show promise and possibility, but with each level you have things fall off or they kind of like go through an infinite loop until they end up falling off too.

And taking a look at sandbox, you see some projects come in and there's a misunderstanding from some organizations or just life happens.

It doesn't find good product market fit or go to market strategy and it just kind of falls off. But you can see some projects like committing all the time, interacting with their communities, really working on things.

And then others that, you know, you could go to devstats.cncf.io, and take a look and see, oh, last update 218 days ago, ah, is, you know, is this project okay?

And you have that visibility, which I think is, you don't get that everywhere, right? You don't get that bird's eye view into an organization, especially private ones or other kinds of startups.

So for those that love to look at the data, that's very fascinating. A good way to spend the evening instead of reading "Goodnight Moon." That's what I try to do anyway.

John: There you go.

Speaking of a bird's eye view, I don't think any retrospective would be complete without, you know, looking at what went right, you know, what you would do again and maybe what went wrong, and what you wouldn't have done again.

What's your perspective on that?

Taylor: So, I was at the CNCF for three years as the Head of Ecosystem. And when I joined, I was incoming from HashiCorp, and I was on the track to become a manager over there and lead the developer relations team.

I was a senior DevRel, I forget the actual title. I'd have to go and refer to my LinkedIn. But interacting with people, community talking.

I was coming from this place of working across many different internal focuses: silos, engineering, product marketing.

And so when I joined the CNCF, it was kind of like me going to Merly now. It wasn't something I was thinking or planning on or had this like elaborate kind of sequence of events.

It's just like life, you know, serendipity happens. And I went to AWS reinvent. Priyanka had reached out to me and said like, "Hey, I have a couple roles open if you're interested."

And I'm like, I'm going to take this call. It's Priyanka calling. Like I'd be silly not to. So when I talked with her, I was like, Oh, this is a lot of fun.

And this came about the time that HashiCorp IPOed, and I saw this shift internally kind of go from open source is awesome and fun to like, let's sell more cloud.

And I'm not against that. Every company should be able to make money it, right? But at the same time, we have customer success teams that DevRel is happy to work with, to support that goal of these external sales or inside sales.

That spirit of open source is what I really wanted to focus on and work with the community, get feedback, bring that feedback to product, or engineering or marketing, and then create that. And then it's this cycle and this flow.

So, I came into the CNCF with that kind of mindset top of mind. And I had a couple predecessors that... The end user program didn't really have a leader for a couple of months.

And so, I walked into things. It was, yeah, it was a little messy honestly. It took about six to nine months to really get a sense for what was happening.

This was right as the world was kind of like reactivating from COVID, KubeCon, Valencia and the EU. And there were all of these things that happened, some that were really intense too.

Like I joined in March of 2022, in May, we had our first Linux Foundation All Hands after I think it was like six years or a very long time.

Unfortunately during the course of that, the current CTO of the Linux Foundation had passed away at the events, just had a heart attack and unfortunately passed away.

So, there was that going on. There was a bunch of upheaval within the community. It was a very turbulent beginning, being very transparent and honest with you.

And it was a lot to adjust to in the first two months. But it got me ready and laser focused on what's important, right? Like we just had some fires out here in Los Angeles.

It's unfortunate that it takes these kinds of calamities, or tragedies, or just difficult moments. But despite how difficult they can be, it brings out a completely different side.

I wish there was a way to hack that or shortcut it, but there really isn't. You see, you know, the fires and all these things happen. People lose their homes.

But then you see beautiful displays of people getting... I knew there were this many people in LA I've never seen all of them at the same time, donating things, checking on one another.

I've never seen like a friendlier interaction I never expected it from the city. I constantly saw that within the CNCF community, no matter what happened, you could be sure that whenever there was a storm, there was going to be a rainbow at the end of that.

With people coming in, people having levelheaded discussions, respectful disagreements. I think that's what makes the community magical is that it's not this declarative thing, where it's just like it becomes bad and it stays that way or it becomes difficult and you can't get past that.

