Joe Beda: K8s, Culture, and Code That Survives Contact

Episode #23 - Joe Beda

===

[00:00:00]

Matt Klein: All right, folks. Welcome to another episode of Beyond the Noise: Signals, Stories, and Spicy Takes, the show where we dig into the stories of the people shaping the future of app-based computing. I'm your host, Matt Klein, co-founder and CTO of bitdrift, as well as the founder of Envoy Proxy. Each episode, we'll talk with engineers, founders, and technical leaders who've transformed the way their companies build and understand what's happening inside their systems.

We'll dig into the challenges, the breakthroughs, the lessons learned, and we'll wrap it all up with their hottest takes, so let's dive in. Today's guest needs no introduction at all. Uh, I'd like to welcome Joe Beda, who is a technologist with a history of working across many parts of our industry, from web browsers to real-time communication systems to cloud computing to AI.

He started his career at [00:01:00] Microsoft working on Internet Explorer, as well as client platforms, before moving on to Google. During his tenure at Google, he started, Google Compute Engine and helped start Kubernetes. He then co-founded Heptio, and after two years, sold it to VMware. Joe was at VMware for several years, helping to create the Tanzu suite of products.

After a break away from the keyboard, Joe has joined the startup StackLock as their CTO. StackLock is building a Kubernetes-hosted AI platform for enterprises to not just deploy AI, but to actually make it work for their business. Welcome, Joe. Thank you so much for joining us.

Joe Beda: Yeah, thank you so much for-

Matt Klein: Yeah

Joe Beda: having me on.

Matt Klein: Yeah. Um, we've known each other for a while. I'm very excited to chat with you. Um, you know, y- you've obviously had a, a very epic career, so we're not gonna get to dig into every piece. But where I, uh, typically like to start is just tell us a bit of your origin story. Tell us how you-

got into tech. Um, would love to hear a bit about the early years of Internet [00:02:00] Explorer. So just take us from there.

Joe Beda: Yeah, I'm a, a second-generation programmer, and my, uh, my youngest also is, uh, big into programming, so third generation, which, you know, uh, I kn- I know, you know, uh, Matt, you dabble in farming, so, you know, the generations go a lot further back Sure.

Right. Yeah ... there. But, you know, in our industry, I think that's a, that's a pretty storied pedigree. Um, my dad, uh, was an entrepreneur, worked on IBM mainframes and started a company, early on to, uh, add zip codes to mailing lists. This was back when the US Postal Service added zip codes, and you could get a discount on your postage if you actually used zip codes, and so it was a very simple ROI type of thing of like-

Matt Klein: Yeah

Joe Beda: you'd send them your, your mailing list, they'd put the zip code on, you'd save a lot of money, and so really easy, uh, sell there. And, um-

Matt Klein: Yeah, you had... I... You know, it- it's, it's funny. I, I think you've told me this story before, and I had forgotten. And I mean, for people that don't know, I think it's actually a [00:03:00] pretty a- awesome story because this is when they were introducing zip codes, right?

So w- was this like an incentive program to get people-

Joe Beda: Yeah

Matt Klein: ... to use zip codes? Is that the purpose?

Joe Beda: Yeah, I mean, like, a lot of the audience probably doesn't remember, but before the internet, you know, there was... You'd get all these catalogs in the mail, and so there was a ton of direct mail, um, because that was the only way to, to, to sell things short of, sort of physical brick-and-mortar.

And so, you know, it started back in, like, forever with, you know, the Sears catalog. But, but yeah, they introduced zip codes, and then the idea was that you'd get a discount on your postage if you're using zip codes versus not zip codes-

Matt Klein: Right

Joe Beda: ... because it was so much more efficient for them to deliver it that, that it made sense to do that.

And so then that created the incentive for companies to add zip codes, but how do you do that to a preexisting, you know, uh, mailing list? And this is obviously early computer stuff back when, you know, punch cards and stuff, right? So I remember he would work nights because that when he... That's when he would get time on the mainframes, and, uh, [00:04:00] and then, you know, we would always use blank punch cards as shopping lists where we'd write stuff down.

So yeah. So computers have always been sort of a part of my life one way or another.

Matt Klein: Uh, cool. All right. Well then, tell us a bit about your education, how you wound up at Microsoft.

Joe Beda: Yeah, so I went to, uh, let's see. For high school I went to the Illinois Math and Science Academy outside Chicago.

So this is a state-run, uh, boarding school focused on STEM back before it was called STEM. And, and then w- ended up going to, to college at Harvey Mudd in Southern California, and, uh, s- thought I was gonna study engineering. I was already a programmer and sort of, you know, like, you know, spent a ton of time, like, you know, messing around with that stuff, um, as a hobby.

But then, like, when I went to college, I'm like, "No, I'm gonna do something different. I'm not gonna, like, do what, you know, what my dad did." And, um, just turns out everything else was too hard, so just sort of fell into you know, old patterns. And so, yeah. So did a computer science degree at Mudd. Um, [00:05:00] did a couple of internships at Microsoft, uh, and working on Office on this old dead product called Team Manager, which was essentially a way for sort of employees to report what they're working on and track usage and stuff.

Um, you know-

Matt Klein: And, and-

Joe Beda: ... time tracking.

Matt Klein: Right. And, and, like, just for time reference, this I'm assuming would've been the early '90s? It, it- Yeah, '

Joe Beda: 96 ... is that- Or mid-'90s ...

Matt Klein: is that... Okay. Yeah. Mid, mid, mid-'90s. So-

Joe Beda: Yeah

Matt Klein: ... so, so you were there during, at least you started your career during the boom times, I mean, of, of w- when, when Microsoft was-

Joe Beda: Yeah

Matt Klein: was,

was really tak-

Joe Beda: I joined Microsoft after college ...

Matt Klein: taking off.

Joe Beda: Yeah ... and, yeah. And it was definitely, like, like, after Windows 95 when things were really exploding. It was a weird time to be in the industry 'cause this was, you know, i- heading into the, the dot com, uh, uh, bubble, boom, whatever you wanna call it.

And, you know, the... you know, in terms of financials, there's like this sort of formative experience. It's like, 'cause when I joined Microsoft, I got, I think, like, [00:06:00] my, my initial... Like, I was paid, like, $45,000 a year, and my initial grant was 12,000 shares of Microsoft at the time. Um-

Matt Klein: You actually know that. That, that

Joe Beda: is- I, I somehow that is burned in my brain

that,

Matt Klein: that is very impressive. Wow.

Joe Beda: Right? And so, and I was, and I was, I was offered a choice between joining the NT team and the Internet Explorer team. And, uh, and I decided to go with IE because the NT team was full of, like, these old people, and I'm like, "I don't know where this internet stuff is going," but, like, it's definitely gonna be...

