Episode 26: Matt Moore, Co-Founder and CTO of Chainguard

Today's guest is Matt Moore, Co-Founder and CTO of Chainguard — the software supply chain security company on a mission to make the open-source ecosystem safe for enterprises everywhere. 

Matt's background spans some of the most consequential corners of the cloud-native world. Before co-founding Chainguard, he was a core contributor to the open-source ecosystem and played a pivotal role at Google, where he helped shape the Kubernetes and supply chain security landscape that underpins much of modern software infrastructure today.

In today's conversation, we'll dive into Matt's journey from open-source engineering to founding a company, how Chainguard is tackling one of the most critical — and often overlooked — challenges in enterprise software, the philosophy behind building security into the foundation rather than bolting it on, what it takes to turn deep technical expertise into a venture-backed company, and his vision for the future of software supply chain security in an AI-first world.

Transcript

Welcome And Guest Background

SPEAKER_00 0:06

Hi everyone and welcome back to the Finance E podcast. Today's guest is Matt Moore, co-founder and CTO of ChainGuard, the software supply and chain security company on a mission to make the open source ecosystem safe for enterprises everywhere. Before co-founding ChainGuard, he was a core contributor to the open source ecosystem and played a pivotal role at Google, where he helped shape the Kubernetes and supply chain security landscape that underpins much of modern software infrastructure today. Prior to Google, he was at Microsoft for seven years. In today's conversation, we'll dive into Matt's journey from open source engineering to founding a company, how ChainGuard is tackling one of the most critical and often overlooked challenges in enterprise software, how Claude Mythos is affecting cybersecurity, the recent Mercor breach that happened a few months ago, the philosophy behind building security into the foundation rather than bolting it on, what it takes to turn deep technical expertise into a venture-backed company, and his vision for the world of software supply chain security in a AI-first thinking world. Now, without further ado, let's get into it.

Why Matt Started Chainguard

SPEAKER_00 1:19

Alright, Matt, I am super excited to get this on the calendar and talk with you. I've been following Chain Guard for a little while now. So really excited, and thank you for joining the show. Yeah, thanks for having me. Um just kind of start off and begin our conversation. Wanted to briefly talk about your career. You spent 15 years across some of the biggest companies in the world, Microsoft, Google, VMware. A lot of people would say that's kind of their end goal. I was curious before um, you know, talking, reflecting on that period, but can you tell us about the spark for you to leave VMware, um, start Chain Guard? And what was kind of the origin story?

SPEAKER_01 1:58

Yeah, um, I mean, I honestly I never really thought that I'd start up. Uh I had some friends that tried and and failed early in their in early in my career while while I was at Microsoft. And I uh I saw how hard it was for them, and it gave me a sort of uh appreciation for the challenge. Um uh during the tail end of the pandemic, I had actually left VMware. I was taking some time off and uh, as I say, smoking meats and playing video games in my backyard. Um and uh I I sort of started to get bored. Uh and I started to sort of search for what I wanted to be when I grow up. Uh and I talked to companies of all sizes, you know, some of the biggest companies in the world, as well as, you know, um some pretty early stage startups. I think I talked to one one group who um, or you know, one founder who had like just graduated from Y Combinator. Um and none of the none of the startups I talked to and what they were working on really like uh spoke to me. Um, but I did start to feel the the sort of itch to uh try and start something uh you know of my own. Uh at the same time, yeah, our CEO Dan had was uh who I worked with for many years at Google, uh kept trying to get me to come back to Google. Um and I finally told him I'd start working again, but only uh to start something new. I because really I wanted to bet on us in a way that rejoining big tech really wasn't. Um around the same time, um, solar winds had just happened. Uh it was a real problem to be solved with an enormous market. And um, Dan and I had spent years discussing the supply chain uh problems at Google before I left. So the timing felt right.