So, that was a very interesting onboarding to what it really takes to be a leader in the open source space and how to be a steward, right?

And be a servant leader rather than, right, 'cause the CNCF doesn't come in and like punch down out of power and say, "No, you can't do," that's not our place to do that. It's to enable.

So, how do we bring these people together to solve things? So, that was how things started.

It took a while for me to understand things like, oh, end users really don't have all this extra time in the same way that our vendor members might.

Vendors are focused on product and features and selling that, the CNCF can pick up a couple things to help out with like education or making some reports or other things to help out with the end user community.

And so, that took about a year, year and a half to learn. And then with all these learnings, I really wanted to put these things to use. So, we started up Zero to Merge, to get people interested in contributing for the first time.

At Disney and other places that I'd worked, I saw that a lot of people might have been senior principal engineers sitting in a role for, yeah, a long time, but afraid to take that first step and actually contribute 'cause they might be wrong, right?

But again, good real learning is difficult. I can tell you it's worth pushing through as I feel like both of you know, a startup isn't easy, right?

Like, and so that was really encouraging to see so many people that joined that program and we taught them how to add a profile picture.

Here's how to make yourself marketable. Here's what to keep in mind. Yes, this is going to be slower working in the open source space, but taking the time to have these discussions, it's going to make coding so much easier.

And you're going to have everyone's buy-in. So you have people rooting and cheering for you as you're working on this.

That's going to nullify all of this worry hopefully that you have about working in the open. And so, that went really well.

I did two cohorts of that and George Castro, who is standout, phenomenal, amazing person is going to be-

John: Legend.

Taylor: He's going to be picking up some of that. And this shouldn't be news to him if he hears this.

John: George, you're hearing it first here on the Open Source Ready podcast.

Taylor: Yeah, George, I just assigned this to you. No, he's got some idea, but that was great.

But that also created this discussion about something that the CNCF might release later for like merge to maintain, maintainer camp and kind of like these tracks to be able to more definitively join open source in a structured way.

'Cause that was another thing that we saw too, that I think is still not so great. Talk to any of these prominent open source contributors, people maintainers, and it's an individual story.

There's a lot of people around the edges of those stories. But it's typically that standup person that you hear about, right?

You don't hear about like, oh I did this, you know, accelerator and then my entire class is now part of my contributing community.

I lean on them, they're my friends. I called them up with hard questions.

I got this really cantankerous or prickly issue that opened up, I just dunno the respectful way to respond, right?

They're these very human things that it's difficult to talk about and you don't want to talk about out in the open, right? You can't process or buffer that.

And so you need people, you need people that have either gone through this or someone that has your same value system that you can riff with to be able to accomplish something.

So, that was a really valuable lesson to learn at the CNCF. And then at the latter part of my tenure at the CNCF, it was capitalizing on all of the momentum.

Like finally we have, I'm skipping over tons obviously, it's hard to condense three years of all the learnings into such a short amount of time.

But the end user technical advisory board, that was a huge win because we finally got this governing body together that represented end users rather than saying end users are a pillar of the CNCF.

That was true, but now there's actual people consistently that you can speak with beyond the CNCF staff that can go do the things rather than be the steward or be kind of impartial.

And the Switzerland-like entity. In Salt Lake City, the most recent KubeCon, the end user tab announced and launched the reference architecture website.

So now end users finally can publish what works for them in production, which I would say is even more important than a case study or an end user journey report.

'Cause it's like, this has been tested. This is what this organization uses. You can go check it out. You can try it. You can iterate on it.

You can not use it. This is might just be, you know, icebreaker conversation really. So, I think that was a huge win.

You know, I hope that there's going to be hundreds of examples on that front, but the reality is it takes time and again, not all end users kind of have that luxury of time.

We also brought back the end user tech radar. So, that was also really good with like AI ML, and then batch workloads.