There's gonna be a lot more mo- movement and action, and so that's gonna be much more, you know, uh, uh, a better place for my career to be someplace dynamic versus someplace that's relatively... calmer. But yeah. But so, like, but then I was-- And when I joined Microsoft, there was a certain stretch there heading up into 2000 where it was like every 18 months, the stock doubled and split.

Yeah. And it just felt like this, like, Moore's law, law of nature, um, that this would happen. And, and the psychology there, you know, I think, you know, this is not really career, [00:07:00] but, like, when you see that sort of steady rise, it just becomes this sort of rule of nature. And then to see the decline of Microsoft stock- Yeah, for sure

'cause it took it, like, 15 years to recover after-

Matt Klein: Yeah

Joe Beda: ... the, the dot-com bubble, um, that was a formative experience for me, sort of on a financial sort of literacy type of thing of like, you know, you cannot predict what's gonna happen because your psychology plays-

Matt Klein: Absolutely.

Joe Beda: Yeah ... such tricks at you as this stuff happens.

And so now people are like, "Oh, such and such is gonna happen," and I'm just in here going like, "Yeah, I have no idea what's gonna happen."

Matt Klein: Yeah. I mean-

Joe Beda: Whatever it's gonna be, I'm not gonna predict it.

Matt Klein: Right. Yeah. We will, we will get there for sure. I, I think there are some parallels in terms of what is going on right now.

But I, I, you know, I think w- what is interesting, to me at least, is you were there. I mean, again, like most people who are listening to this podcast d- don't understand or remember the fact that software used to be shipped on physical- ... CD-ROMs and, like, floppy disks and stuff like that. And I think what's interesting about when you started in the industry, which is, which is a [00:08:00] bit before me, is you were really there for that transition from the, you know, you're shipping a thing in a box, like you go to Newegg, you know, y- you, you have your CD-ROM, right?

And, um, that CD-ROM, if you get it wrong, is, it's like big cost, you know? Like, versus-

Joe Beda: Yeah ...

Matt Klein: the, the way that we operate today. And I think-

Joe Beda: So at Microsoft, whenever you were shipping, they called it RTM-

Matt Klein: Yeah

Joe Beda: ... uh, reverse-

Matt Klein: Yeah

Joe Beda: ... uh, release to manufacturing, right? And so, and, and IE, you know, around IE4 was the time when I think I still have buried somewhere, I just moved, I have in a box someplace an old IE, uh, CD.

But, um, but they were still shipping out physical media, 'cause this was like when AOL was carpeting the world-

Matt Klein: Yeah

Joe Beda: ... with AOL CDs and all that.

Matt Klein: Yes, I remember. Yeah.

Joe Beda: But it was definitely shifting, and we started talking about RTW, release to web, as sort of being the, the alternative to RTM.

Matt Klein: Yeah. Um, I mean, and I would imagine that during that era with the web [00:09:00] moving so quickly, I mean, with, uh, different standards and all of those things, I mean, what, what, what was it like to be a new programmer and w- work on the browser?

I mean, I'm sure you have some, some pretty interesting stories from that era.

Joe Beda: It was, it was a strange place to be. And I think, you know, one of the things that we forget, or at least as we get more senior, and I don't know if this has happened with you, Matt, is that you kinda understand how the company and the world and the industry works.

And so there's a certain level of like, like y- you have a deeper understanding of the systems that you're working in, right? I'm, I'm fundamentally a systems person, both like technical systems but also people systems, and a company is a system. And, um, but when you're starting out in the industry, you know, you're, you know, plugged into some team with a manager that has like, you know, maybe a couple other reports at the time.

The teams were relatively small at Microsoft. Um, and like, you know, maybe 10 levels between you and Bill Gates. And, you know, you really don't have that sort of larger view of how you fit into the world, right?

Matt Klein: Sure.

Joe Beda: And [00:10:00] so it was both an exciting time because there was a lot happening, um, but it was also confusing because it's like I, I, I was not privy to...

And you can read books about sort of early IE-

Matt Klein: Yeah

Joe Beda: ... but I was just sort of like junior person on the team. I wasn't privy to a lot of the stuff that was going on. And then especially through the, the, the antitrust trials and all that, there was a real lot of cognitive dissonance there because I was sort of so low.

I'm like, "I like the people I work with. These are good people. Why are we being vilified by so many people in the outside world?" Without understanding, you know, the, the, uh, you know, the, the layers and layers and layers of stuff that's on top of you there. So, so on a sort of, company, sort of career social level, that was, um, tricky to navigate.

And then on a technical level, you're just drinking from the fire hose figuring out what it means to actually build production software. That's not something that, you know, was taught in schools at that time. Like, even when I was in, in college, like source control was not something that you learned in college.

Matt Klein: Of course. Yeah.

Joe Beda: Um, and [00:11:00] so that was all sort of like skills that, you know, you pick up on the fly during the, those early times. So yeah, it was a really weird time to actually... or interesting time to be pr- uh, join in the industry.

Matt Klein: Well, for sure. And you know, I, uh- I also started my career at Microsoft a bit after you.

I was in the early 2000s a- after the, the boom and the bust had happened. But o- one thing that, that I think about often that I would love to get your opinion on is even though I also moved on from Microsoft after a few years, I actually feel like I, I, I learned a lot, like, during that during that formative experience and, like, prior to us talking about your stint at Google and other things that you've done.

Um, I, I don't know. I mean, do you, do you think that starting your career at Microsoft y- you know, was formative for you? I mean, it's like... Because I see pros and cons of actually doing that, and for me, I actually think it was a net positive. But I'm curious to get your, your thoughts on that versus having done something else.

Joe Beda: [00:12:00] Um, yeah. I mean, in a lot of ways i- is, uh, you can't help but soak in sort of the, the... sort of the value functions of the organization that you land in, especially early in your career. And so Microsoft, more than every other company that I've ever worked at, is a platform company. They understand the power of platforms in a way that, that very few other people do.

Um, both developer platforms, but also things like Office. The, the idea that, you know, Office and Windows sort of reinforced each other. There was sort of like systems-level thinking around how the market dynamics work and how platforms actually create long-term value that Microsoft understood in a very, very, very deep way.

Matt Klein: Yep.

Joe Beda: I think if you look at something like the developer division at Microsoft, which has been spun in and out and twisted and reorged a gazillion times, I think it maybe has broken even from a revenue point of view across the history of Microsoft. Um, uh, you know, maybe with AI and ties to Azure, it's maybe making money, but it was [00:13:00] always seen as a bit of a, you know, supporting sort of loss leader for the rest of the platform.

Matt Klein: Sure.

Joe Beda: Whereas this was at a time when people were charging hundreds of dollars for compilers, and Microsoft really understood the power of, of, of making it easy and cheaper to build on their platforms. And so that has definitely stuck with me throughout my career, and it's, it's something that it's, it's hard to sort of set that view aside once you really internalize it.

Matt Klein: For sure. Yeah. Okay, cool. Well, I mean, you obviously worked for Microsoft for a number of years. Tell us a, a bit about why you ultimately decided to leave and, and how you wound up at Google.