Software Supply Chain Security Simplified

SPEAKER_00 3:54

Um and I guess for you, before we dive deeper into kind of what Chain Guard does, um for someone who doesn't live in your world or said you know, security, um what what would you say is software supply chain security? What does that mean in plain English? You guys obviously your biggest customers is Snap, HPE, they all kind of use it. And uh, I guess why does it matter to a Fortune 500 CTO and in and in general?

SPEAKER_01 4:23

Yeah, so I I like to think of sort of modern software as the proverbial iceberg, right? Like there's you see a tiny bit above the surface, but there's this huge, you know, uh block of ice that's under the water, right? Open source software makes up 90 plus percent of the code that uh runs in applications today. The stuff sitting below the surface in that iceberg, right? When you're checking your bank account balance, when you're mapping out a route on public transportation or you pay for something online, open source is helping to power and underpin all of those interactions. Um it's it's really helped companies innovate and build faster. Um, you know, I I think about like all of the work people had to do. Um, you know, it used to be like racking your own servers and things like AWS and GCT came along and made it so that you didn't have to do those things. You could sort of outsource it. Um, but in in much the same way, right? Um what you used to do before you know the open source ecosystem became so um powerful, uh a tool for folks to leverage, you ended up writing like 100% of the software that you were gonna ship, right? And as open source has grown, your ability to reach for sort of off-the-shelf libraries and stuff like that and compose those into powerful applications has grown and grown and grown, right? And so you know, that you went from writing 100% of the software to writing less than 10% of it. But right, even having avoided writing that 90%, uh, you still are accountable for 100% of the software that's running in your environment. And that software can carry vulnerabilities, it can um embed malware. Uh Dan likes to use the analogy that the way folks used to consume open source software prior to ChainGuard was the moral equivalent of taking a USB drive you found in the parking lot and plugging it into your computer. Not a not a good thing to do. I wouldn't recommend it.

SPEAKER_00 6:28

And I guess so how I like to think of it, so you're basically saying it's um if I were to create like a Python app, um, I would use like your container images or your library to basically use uh say source code that is nearly close to zero vulnerabilities. I think you guys call it uh CV CVEs, is that correct? And so that's kind of that's kind of the idea of the software supply chain. Am I am I get right?

SPEAKER_01 6:54

Yeah, so so I mean Python's a great example. Um, you know, I think there's a question of like you're not gonna write your own Python interpreter, right? You're gonna you're gonna reach for a Python uh image that provides that Python interpreter. You're likely not going to write all of the Python that runs in your application. You're probably gonna pull in um application dependencies that, like, okay, maybe you're writing a service. You're probably not gonna write your own Python HTTP stack or you know, REST layer. You're probably gonna use some off-the-shelf libraries, you know, and if you're dealing with um, you know, structured data, libraries for parsing that structured data, all this stuff, you're not gonna write that stuff from scratch. You're gonna start from a Python interpreter, you're gonna install a bunch of Python dependencies for serving an HTTP server, maybe serving a REST API on top of that, and then you know, parsing that structured data so that your application can deal with it, right? And so you know, you end up writing a tiny bit of logic that leverages these serving stacks, you know, the interpreter, you know, middleware, et cetera. Um, and you know, you you, I mean, today to write uh you know a modern HTTP or API app, you know, you end up writing a pretty tiny amount of code. You think of services like Lambda, right? Like you write this tiny little bit of code, and then a lot of stuff is taken care of for you, right? Um, you know, open source is the same way, but where are you gonna get that Python interpreter? Where are you gonna get a safe source for those, all those language-level libraries that you're pulling in? Um, though those are the places where ChainGuard plugs in, right? We have a version of the Python-based image that you can take and run where we've applied all of the patches. And so there are no uh known vulnerabilities, right? CDs is uh common vulnerabilities and exposures, um, it's a way of sort of classifying vulnerabilities, and um uh that's standard. It's uh what a lot of people track. Sometimes they have um you know equivalent names. So GitHub has their own sort of security advisory mechanism that tracks a lot of them. But um, but yeah, I mean, and a lot of compliance frameworks um uh you know have stipulations around uh tracking these kinds of things and in some cases timelines for remediating them. So if you're uh trying to do business with the federal government, for instance, um you you typically need to um follow what's called FedRAMP compliance. Uh and those compliance frameworks uh require folks uh to um basically mitigate vulnerabilities that are critical within 30 days and you know um uh high and medium on you know long longer timelines. But so uh, you know, I I used to joke that I didn't know how anyone could get compliance with these things before we exist as a product. Because if you just take an off-the-shelf, you know, Python image off Docker Hub or wherever else, um, you know, it's gonna have CVEs uh that are unpatched because a lot of the distributions and and whatnot aren't applying all of the patches in the same way that we are at ChainGuard, right? We we've bootstrapped our own distribution, built everything from source so we could have the control