Really helpful to be able to see that data and finally bring that back, so that we can't make these determinations as CNCF staff or LF staff.

But we can run a survey, collect data, and share out these findings that we have with people. We can observe what's actually going on, right? Much like we did today.

I can tell you anecdotally all the things, is that true? It may be, it might not be. But I hate giving the, "well, it depends" answer, 'cause it's not definitive enough for conversation.

So, lots of wins on that front. But I think what could have been better is I've worked in the community for a long time. I've led Kubernetes releases.

I started my open source journey probably around 2014 in earnest with like Ruby on Rails, and then Docker before getting into Kubernetes.

But never underestimate the amount of time something might take when you involve the community versus yourself. And never feel bad about that. I think that I expected that to happen, but in some cases, it's like you really do have to have that patience and empathy to wait for all the pieces to align, you know? 'Cause sometimes it moves way more slowly than you might anticipate. Sometimes it's way more quickly and you got to, you have to hustle to catch everything.

John: Yeah, absolutely incredible retrospective. And you know, your impact has been super clear throughout the years at the CNCF. But Taylor, now are you ready to do some reading?

Taylor: Oh my gosh, yeah. Of course.

John: Okay, so this week I had a pretty quick one that was just kind of a random late night read. I don't even know how I stumbled on this.

So it's titled, "I Tasted Honda's Spicy Rodent-Repelling Tape."

Taylor: Wow. And I know, what it's about is, you know, this kind of weird food blog and like the weird things that she finds to I guess putting food or eat.

And there's this tape apparently that has little mice on it and that Honda uses to like wrap wires, so that rodents won't chew it.

But you know, the primary ingredient of this tape is I believe some sort of like cayenne peppers or something.

So, it's a spicy tape and she uses it for a Bloody Mary, which is pretty hilarious. So if you want a quick laugh, this was a really good one.

Taylor: That reminds me of like the Nintendo Switch cartridges and stuff like that. They're supposed to be bitter, I haven't tried them yet, but you know, maybe I will in a dish.

John: Very tempting.

Brian: Tempted to see what Super Mario tastes like.

John: Yeah. Brian, what about you? What'd you have for a read?

Brian: Yeah, actually it's a book, and it's a recommendation I got from you John, which is "Hyperion," from Dan Simmons.

So honestly, I'd even, you're like, hey, if you're interested in sci-fi, 'cause I'd mentioned I was reading some sci-fi stuff, read this book.

So, I work in the library every Friday afternoon. I think it's like a nice break for being at home, working remote to go to the library.

And they've got desks, got WiFI. It's great. They've got a taco shop across the street and it's in Oakland. So the book was there and I'm like, oh, check it out.

But I like the story because it actually... These guys are on a ship called Hyperion, headed to a certain point to like see a grave site and there's like a historical reason on why they're doing that.

And everyone shares a bit of their story along the way. So if you're familiar with like old English Canterbury Tales, it's very much like that.

And I think like Canterbury Tales is actually personal to me or like freshman year in high school, I actually read that, 'cause it was required reading, but I actually really enjoyed it and really enjoyed the story parts.

And so like for me it's like, man, this is like special 'cause it's like the movie "Alien," but it's a book. And it's kind of like memorable to like how I read in my early years.

So if you like sci-fi, highly recommend it. It's actually a good read, great writing. And yeah, Dan Simmons, I'll definitely be checking out more of his stuff in the future.

Taylor: Interesting.

John: Yeah, he's one of my favorite authors. He does a bunch of like, it's kind of historical horror now.

The one I'm reading right now from him is called "The Terror." And it's about this like ship that gets like stuck in the ice, which really happened.

Like that's the historical part of it. And to this day, they've yet to really understand like what happened to this, you know, expedition.

But you know, in the Dan Simmons retelling, it's basically the abominable snowman comes and eats them all, it's very good.

Taylor: I like that, I don't think I've ever looked at like a historical horror. It feels like a fun genre.