Joe Beda: I think it was, it was career velocity and stuff, um, you know, was a big driver. I stayed through IE, transitioned into working on what eventually became Vista, um, uh, got more leadership experience.

But I looked around and I'm like, "This isn't going anywhere fast." It definitely felt like things were slowing down. And the way that Microsoft worked at the time is that, you know, you'd ship a big product like a version of Windows, and then there'd be a game of musical chairs as everybody moved and got promoted, and there was more [00:14:00] opportunity.

And so the slower that they shipped, and Vista was taking forever, the slower sort of career advancement there was. And at the same time, a bunch of old colleagues from Internet Explorer, uh, joined this, like, really crazy company startup called Google and were starting the office here in Seattle in the, the Kirkland office.

And, uh, decided to join them having no idea what I was gonna be working on. And, uh, that was, like, also a very crazy time to, to join Google.

Matt Klein: Yeah, for sure.

Joe Beda: But this was around 2004, right around the IPO is when I, when I made that jump. And this was when, you know, when Ballmer was throwing chairs and all that, and there was an enormous, you know, amount of bad blood between Microsoft and Google at this time.

Matt Klein: Yeah, I mean, and, uh, I have, I have never worked at Google myself. I, I have been exposed to Google a lot. I mean-

Joe Beda: You kind of feel like you have,

don't

you, Matt?

Matt Klein: I, I know, I know. Um- ... but, y- y- you know, what I'm, what I'm most interested in is, I mean, at, at least from my perception, the, the, [00:15:00] the culture of these large companies.

I, I, I think what a lot of people don't understand is that, you know, these are all mega, like, successful companies. But the cultures at all of these companies, like whether it be Google or Microsoft or Apple or any one of these huge... They're really different. Like, they're-

Joe Beda: Oh, yeah

Matt Klein: ... different places. And it's like, I feel like in many ways Google is almost the polar opposite of Microsoft.

So, you know, what, what was it like to both be exposed to that culture difference, but also obviously Google was born out of the internet era, was born out of web software. Like, we talked a bit about the shrink-wrap CDs and all this.

Joe Beda: Yeah.

Matt Klein: Like, Google never lived in that world. Like, Google was a new company, you know, that never, never did that.

It was all about internet services. Um, so I don't know, just like talk a bit about what it was like. Was it, was it culture shock? Was it easy for you? I mean, like, what, uh, how did it feel?

Joe Beda: I

mean, there was definitely some culture shock there.

Matt Klein: Yeah.

Joe Beda: So, you know, the development environment for NT, which I [00:16:00] was working on, the sort of Windows NT, was called Razzle, and it was sort of a set of, of, um, uh, shell scripts or whatever passed for a shell on Windows at the time.

Um- to sort of set up your development environment. And I was so used to working with that, that when I got to Google, I sort of kind of recreated something called Razzleish type of thing in my, like, Unix environment as I was programming at Google. But I think on a cultural level, it's very different also.

I, you know, I think that company cultures are often defined by a caricature of the character flaws of the founders, right? So when you were at Microsoft and you went into a review with Bill Gates and, and I never had the pleasure, but you hear these stories, um, you know, he was just hyperaggressive. Um, you know, "That's the stupidest effing thing I've ever heard.

How do you not drown in the rain?" Like, "Why do you still work here?" Right? Just, like, really brutal. And you would see all the lieutenants sort of try to out-Bill Bill-

Matt Klein: Sure

Joe Beda: ... and sort of created this super hyperaggressive culture inside of, inside of Microsoft. [00:17:00] Whereas Google, Larry and Sergey were just fundamentally passive-aggressive people.

Like, they'd be in a big review and, like, get bored and start throwing bits of paper at each other, and then, like, Eric Schmidt would be like, "Guys, you gotta pay attention. This is important," right? And it's just a super weird dynamic. And, um, and I had enough time and sort of the people around me to be able to adapt to that.

But we would see leaders come in from Microsoft, go to Google And like, you know, in a, in a, in a Microsoft, if you have a bad idea, somebody'd be like, "That's the stupidest thing I've ever heard," where at Google they would just ignore you and hope you'd go away. And so these people would come over and they'd be doing stuff and everybody...

And they'd be thinking that they're killing it until the rubber meets the road and they need to ship or they need to like get resources or you know, support in certain ways and then they'd find out that nobody wanted what they were doing and they were kind of dead in the water. And so I'd see that happen multiple times as, as people sort of transition between these cultures.

Matt Klein: Yeah, but at the same time, I mean, for both of these companies really, they're making so much money in their [00:18:00] primary businesses. At the time for Microsoft it would be Windows. Obviously Google started their ads business and started making an ungodly amount of money. And I don't know, it's like I think what's interesting about these companies is it certainly gives some air cover for a lot of people to maybe do stuff for a very long time that might not be- Yeah

the most productive thing. I mean, it's like, and I, I think that probably applies to both companies. Um, so I don't know. I mean, it's a, a, um-

Joe Beda: I, I think, you know, there's always this push and pull between staying on task at a company versus giving engineers room to experiment.

Matt Klein: Sure. Yeah.

Joe Beda: And like does innovation come from the top down or from the bottom up?

And I think, um, Google definitely believed in bottom up innovation and I think, you know, people don't talk about it now, but like, you know, 20% time was a thing at Google-

Matt Klein: Yeah

Joe Beda: ... where this, it was this idea that you'd spend, uh, roughly 20% of your time doing something that wasn't your core job. Now I always took that to mean something that could still benefit the company in some way, so I like, like for me my like work ethic, I wanted to see it related to the [00:19:00] business.

But there were people that would do like underwater basket weaving type stuff-

Matt Klein: Mm-hmm

Joe Beda: ... um, you know, on Google's dime and call it 20% time. Um, I, there's a blog post I wanna write called The New 20% Time about how with AI and agents there's actually room to actually do experimentation, but it's a different-

Matt Klein: Yeah

Joe Beda: way to approach it because you're sort of time slicing with multiple agents.

Matt Klein: Yeah.

Joe Beda: But that's, that's something we can talk about later or I'll-

Matt Klein: Yeah, sure

Joe Beda: ... I'll write something about it.

Matt Klein: Yeah. Um, okay, so when, when you were at Google o- obviously you worked on some different stuff. H- how did you wind up getting involved with Google Cloud?

Joe Beda: Um, so twisty path. So, um, started off working on Google Talk, which was the, you know, essentially Google's chat product. And at the time people had-

Matt Klein: Sorry, you say Google's

chat product. There's been, like, like 45

of them.

Joe Beda: It was the first one.

Matt Klein: Oh, the first.

Joe Beda: It was the first one.

Matt Klein: The first, okay.

Joe Beda: Um, it was built on XMPP.