PyPI Typosquatting And Malware Risks

SPEAKER_01 10:12

we needed to apply all of those patches, make sure folks are secure, right? And then at the library level, um you know, yeah, you you may pull down an HTTP server or um something to parse data, right? But you're pulling that down from PyPy. It turns out anyone can publish packages to PyPy, right? It's not like a distribution where there's like a strict set of maintainers. Um, and it's been uh PyPy, NPM, all of those equivalent like language package manager ecosystems, um, you know, has been a hunting ground where uh attackers are finding ways to publish packages as either popular maintainers or doing things like typo squatting. Right. So say you wanted to use um, you know, uh some Python package and you accidentally like swap two letters, right? An attacker may have published a package with malware uh with those two letters transposed. And if you just make a typo, right? You now pull down that malware and your app when it when it runs, or maybe even at install time if you're using something like node with install hooks, it could, I mean, it shutting down, it would be easy, right? Like it probably uh will run, it will probably scan your system for credentials and it will send them somewhere so that the attacker can then go have fun uh as you. Uh and nobody nobody wants that, right? Um, it could steal your crypto wallet and it could do malicious things locally. Um, like uh there was one um uh there was one that actually targeted uh AWS's agentic developer tools that um I think it had a bug in it, so it wasn't particularly effective, but it basically tried to wipe out not just your local file system, but your AWS account and clean up all of the resources. Um and so, you know, uh these these attacks are in very malicious, very dangerous, and they tend to chain together, right? Because what happens when you get get a hold of someone's npm credentials? Well, hey, I can publish packages as them, right? And so these things can can chain. And we saw that with attacks like Shai Hulude. Um and uh but yeah, I chain guard libraries provides a source for that you can slot in in place of the upstream language library dependence so that you don't have to worry about um those attack paths, right? Because we're taking care of making sure that uh every Python library, every node library is built from source in a controlled environment that isn't susceptible to these types of attacks. Um and then you can take your Python base image, save from ChainGuard, install Python libraries, safe from ChainGuard, and then overlay your little bit, your little 10% top of the iceberg application on top.

SPEAKER_00 13:03

Yeah.

SPEAKER_01 13:03

And you know, you're you're accountable for the code you write. There's we're still working on ways we can help you with that. But every line of code that you didn't write is now uh safe and secure and sourced from Chain Guard.

SPEAKER_00 13:16

Awesome. And I guess building off of that idea, I mean, uh obviously solar winds um and just other attacks. I I thought, you know, it was so funny when I when I heard about the Mercor breach, and I know I'm about to ask you this because you were the first name that I thought of, and I was like, okay, what the heck? So I wanted to ask you more on that. Um so Mercur, they had a security breach, it was traced back to supply and chain attack on Light LLM. Um, would you briefly kind of explain what happened and what incidents

The Mercor LightLLM Supply Chain Attack

SPEAKER_00 13:47

like these tell us about the industry's security model? Is it fundamentally broken and I guess how you guys also play a role in that?