John: Yeah, it's kind of a niche genre. 'Cause I actually really love like sailing novels, which I'm like becoming my dad.

But you know, it sprinkles in a bunch of kind of mystery and you know, other worldly like what is going on?

You know, but then he'll spend a whole chapter talking about like, you know, rigging the masses and, you know, doing all this sailing stuff.

So, a very niche one. But Taylor, what did you have this week for a read?

Taylor: This week? I have, it's, yeah, wow. We really just have a motley crew of reading this week-

John: Yeah, I love it.

Taylor: I have a technical paper and it's from Microsoft, and it talks about the impact of generative AI on critical thinking.

And I thought it was really interesting, because it's a conversation I've gotten to have with lots of folks. Is do you think this is making you better or worse?

There's lots of conversations about junior developers and you know, what their future looks like. What does being a senior plus principal engineer look like for teaching and the expectations?

So, I think it was a really... I really hope to see more research around this, but I think it's a fascinating read that kind of helps validate some of those anecdotal conversations that I've been having.

Not as spicy or as horrific, but maybe actually.

John: Yeah. So is the takeaway that, you know, using these gen AI tools, you know, cognitive effort goes down or that like the perceived like understanding of what is being generated goes down or what's kind of the hot take here?

Taylor: Exactly.

It's if you start leaning and kind of like relying more on your tools, that's not necessarily a bad thing. But if you don't understand it in the first place and you rely on it, then you've completely bypassed this moment of learning. So, you might introduce something into code base that works, but then it's time to fix it or it breaks and you're like, ah. And it ends up taking far more time than, you know, having done it the right way or not at all.

John: Yeah, this is kind of my like worst nightmare scenario for like where gen AI will take us.

'Cause like very clearly the bottom has sort of fallen out for like the new engineer pipeline, you know, and I don't believe that's because of gen AI.

You know, there's macro economic reasons, but, you know, what's the future of this going to be like when we're left with a lot of AI generated code, not very many junior or mid-level engineers to, you know, rise into senior positions to fix these things.

And then all the senior engineers have, you know, gone and started at the farm. And they're not interested in fixing gen AI code anymore, right?

Taylor: That's when you just give it back to the computers and they write the rest of the code for you.

John: Yeah, exactly. At one point, I swung very hard into gen AI tooling and I was using it for like, writing, and code stuff, and all this stuff.

And especially for writing, I was like I can't be using this for writing anymore.

Like it's doing a disservice to myself 'cause it's like I'm stealing that thought process of going through writing a blog post or writing a proposal or something, or writing my thoughts in an issue.

Like it just, it was too much where I was just like I don't even understand what, you know, the overall thought is anymore. And I can't craft that narrative.

So I've sort of, maybe the pendulum is swinging it back a little bit for me, but Taylor, you know, working at an AI company, where is your use of AI gen technologies?

Taylor: It's measured. It's similar to you in that I think I definitely leaned hard into a lot of those features.

I do think it's worth, you know, I've seen some people that they're like, ah, too much. And, you know, I've fallen into that category a little bit too, but I do think it's helpful to go try to use the tools, I do.

I'm bummed for folks that work at specific companies, big clouds and others that they're like, you can only use our tool. I think that's a reductive viewpoint and I heavily disagree with that.

I think you should go use all the things, because that's how you can understand what's useful, what's not, what you can adopt, what might be a good or bad idea.

And you can have your revelations that you had John, like I kind of want to still have that muscle, that mental muscle to be able to go write, or I feel like I'm overextending myself here or not getting, you know, leg day, arm day, you know, exercise it, you know, to make me complete as a person.

So there's a unfortunate, it depends, but you know, with what you use. But yeah, I'd say sparing, but I try to have a good context and use case for it.

If I can speed up my time and get it, you know, time is a thing I'm trying to optimize on that front.

John: Yeah, well, that's a great perspective. And I think with that, we'll finish it up.

Listeners, don't forget leg day and don't forget to stay ready.