There was a Windows client. We eventually, uh, uh, merged it into Gmail, and it became Gmail Chat, um, through Google Plus, and this was after... You know, I, I did a bunch of work to hook [00:20:00] this up to the telephone system, but then there was the Grand Central acquisition, and there was a lot of conflict there. And so, uh, but that eventually turned into...

Let's see. Their, their, the P2P stack that we had in the Windows client eventually became libjingle, which is the underlying library for WebRTC, which is actually recording what we're doing now. Uh, and then, and then eventually sort of, you know, through Google Plus turned into Hangouts. So you never know where these things-

Matt Klein: Yeah, of course

Joe Beda: will actually take you.

Matt Klein: Yeah, for sure.

Joe Beda: Um, which is always fascinating. Um, and then did some machine learning stuff, some ads optimization, uh, which I felt like, you know, I had to pay my dues a little bit and do something in ads. And then we had our second kid, and I came back to work after paternity leave, and I just could not make myself give a shit about ads anymore.

I just, like, I could not care. Um, and there was a director in the Seattle office that wanted to do systems types projects, and so we looked around and we're like, "Hey, there's these sort of, like, super cheap laptops that are coming out." They were called netbooks at the time. And [00:21:00] we're like, "Could we do a sort of internet native operating system for these things?"

And so we started what would kind of sort of eventually become ChromeOS-

Matt Klein: Yeah

Joe Beda: ... um, with Chromebooks. But there were some ideas with the Chrome team down in Mountain View, and so there was sort of company politics, and we kind of got frozen out of that. And so we're like, "All right, what are we gonna do now?"

And so we literally called the project Plan B, and it ended up being infrastructure cloud, what eventually became GCP. Um, and so, uh, started out, and the f- the first three projects there were, um, uh, we called it, uh, BigStore, which became Google Cloud Storage. BigQuery, which was essentially an internal system.

Dremel, which was already almost a service-

Matt Klein: Yeah

Joe Beda: ... so it was unique in a short distance. And then I'm like, "If we're not doing virtual machines, we're not really in the game," and so started a project called BigCluster. Horrible internal name, right? And so that eventually became Compute Engine.

Matt Klein: Uh, it, uh, I mean, this is probably what?

Like 2007, 2008, something along those lines, right? I mean, or 2000-

Joe Beda: Yeah, [00:22:00] something like that. Yeah.

Matt Klein: Yeah. And, and I mean, how much of this was a response to what you were seeing across the, the 520 bridge at Amazon, right? Like, because obviously they, they launched EC2 in 2006. Um, I, I mean, I guess what I'm asking is was it clear to people that that was gonna be huge and that there had to be a response, or was it hard to convince people that this was gonna be a real thing?

Joe Beda: Um, to me it was clearly like there was something happening there. Um, and I think it was both a response to seeing what was happening with AWS, but also seeing Google's internal systems and sort of how you can build something really cool and useful with things like Borg and what have you. But there was a lot of resistance inside the company.

Matt Klein: Mm-hmm.

Joe Beda: There was-- At that time, Google could not build data centers fast enough, kinda similar to, to today. And there was a little bit of thinking of like, "Hey, if, if we're gonna have a new machine, uh, in a data center, we'd [00:23:00] rather use that to build the next Gmail that's gonna be higher value versus just rent it out for parts."

And I don't think Google realized how lucrative the sort of, you know, arbitrage around sort of like what they can build data centers for versus what they can sell it for cloud could actually be. And they didn't necessarily have the skills to do it correctly. I mean, fundamentally, Amazon is a logistics company.

Matt Klein: Yeah.

Joe Beda: And, um, and being successful at cloud is, is largely a l-logistics problem, having the right number of machines at the right time, because any machine that you're not selling hours for is essentially like bananas going bad in the grocery store, right?

Matt Klein: Yes.

Joe Beda: It's spoiled produce. Um, and so, it took a while for Google to really come to the realization that cloud was a real business.

And in the meanwhile, we built stuff anyways, and this was Google's passive-aggressive nature benefiting us. We were up in Seattle. Nobody gave a crap about us. They just ignored us and hoped we'd go away. And it turned out that, you know, we didn't go away, and they actually needed what we were building, and so that's kinda why cloud turned into, to what it turned into.

Matt Klein: Yeah. I, I, I mean, it's like I think that few would [00:24:00] dispute that a-at least during that era, Google's internal systems were b-by far the best in the world, I mean, in terms of building web-scale businesses. I mean, most of the formative research that has-- or, you know, some of the formative theories that have come out and the systems have all been built at Google.

And, you know, we're probably far enough removed that I can ask you this question and, and get your honest take, 'cause I find it fascinating j-just how there are so many instances of the company that probably should have been able to do something the best, right? It's like at least on a pure technical basis, Google had the technology and the people to basically be AWS.

But I would argue from a cultural perspective, it, I mean, it, it just didn't. And, and I have my own theories on why that is, but I would love to get your, your thoughts on how, why, you know, like, what if, what actually happened back then?

Joe Beda: [00:25:00] I,

I think Google is fundamentally a consumer company, and I think, uh, you know, and that, that comes at you a lot of ways, and it also is a company that believes in automation in a deep way, even before AI.

And so I think Google had these armies of people that would essentially be ad account executives that would actually partner up with people selling ads, and there was always a, I think, a, a desire to eliminate that position and automate it and sort of m- they, they always begrudged the human cost.

Whereas I think Amazon has never been afraid, whether we're talking about warehouses or people on call, to sort of throw people into the grinding machine to actually make stuff work. Um-

Matt Klein: Absolutely.

Joe Beda: And so you'd see that like, you know, being successful in, in a cloud is, is fundamentally working with enterprises because those are the ones with the big bucks.

And, um, and Urs was the head of TI, technical infrastructure, the opposite of IT. Um, and, uh, uh, and cloud ended up sort [00:26:00] of rolling up through him. And mostly this was actually servicing internal customers at Google, but this was kind of one of-- gonna be one of the first sort of external products coming from this group.

Urs was running it. For a while, Urs also... This is Urs Hölzle, who's like, I don't know, he was like early employee at, at, at Google. Um, still there, but working from New Zealand even though he thinks that nobody else should be allowed to work remote. Um And, uh, so for a while he actually managed Google Apps essentially like, you know, uh, uh, you know, Docs and Gmail and all that stuff in the, in the corporate account there.

And they did a big deal with the, I believe it was the LA sheriffs or LA government, LA, LA police, something like that in the LA area. And, and it hit into one of those situations with enterprise sales where they think you work for them, and they just keep asking for more and more changes and more and more special stuff.

And it's just, that's just normal doing enterprise sales. At some point you're gonna hit somebody-

Matt Klein: Yeah

Joe Beda: ... who wants you to be their sort of externalized IT department and build stuff custom for them. The [00:27:00] lesson that Urs took away from that was, we're not gonna do any customization. We're not gonna do any sort of ass-kissing that's actually necessary for enterprise sales.