SPEAKER_01 13:54

Yeah. Um yeah, so light LLM, uh, for folks not familiar with it, the a popular AI library, it's got something like 97 million monthly downloads, right? So it's an extremely attractive attack target, right? And one of the things that we see, you know, I mentioned the the um AWS Agentic developer tools, right? Like the the agentic tools aren't necessarily the the attack path, right? But AI is so hot right now that AI projects make really good bait, right? Because everyone's installing them. And and it may not be the um actual attack path, right? But like by uh getting malware into it, right? It's not AI's fault, right? It is AI's the bait, right? And when you install that, you then you know have sprung the attacker's trap, right? So in March, uh basically, attackers published two compromised versions directly to PyPy, and it was out there for about three hours undetected. Right. So during this period, any developer who updated that tool uh unknowingly ran code that uh, you know, I mentioned all the nefarious things it does. It would silently vacuum up uh passwords, credentials, sensitive files from their machine. Um, and uh from there, right, the attackers could use these stolen credentials to move deeper into your systems. Um, or you know, we've seen uh folks like I mentioned Shai Halud, which I mean that seems like 12 attacks ago, it probably was, um, but it really was fairly recent. Um it actually was self-replicating. It would it would look for credentials and then publish more malicious packages, right? And so these things chain together, right? And the the credentials that it gets from your system then get used for the next attack, right? And um, it's it's definitely um yeah, a big problem. But this is one of the reasons that throughout ChainGuard, um basically since our inception, we have banned the use of long-lived credentials. We we do not have long-lived credentials that we use uh basically anywhere we can avoid it. Um and uh it has served us very well because it is one of the biggest attack paths basically everywhere um these days.

SPEAKER_00 16:11

So and I guess I gotta ask you this because Merkers seem to have really large consequences of this attack, you know, kind of losing partnerships with, you know, the open AIs of the world, that sort of thing. And obviously, you mentioned we we we live in an AI first world, and uh Merker hander handles like the AI training data, um, which is the most sensitive IP um in the industry right now. So it was curious, does a breach like this change how seriously that the AI world needs to take supply chain security seriously? Um and I guess your outlook there.

SPEAKER_01 16:45

Yeah, I mean uh it's I mean the AI space is moving fast. It is you know, every company uh involved that I that I've talked to, right, like they're they're approaching this like there is no second place, right? Like they they view, you know, everyone uh operating in this space as an existential threat to them, right? Um and what does that mean? That the pace they're moving at, right? Like often velocity trumps almost everything else. I agree. Um and you know, it's those kinds of environments where you know supply chain uh risk compounds the fastest, right? Um so you know, I I think that you know there's there's all kinds of different like categories where there could be damage, right? There's you know reputational damage. Um, you know, I you mentioned that uh they lost deals over this, right? Um that they so they lost uh revenue, they lost partnerships, right? Um and you know, I mean, I I don't know uh of a lot of people who knew of the the the brand name Solar Winds before that breach. And it now has become synonymous with supply chain attacks, right? Um and so I think that um I think that folks need to appreciate that. And you know, I do think that with the adoption of almost any tool, like this isn't unique to AI, um, it typically takes a while for folks to figure out how to get security right. And when there's something shiny and new, folks tend to rush into it and like security is sort of an afterthought. But I think with the power of um, you know, some of this AI tooling, it definitely has a lot of folks in the security space um, you know, concerned uh about just the power and the level of credentials that uh folks are handing to these things.

SPEAKER_00 18:37

Um and I guess taking all of that, what we've kind of discussed, um, I wanted to mainly get your thoughts on, you know, you mentioned speed and power of AI. It's it's incredible. It's incredible how much

Claude Mythos And Exploits At Scale

SPEAKER_00 18:52

like life has changed since like beginning of 2024 to now. And I wanted to talk to you a little bit about Cloud Mythos, AI, but also, you know, everyone's talks about the SaaS apocalypse. So my first question was more on Mythos. Um they reportedly went from finding vulnerabilities to building working exploits without human help. Incredible. So when you see a capability like that that jumps into such a steep single model generation, like it's it's the strongest model they even claimed. What is your thoughts there? What does that mean for cyber companies over the next few years? Um, I guess generally, and then if you want, also diving into just supply chain as well.