And so what you would end up with is that with Google Cloud, like for the longest time, I wanted to build a product for GCE that was essentially a, um, a, a scalable Google quality shared file system mounted over NFS. 'Cause I'm like, if we can solve this and actually up the scaling levels by 10X and make it fast in performance, I knew that we could sell the shit out of this.

'Cause we were visiting a, a company that was... It was down in Hollywood doing rendering. And, and I, I... As we were visiting them, I was exiting the building, I saw a literal pallet of, of filers, right? Yeah. And I'm like, "If these guys are buying filers by the pallet, then they are addicted to this, and there's a lot of other companies that are addicted to this architectural style.

We should build a product for that." I pitched that internally, and there'd been multiple false starts inside of Google, and they're like, "No, if you're doing a shared file system like that with a POSIX, shared POSIX file [00:28:00] system, you're doing it the wrong way. Customers shouldn't be doing that." And that is the attitude of like, I saw a pile of filers.

Matt Klein: I hear it.

Joe Beda: I know there's a market there.

Matt Klein: Yeah.

Joe Beda: And they're like, "No, they should rearchitect their applications. They're doing it the wrong way." And I'm just like, "You guys are useless," right? Like, like you're just... There's no hope. Now, eventually, Thomas Kurian came in and actually sort of whipped things into shape and actually said, "No, you will actually do things that customers want."

But it took them a long time to get there, and I think they really lost a lot of, uh, momentum early on-

Matt Klein: Yeah

Joe Beda: ... because of it.

Matt Klein: I, I, I mean, that, I, I think that actually brings us back to what we were talking about, just about company culture.

Joe Beda: Yeah.

Matt Klein: And I, I, I think that in that era... I mean, again, like no one would dispute that as of this year, like 2026 Amazon and, you know, they, I mean, they have built a, a ton of incredibly scalable technology.

I, I mean, you know, I- they, they've done absolutely amazing engineering. When I joined EC2 in 2010, I mean, the level of shitshow bash scripting, like duct tape, [00:29:00] you know, that I witnessed-

Joe Beda: I mean,

that's Amazon engineering right there.

Matt Klein: No, I understand. But, but at the same time-

Joe Beda: But it

works ...

Matt Klein: and it was shocking to me coming from Microsoft, but holy shit, I mean, it was, it was effective.

I mean, and they were just laser focused on delivering what customers asked for and not shoving their internal views down people's throats. And that's the part of this that I find so fascinating is that at the time, Google absolutely had better engineering, better technology, better everything. I mean, but, but that doesn't matter in the end.

Like, if you're, if you're not delivering-

Joe Beda: I mean, the original data centers that Microsoft built compared to-

Matt Klein: Right

Joe Beda: ... like Google data centers at the time, I mean, Google was building its own bespoke Clos fabric network switches that-

Matt Klein: Right. Yeah

Joe Beda: ... you know, and they were on the third generation by the time Microsoft even started building data centers.

Matt Klein: Yep. Yeah. I mean, so I don't know. I, I, I just like, I find these stories to be so fascinating because a lot of times in our industry, we talk a [00:30:00] lot about technology and cool things, and it's like at the end of the day, it's a small part of what makes for a successful-

Joe Beda: Yeah

Matt Klein: ... business. Um-

Joe Beda: I, I think there's also a lesson there from that is that, you know, the sort of pile of bash scripts and, you know, chewing gum and bailing wire and duct tape is that I think we both, I, I'm sure you have this sentiment, that like large systems, like you can design them well and that matters, but at the end of the day, you have to beat them into submission.

You have to battle test them, you have to fix the bugs.

Matt Klein: Yep.

Joe Beda: There's always gonna be stuff that you didn't foresee and, and there's, you know... And in the AI era, th- it's one of these things where it's like to some degree, you know, it doesn't matter how good the code is, if it works, it works.

Matt Klein: Yep.

Joe Beda: And if you can actually sort of stay on top of it and beat it into submission, then you actually end up with a usable system.

And so, you know, elegance is not always the, the, the thing that matters. Sometimes-

Matt Klein: Nope

Joe Beda: ... it's just, you know-

Matt Klein: Yeah

Joe Beda: ... miles on the odometer.

Matt Klein: All right. Um, tell us briefly about the, I'm sure, large internal fight of, of [00:31:00] starting up Kubernetes. So what was, what was that like?

Joe Beda: Well, so at that point, cloud was starting to grow and it was getting stifling and, you know, and this is another sort of aspect of Google's culture is that they were obviously struggling with cloud and they're like, "Oh, the way to solve this is to bring in external leaders."

And so they would have this, this cavalcade of people coming in externally and they'd want to put their mark on stuff and you would end up seeing more and more process and it was stifling for me. And so there were a bunch of us that weren't having a good time and, uh, uh, and I know Brendan, one of the co-founders of Kubernetes, would have a daily meter of like how close he was to quitting, like on a scale of one to 10.

And, and so like part of our motivation was we want to do something fun where we can create this bubble where we can write code and, and, and build something and not deal with all the sort of like new layers of process and bullshit that was coming in with cloud. At the same time, it definitely felt from a, uh, a competitive point of view that we needed to change the game, right?

Just sort of fast [00:32:00] following Amazon, I don't know how fast we were, wasn't gonna get us there.

Matt Klein: Yep.

Joe Beda: We needed to do something different and the experience of using containers sort of like higher level constructs of Borg inside of, of Google was, was pretty magical at the time. And then Docker came out and you're like, "Oh, we're taking all the same technologies that Google was using on, on the node," but they packaged it up in a super easy to use way and you're like, "This is solving like a big part of the problem.

We see what the rest of it looks like. Let's go ahead and build it." Um, but doing that as open source was very... was seen as very risky, because Borg was seen as sort of this secret sauce of like... And, you know, and at the same time Mesos was coming out. Like, anytime you see these things happen in companies, people move around, there's th- these diasporas, and it's like these technologies never stay contained in a single company.

And so we eventually convinced Urs to actually let us do it as an open source project, and that was unique because a lot of the sort of previous sharing of technology from Google, whether it be, uh, Bigtable or GFS-

Matt Klein: Yeah

Joe Beda: ... or MapReduce, had been in the [00:33:00] form of academic papers.

Matt Klein: Papers, sure.

Joe Beda: And then you would see that, like, there would be communities formed around these things, whether it be the Netflix OSS stuff or whether it be Hadoop, and they would just do it in a slightly different way.

And so there was innovation that took things beyond what Google would do, and there was zero way for Google to re-import that expertise and that, that, that code back in.

Matt Klein: Yeah.

Joe Beda: And so we thought, like, well, let's do it as open source, and that way we can stay involved and we can actually benefit from releasing this stuff into the wild.

And, um, and there were a lot of people who thought it was a horrible idea. They didn't get it. Um, they're like, you know, "Why don't you just use App Engine? Why don't you just use VMs?" Just understanding sort of the-

Matt Klein: Yeah

Joe Beda: ... that sort of middle layer and how that would appeal to people was not obvious to a lot of folks.

I know one product manager marketer guy was like, you know, right after we launched and there was a lot of excitement, was like, "I know there's a unicorn in there somewhere because I see all the unicorn shit, but I don't know where it is," right? So a lot of really sort of like derisive ways of looking at this.

Matt Klein: Yeah. I mean, we could spend the rest of the [00:34:00] time talking about- Sure ... the cloud-native era. Um, uh, and, you know, obviously you, uh, you know, for I'm sure everyone knows, you obviously s- spun Kubernetes out into Heptio and had some good, ha- had some good success there. Um, I, I, I do wanna spend the rest of the time talking a bit about your, about your retirement.

Okay. Um, so I don't know, you know, te- tell us a bit about, um, I guess, what, why you decided to leave tech, and I guess more importantly, what, what has brought you back?

Joe Beda: Yeah. I think well, you know, so both leaving Google and then leaving VMware I was, I was burned out on both of those things and sort of like after leaving Google, uh, I took about a year and a half off and sort of, kinda worked with Didno as an entrepreneur-in-residence, which means you just sort of pal around with VCs for a bit and get to know each other, um, on their dime.

Uh, which is not a bad gig if you can get it. Um, and so there, there's different things that brought me back in. So after that first sort of, you know, retirement, you know, my dad being an entrepreneur, I always wanted to do [00:35:00] that, so I always had the bug. And with Kubernetes sort of on the ascent, uh, uh, I'm like, "If, if I'm not gonna do it now, when am I gonna do it?"

Right? And so it was really about sort of a bucket list item of like, "Hey, I wanna, I wanna do a startup," and we ended up being phenomenally successful, got acquired after a couple of years- Yep ... um, which is kind of unheard of. And, um, and then being at VMware was also very interesting because it was a very different company than, than Google or Microsoft, and, and a lot of lessons there that we could go into.

Matt Klein: Yep.

Joe Beda: Um, but then, uh, you know, but there was a lot of stressful change there. The CEO that brought us in, Pat Gelsinger, left to go run Intel, so we lost some sponsorship there. Um, VMware bought Pivotal and tried to merge it in with us. Very different culture there, like, uh, lots of stories around that.

And being part of that merge and trying to find a way to sort of bring those cultures together was incredibly stressful. Plus, there was just a lot of people inside of VMware that didn't get it, and so I was burned out, and I'm just like, I, I... And for me, burnout is a [00:36:00] driver of... feeling responsible for something but not having the tools to impact it.

So you feel like you're pushing a rock that's just never gonna move. And, and, and I just, you know, the, for me, I just had to stop. And so, um-

Matt Klein: Well, uh, uh, yeah, I mean, I, you know, when you, when you do a startup, at least in the beginning, I mean, you're... The, the amount of change, the amount of things that you can do on an almost daily basis is absolutely huge, and I, I have not myself had this experience, but I would imagine that going very quickly from that type of environment to, like, an almost bureaucratic quagmire, I mean, that, that would be, that would be very difficult, you know?

Joe Beda: I mean, I think it's, it's, it's... So when you're at a startup and you bring on a new employee, it's so empowering 'cause you're like, "Holy shit, we have another s- cylinder in the engine. We can go faster." And so there is this sort of shared fate, this shared sense of purpose of like, we're all in this together, we work together, and, and we all sort of, you know, [00:37:00] succeed and fail together.

Um, at some point, companies get big enough where it becomes more personally efficient to play against other employees-

Matt Klein: Sure

Joe Beda: ... than to actually-

Matt Klein: Yeah

Joe Beda: ... you know. So, so it becomes a zero-sum game inside the company. And that's always sort of a depressing thing to see. And I saw some of that happen at Google as Google grew, right?

The, the sort of the zero-sum thinking was on the ascent. And, um, yeah. And it's a very different environment, and it... And you get stuck in the middle, where it's like you don't have that control. Not everybody is, is, you know, uh, uh, uh, aligned with the same sort of motivations.

Matt Klein: Yeah.

Joe Beda: And it's, it's just, it's, it's, it's kinda depressing, really.

Matt Klein: Yeah. Okay. Well, tell us a bit about what happened then. So obviously you, you left VMware, took some time off, and n- now you're back. So what-

Joe Beda: Yeah

Matt Klein: ... what brought you back?

Joe Beda: Um, so I mean, uh, my, uh, both my parents passed and so dealt with that. Had a, had a good friend die, you know. We're, built a new house that I'm in now.

So there was a lot of [00:38:00] stuff going on, and, um- And I left the industry right around when ChatGPT launched. And so I missed out on, on a lot of the sort of early AI stuff and, uh, got to the point where things were calming down. My wife wasn't ready to quit and sort of goof off with me. And so, you know, kicking around the house, my wife and my youngest kid were like, "You need to go do something.

You're bugging us." And, and I also sort of self-identify even after all this time as, as an engineer, uh, both a businessperson and an engineer. And looking at where the industry is now, I'm like, I, I, I can't call myself an engineer unless I actually keep up with this stuff. And I would like to think I'm the type of person that can be motivated by having personal projects and sort of get up and sort of spend several hours a day, like, self-motivated.

But like, you know, three or four years of being semi-retired has shown I- that's just not me. I tend to sit around and watch YouTube and goof off. And so without that [00:39:00] external structure, it was very hard for me to, to really take this stuff seriously and immerse myself into it. And so, um, I joined Craig, my co-founder, with Heptio, uh, his company that he'd been running for a while.

And, um, you know, I'm having a great time just sort of learning a ton- a bunch of stuff, working with some great people, and that's what I was looking for is, is, you know, good people to work with, learning a bunch of fun stuff, um, you know, trying to, to make it happen again.

Matt Klein: Yeah, I mean, tell us a bit about what you're up to.

Like, what does the company do? Uh, tell us, are you a Claude Code fan or a Codex fan? Like, tell us a bit j-just about what-

Joe Beda: Mostly been using Claude Code

Matt Klein: ... what you've learned. Yeah. Yeah.

Joe Beda: Um, the, uh, uh, the company, started off doing supply chain security, uh, with some open source projects. Um, this stuff is still important, but the industry changed so much with AI-

Matt Klein: Yeah

Joe Beda: that a lot of the original approach wasn't really viable anymore, combined with the fact that, you know, some of it was this idea of building a community around [00:40:00] security products, and it turns out nobody gives a shit about security. Like, they, like, they... You know, nobody builds a security home lab. I'm sure there's people out there that do, but it's not nearly the same type of thing as like- I mean, like-

you know, the Kubernetes community.

Matt Klein: No, no one cares until they care.

Joe Beda: Until they do,

Matt Klein: yeah. Until they care, but that I agree. It's a, it's a, it's a very difficult situation.

Joe Beda: It's a

very different dynamic around community and company building.