SPEAKER_01 19:31

Yeah. I so I the way I look at it, right, like turning turning a vulnerability into a working exploit used to require expensive, skilled humans, um, but you know, uh AI removes that bottleneck, right? Uh and the window between vulnerabilities disclosed and an exploit in the wild um is shrinking from months to hours, right? Right before our eyes, right? Um, I mean, even before you know Mythos came came on the scene, there were papers uh uh there's one called CVE Genie that talked about um, you know, basically taking known vulnerabilities, disclosed disclosed vulnerabilities, and producing exploits from them. Um I think that combination of the lag time that it would that it used to take folks to produce vulnerabilities, as well as the skill requirement of like how how skilled a person had to be in order to do that, um, is what allowed sort of traditional distributions to take so long to patch vulnerabilities and organizations to pick up those patches and roll them out, even with compliance pressure from uh frameworks like PCI, FedRAMP, HIPAA, um, those barriers are gone now. And I, you know, I believe that running unpatched software with known CVEs is going to shift from um, you know, fairly common practice to you know approaching negligence, right? Especially if it leads to a breach. If you knew about it, you didn't, you know, do everything you possibly could to remediate that vulnerability. There's a patch available, you just didn't deploy it, right? And you get breached as a result, your customer, your customers or user data gets uh compromised as a result, it starts to approach negligence. So I like I I applaud Anthropic um for taking the steps to to help the industry's largest software vendors get ahead of this, right? The you know, if you think about the number of Apple devices out in the world or stuff running Microsoft's software or um, you know, uh the just the sheer uh if you think about how big cloud is, right? All of the major cloud vendors are just enormous targets, right? So like working with them to get ahead of this is definitely um uh you know um very uh uh a good thing, right? But I think the the key thing is it's not just About you know disclosing the vulnerabilities and getting them patched, every single downstream organization of all of all software, right, needs to start preparing to release software updates at machine speed because that is the speed at which attackers are going to start exploiting known vulnerabilities.

SPEAKER_00 22:21

And then I guess on uh the more the broader side of software, more on the the SaaS side, I guess I I mean I've I've listened to so many podcasts, I think everyone talks about MOS and just in general. So I was just curious, like, what do you think creates the defensibility for these broader software companies and security companies? Maybe on like the larger enterprise, like a sales force versus like startups that are a bit smaller, still developing. I guess is it the data moat or the fact that you know you have large enterprise clients that are so used to your software? So was very, very curious there.

The Only Real Moat In Cyber is... Trust

SPEAKER_01 22:57

I at least within the realm of cybersecurity, I think the most important moat is uh trust. Um, I think existing relationships that are healthy, which is not a given. Uh some people hate their vendors. Uh, I hope not us. Uh, but uh those existing healthy relationships, right, can give companies a leg up, but you they still have to continuously deliver competitive value and continuously grow that trust. Um, I think the the so-called um SaaS apocalypse reflects a dramatically shrinking software mode. Um, if your competitors and and upstarts can basically clone your feature set from your launch announcement blog post, then clearly that alone isn't a moat. Um, however, a competitor with a carbon copy of your features can't necessarily replicate that uh trust moat overnight.

SPEAKER_00 23:56

And I guess then what is um I guess in regards to more of the the positioning of Chain Guard, going kind of back to Chain Guard, you know, uh you guys mentioned I I forgot which exactly was the article, but you guys are now kind of explicitly around the age of AI, the idea that modern software is being written by AI agents, and also the idea of vibe coding. My roommate loves to vibe code. So I was just curious, um, how do you think that fundamentally changes the supply chain security problem? And does how I guess how does ChainGuard play into those dynamics?