Matt Klein: Right. Right.

Joe Beda: So, you know, so with, with everything that happened, you know, before I joined, they started, you know, pivoting towards AI, and I think we saw a lot of companies do that, so reinvented themselves around MCP.

Matt Klein: Right.

Joe Beda: Um, and, uh, uh, with, with a few other things sort of, you know, that happened along the way around MCP governance. And so a lot of the sort of thinking there is that the, the key difference between AI being successful at a company, sort of, you know, by and large, and not being successful is access to internal systems, and MCP is that gateway there.

It ends up being a, a, a, a good sort of governance and choke point. And, um, and so yeah, so there's, we're seeing some success [00:41:00] around that. We have an open source project called Tool Hive, but we're also finding that with the AI era, with the ability to write a lot more software so much faster... what used to be a company is now a feature of a larger platform.

And so, you know, the traditional thing is that, you know, startups would build relatively narrow and deep expertise, and then over time, as markets matured, these things would sort of conglomerate together and, and the definition of products would change from being sort of these narrow things to larger platforms.

Um, I think we're seeing that accelerate in the AI era just because the capacity to build is so much greater for companies. And so we're now moving from just focusing on MCP, and we haven't talked a lot about this publicly, but like, um, uh, to looking at sort of like what are the other ancillary thing, 'cause we think MCP governance is actually part of a larger sort of governance AI platform.

Now, everything we're doing is on-prem because we find that there's a lot of companies there that, um, they don't wanna let... This is sort of the lifeblood of their company. They don't wanna let it outside their [00:42:00] firewall.

Matt Klein: Yeah. Makes sense.

Joe Beda: And so, um, so we're running all this stuff on Kubernetes, uh, with the idea for, you know, large, mature companies that really value keeping their data private.

Um, so a lot of banks, regulated industries-

Matt Klein: Yeah

Joe Beda: ... that type of stuff. And we're looking at, like, what does it mean to complete that platform? What are the other pieces of the puzzle that we need to build? And so that's what we're working on right now.

Matt Klein: Yeah.

I, I mean, it, it is, at least for, for me, and o- obviously, like every other company, we are doing our own AI thing at bitdrift.

And I mean, you know, it's like everyone's heading in this direction. But, you know, the, the level of change on an almost weekly or monthly basis, n- not only in terms of, like, who's ahead, you know, is, is, like, Opus better or is Codex better? It, like, changes every week, right? Or, um, you know, the, the topic that I wanted to ask you about, because it's what you all do, is, like, MCP is dead.

CLIs are in. CLI... You know, it's like I, I, I can't keep track. So, you know-

Joe Beda: It's gonna keep changing

Matt Klein: ... and, and, well, a- and I was actually [00:43:00] wondering that, I think, um, I was reading the other day, GitHub actually did a, did a blog post about how they've optimized their token spend and a bunch of other stuff. It was actually very interesting.

And, you know, they are very anti-MCP, I, I think due to token costs and, like, various other things versus having, like, a custom CLI and all of those things. And at the same time, it doesn't, to, to me it doesn't, uh, whether you have an MCP, whether you have a CLI with appropriate auth and policies and all of those things- I, I'm not sure it matters, but, but I wanted to get your expertise-

Joe Beda: Yeah

Matt Klein: on this and, like, how are you all thinking about just this, like, fast-moving landscape on how people are even interacting with the harnesses and whether they should be using CLIs or APIs or MCP I just like I can't, I can't keep track right now.

Joe Beda: I mean, there-- Yeah, I got a lot to say there. So, you know, first off, I think, you know, MCP being dead or not, I, I mean, there are problems with MCP.

It's very chatty. It sends a lot of stuff, the [00:44:00] huge token cost there. Um, there's techniques, like one of the things we have built into the, the Tool Hive gateway is a virtual MCP server that trades off ch- uh, uh, chattiness for verbosity. And so essentially the AI, you know, the, the model can ask, can-- Essentially, there's a tool saying search tools that can find-

Matt Klein: Right.

Yeah

Joe Beda: ... help you find other tools, right? So you, you add a level of indirection there, it reduces the token cost dramatically. And so there's ways to, to, to sort of work through that. Um, and w- and we're working on that. Um, but I think as we're targeting larger enterprises, I think, you know, and, and, and I'm sure you've seen this also, is that when you're in the sort of tech-forward company bubble, you think this stuff is moving really fast.

Once you start talking to enterprises, you find that, that, um, adoption cycles take a, a long, long time. So, like, I, you know- After I-

Matt Klein: Sorry

Joe Beda: ... after I left Google, I started this project with essentially a white paper and other people took it on, called Spiffy, um, which is one of the [00:45:00] sort of identity frameworks that's starting to sort of emerge for agent-to-agent stuff.

And that was, like, about 10 years ago. So it's like, I saw that it was based on some internal Google sort of ideas and, and, you know, rendered in a way that I thought was more palatable outside of, of Google. And it took about 10 years for people to see the need for this, right? And so it takes a long time for these technologies to reach, to reach mainstream.

And so one of the things that we're looking at is like, hey, there's so much action around the developer persona and what it means to do things like Claude Code and the harnesses and development around that, right? And so even your question of like CLI versus MCP assumes that you're running an AI agent that has an environment where it can run a CLI.

Matt Klein: Yeah.

Joe Beda: And that's not always the case, right?

Matt Klein: For sure.

Joe Beda: Not everybody is like-- And that, you know, right now, unless you're running that in, in a, in the cloud somewhere, that then ties you to a desktop and then there are, you know, you have people buying Mac Minis to run OpenClaw, which is kind of wacky, right?

And so- so I think that, you know, uh, a couple of things is that, like, enterprises are gonna take longer than everybody thinks to really truly [00:46:00] adopt this. Even if they're accelerating it, maybe that 10 years gets cut down to five, but it's still gonna take longer than we think. And we see trickle-down of ideas from what's happening, I think b- I believe in, like, for, for developers.

I think that'll start applying to knowledge workers in general. Um, but it won't be in exactly the same form. I think we're gonna see ideas trickling down.

Matt Klein: Yeah.

Joe Beda: But in the same way that we saw the transition from, like, Word to Google Docs, where it's like a shared collaborative thing, it's the same thing but different.

I think we're gonna see a lot of places where the way AI is used by developers will influence and trickle down to knowledge workers and other people, but it'll be inspired by, it won't be exactly the same as Claude Code. I don't think we're gonna expect, like, you know, your finance person to be checking stuff into Git and-

Matt Klein: Right

Joe Beda: you know, m- you know, approving command lines directly, right? We, we can't have that.

Matt Klein: Yeah.

Joe Beda: But, but something similar is going to have to happen.