SPEAKER_01 24:32

Yeah, so I mean, I when I when I sort of take a step back, uh many of the problems that ChainGuard solves behind the scenes are sort of operationalizing and scaling best practices across all of open source. Um, you know, it early

Vibe Coding And The New Risk Curve

SPEAKER_01 24:49

on, like agentic AI didn't exist when we were getting started, but we leveraged traditional automation to push this further than really anyone had before us. Um, and we're now starting to leverage agents to push this even further, right? If traditional automation could get you to say 80%, uh, that's great. Our engineers are five times more effective, right? But I think as we scale, like we want to look at how we can leverage agents to make that 90% or 99%. Um so the way I see that organizations uh struggled to operationalize and scale best practices. Um uh so um it's uh it's why so many organizations struggle with technical debt, right? Because they haven't figured out how to scale and operationalize those best practices. Um and uh that was that was before agents, right? Like uh organizations struggled with the mountain of tech debt they had before agents were there, right?

SPEAKER_00 25:47

Yeah.

SPEAKER_01 25:48

And now agents are enabling their engineers to produce 10 to 50 times that volume of code, right? And so um I, you know, I talked to customers that had uh you know millions of CVD CVEs running in production before they adopted ChainGuard. Um and there's no way any of them would sign up for 10 to 50 times that number, right? Uh but I think the nice thing is 10 to 50 times zero is still zero, right? And so if we can get them to zero, and if we can get it to the point where if your roommate is vibe coding a Python app and he's getting his base image from Chainguard and his libraries from Chainguard, um, you know, it if he's not writing one app now, but he's writing 50 apps now, and all of them getting those those bits from ChainGuard, you know, he's gonna be at zero um, you know, regardless of how much more software he's writing. So I think I think the key thing is helping organizations to sort of um operationalize um and scale best practices. And you know, we are trying to do that for that 90% of the iceberg that is software that you and your roommate aren't writing yourselves, right?

SPEAKER_00 27:00

And I guess um, so when you think of like vibe coding, agentic AI, um obviously not not correct me if I'm wrong, and I I'm not the most knowledgeable exactly there, but what like what what more vulnerabilities does it kind of create than typically what's already known or what already happens? Like what does vibe coding and kind of that aspect do? And yeah, more curious there, because I don't know too much about it.

SPEAKER_01 27:24

Yeah, so I mean, I I think I think to some extent it depends on how how true to the like you know uh vibe coding mantra uh you are, right? Like I think, you know, um there are folks who vibe code and um you know don't look at the code at all, right? Like it's it's purely output-oriented. Um uh I think that we'll over time uh we will see that more powerful models be able to produce better and better code. Um but also right, like it depends to some extent on what kind of like, even if you aren't reviewing it, you might be telling it to or telling other agents to give it feedback in um tactical ways that make it better. You know, there's this idea of sort of back pressure um where the agents, the agents, you know, push towards some goal state, right? Like you telling it, hey, I want this web app, right? Um, but the more like uh guardrails you can give it, right? Like, okay, you you need to pass these lint checks. You need to, you know, it needs to build, it needs to pass all of this suite of tests, right? Um, the more um, the more things that you can give it, uh, it's you know, basically a feedback loop, the more feedback it can get um about what it's doing wrong, um, you know, the further it can get uh towards a good state, right? So I think that um it depends a lot on what back pressures you can build into the system. And where we where we are leveraging um agents, uh you know, we we are investing in many, many different types of guardrails and you know, defense in depth to keep those um uh agents uh you know from doing something that we wouldn't want them uh to do um and to not make mistakes, to not hallucinate. Um so yeah.

SPEAKER_00 29:20

Yeah, and I guess more on the the side you you mentioned obviously you guys use a gentic AI, that helps you a lot. And man, I'm sure you know attackers as well on the offense, um, AI really helps accelerate that in regards to finding vulnerabilities faster. I guess do you think when you think of broader security, um, do you think defenders are keeping pace or are we in a fundamentally asymmetric moment um right now and I guess in the future?

SPEAKER_01 29:46

Uh I think I think in general the answer is no. Uh I don't I don't think that defenders are uh necessarily keeping pace. Um I you know it comes back to my point that like attackers are starting to move at machine speed. Yeah. Um and many, many defenders are still triaging vulnerabilities, extant examining mitigations and context, and um pushing

Why Copying Chainguard Is Hard

SPEAKER_01 30:10

out new releases at human speed.