Matt Klein: Right. I, I mean, not to mention the fact that I, I feel like [00:47:00] we're just now entering the era in which people are gonna care about token costs, right? And I, I, I'm actually curious, like, for the, a- again, for what you're willing to share for the type of companies that you're talking to or trying to sell to, is I feel like we had a, a few years of this stuff being almost free, and at least my personal opinion is that the, the bill is coming due soon, right?

Joe Beda: Yeah.

Matt Klein: And, and the reason that I ask about that is that I, I think that's gonna change actually a lot about how people use the systems. They're gonna have to be more cost cognizant of what they're doing. Like, they're gonna care more about the stuff that we were just talking about, about the fact that MCP is chatty or not.

Um, so I, I, I don't know, like, wanted to get your thoughts on, uh, not the bubble per se, but, but just, like, I think even now in the short period of time, the industry is about to transition to one where the [00:48:00] economics matter a lot more than they did. And I'm just curious how you're thinking about that from yourself, from your company perspective, you know?

Yeah.

Joe Beda: I, I think, you know, when people are looking at AI across a company, I think there's three considerations. Does it actually work? Is it adding value? Is it safe? And, you know, controlling costs, right? And I think we're gonna, you know, the, the importance of those things is gonna shift over time. I think there's a couple of trends that we're seeing is that there's so much action happening what I'm calling sort of left of the LLM.

So if you draw a diagram of, like, flow from the user to the LLM, uh, and I could show you something, the, the LLM, you know, sort of shows up on the right. Everything happening on the left there is where the, the, the, the agents and the, the, the reasoning loops and the harnesses, all that stuff-

Matt Klein: Yep

Joe Beda: ... is happening there.

And we have a lot more openness on the developer side, but when you start talking about the consumer or the knowledge worker side, we have much more vertically integrated, uh, experiences there. And so if you're a company and you buy into Claude, you're now [00:49:00] theoretically delivering Claude desktop and mobile to your knowledge workers, and they're in, and in a sense, you're locking yourself into Anthropic's world there.

Yep. So there's parallels here with, with cloud and how people are actually like, "Hey, you choose your, your partner, and you're stuck with them." But with this stuff changing so fast, with the cost economics are what they are, with open-weight models getting better, and-

Matt Klein: I was gonna say, I mean, I mean, the, the open-weight models, they're three to six months behind, and it's like-

Joe Beda: Yeah, and it's like-

Matt Klein: It, it, I, I'm just saying, like, at least for my personal use, speaking purely from coding, like, I, I'm actually right now all in on the GPT Codex models.

But, like, 5.4 and 5.5 for me are almost good enough right now. And it's like-

Joe Beda: It's gonna, it's

gonna change, right? And, yeah. And it's clearly gonna get to the point.

Matt Klein: Right.

Joe Beda: And so I think that there's a lot of room. So, you know, we're building on a LLM gateway that actually gives people-

Matt Klein: Yeah, sure

Joe Beda: ... a, a governance points for token, you know, f- you can do steering of different requests.

Matt Klein: Yes, right.

Joe Beda: You can actually host models. It gives you a level of agility [00:50:00] there that you're not gonna get-

Matt Klein: Yeah

Joe Beda: ... if you go all in with one frontier provider. But that then also means that all your clients that are calling into that LLM gateway have to be able to do that, and doesn't behoove any of the frontier provider models-

Matt Klein: Sure

Joe Beda: to actually make it easy for you to, say, hook up Claude desktop to an open-weight model. They're not gonna let you do that.

Matt Klein: Yep.

Joe Beda: Right?

Matt Klein: Sure.

Joe Beda: And so I think there's a lot of opportunity there in terms of, of creating a, a horizontal self-hosted solution here for everything that happens left of the LLM- uh, for outside of developers, right?

And I think that's-

Matt Klein: Yeah

Joe Beda: ... that's the place that we're really excited about.

Matt Klein: Yeah. All right, cool. Um, so I, I, I guess what are your three to five-year, uh- ... predictions? I mean, uh, things change every week now, but like, you know, I, I mean, do you, I don't know, like-

Joe Beda: I don't know. I think-

Matt Klein: Give us your final, final hot takes.

Bubble, not bubble. Uh, what is, what is going on right now?

Joe Beda: So we see, so many times we [00:51:00] see this sort of like S-curve where, you know, things are slow, and then they accelerate, and then they even out. And when you're on that acceleration scope it feels, s- slope, it feels like it'll go, be going forever, similar to the Microsoft stock price-

Matt Klein: Yep

Joe Beda: when I first joined the industry, and it inevitably evens out, evens out, unless it doesn't, right? And I think one example of that is Moore's law. For whatever reason, for like 20, 30 years, we had this like incredible exponential growth in computing power. And I think for me the big question is, is that with AI advancements, are we actually gonna see a Moore's law type of dynamic or are we gonna actually see a more traditional sort of S-curve evening out?

Um, and I honestly don't know. I, I talk to people like I'm, like I worked at, at Google with Steve Yegge who's doing the Gastown stuff which is like AI maximalist like-

Matt Klein: Sure, yeah

Joe Beda: ... plus plus plus stuff which is fascinating. And I had coffee with him the other day and he's convinced, like he's like, "Everybody wants me to start a startup but I can't think of a startup that AI just won't make redundant," right?

And so, you know, it turns out that if you go far enough [00:52:00] ahead you're like this reinvents the entire economy, and I think that's how you get these people who will say these outrageous things because they're extrapolating assuming a Moore's law type of, of forever and always increase. And I, I don't know where that ends.

I just know it's gonna be different from anything we're gonna predict right now.

Matt Klein: Well, and also, I mean, I, I think when you talk about Moore's law it's, it's an interesting parallel but at the same time it's like the increase in power does not necessarily mean an increase in profits, right? Because at least for, for me today where I'm sitting is that you have several companies which have basically equivalent products.

I mean, they're all building the same thing. You have the open weight models that are coming in. And what is confusing to me is... how do you capture value where it doesn't become like a Lyft versus Uber situation where you can just switch, right? And, and-

Joe Beda: I mean, you

sell

GPUs is what you do,

Matt Klein: right? R- r- uh, right, right, right.

So, so it's like the people that win, I guess, are NVIDIA or maybe Google with TPUs or Amazon-

Joe Beda: Yeah

Matt Klein: ... with their Trainium or [00:53:00] something along those lines. So I don't know. I mean, it is, uh-

Joe Beda: Or

like the CoreWeaves of the world that can actually arbitrage this stuff, right?

Matt Klein: Who knows? Yeah. Anyway, um, that's a, that's a good place to end it.

Um, thank you, Joe. This was a fantastic conversation. Thank you for joining. I think people are really gonna enjoy it. So, uh, that's a wrap for this episode of Beyond the Noise: Signals, Stories, and Spicy Takes. Huge thanks to Joe for joining and sharing his story. You can find this episode and all past ones on the bitdrift YouTube channel.

If you had fun, drop us a review, tell your friends, or yell your favorite hot take into the void, and just make sure to tag us. I'm Matt Klein. And I will see you next time.