SPEAKER_00 30:12

Yeah.

SPEAKER_01 30:13

Um at Chain Guard, one of our goals is to make uh upstream patches available at machine speed so that our customers can react at machine speeds.

SPEAKER_00 30:24

Okay. And I guess more on the defensibility of Chain Guard. Um like when you think of like Claude and kind of going all back to that, I would love to hear more on your defensibility. Um, you obviously talked a lot about just trust, but who's to say I can't I can't vibe code a really good engineer can't vibe code a uh a chain guard? Um so I was just curious what what is does the defensibility still go back to trust? And I guess how easy is it to vibe code something like Chain Guard? And yeah, I was just curious there.

SPEAKER_01 30:55

Yeah, I so it's definitely an interesting question that um we think about and we get a lot, right? Um I think I think a lot of folks don't appreciate necessarily the um amount of software you have to bootstrap to build out a working ecosystem. And I like that, you know, like when we bootstrapped our own distribution, we um you know, we built like you know, uh our own GCC tool chains and use that to bootstrap the whole distribution from the ground up. And we have blog posts talking about um, you know, the the stages of bootstrapping a distribution uh that we went through. And then like that that parallels to every single language ecosystem has a different like bootstrapping process. But um, if you think about like to to copy what we do um on your own, the amount of infrastructure you'd have to build and the amount of compute you would have to spend building all of that software, right? Um uh as well as you know the token volume uh to you know produce all of uh that both that infrastructure as well as all of the definitions that flow through it, it's it's an enormous cost, right? It's you know uh capex, right? Um and so you know, a part of how we can do it is um we can amortize that cost over many customers that we are doing it for, right? And so could could I mean could someone wake up and do what we do? What we do isn't magic, right? We we are building things from source, we have bootstrapped a distribution and we have built automation to try and make that as efficient as possible. Like, sure, like other folks can do what we have done, but it is um so there's this saying I I like to um bring up that's called the programmer's credo. Um we do these things not because they are easy, but because we thought they'd be easy. Um, right. So I think I think one of the problems as folks try and uh DIY, do it themselves, um is um, you know, they think it's easy. They like we've made it look easy, so they think it's easy, but the amount of software you need to build in order to bootstrap things and to continue, I mean, and there's also building it once, and then there's the continuous upkeep aspect of how do you keep it running, right? Um, you know, we deal with organizations that in some cases have thousands of different like application images running across their environment. And you know, the the work it takes to keep it up, yeah, those thousand images uh up to date, right? Those thousand images probably pull in tens of thousands, if not hundreds of thousands of different components at the um system level, the language level, those things depend on the tool chains used to build each of those language ecosystems, maybe Go toolchains, maybe C toolchains, maybe node runtimes. We talked about Python runtimes, right? So there's just this enormous um you know breadth of open source that um you know uh people have consumed the long tail of, right? We still, you know, we have a catalog of you know more than 2,000 container images, and we still come across customers who are like, oh, you don't have these images that we uh I mean, the entire growth of our catalog has been fueled by customers like that. They come to us, they're like, oh, you have like 90% of what um we need, but you don't have these images. Those images go into our uh backlog and we we add them to our inventory. And so there's just such an enormous wealth of open source that, like, if your organization consumes a substantial amount of open source, I think folks uh generally don't appreciate uh just how much effort it goes into building all of that from source, which is really what you have to do in order to um eliminate all of these classes of vulnerabilities.

SPEAKER_00 34:56

No, I mean that makes a good point. I think obviously probably 10 to I mean, I don't even know, maybe like 10 to 30 years you could have, you know, something like AI, but fully run this. But I also feel like the whole human aspect as well, um, kind of partnering with with AI plays a pivotal role. So I really enjoy that.

SPEAKER_01 35:16

Um even if, even if I mean, even if next year, right, the models were capable of if you prompt it, you say, I want something that matches Chainguard's factory, right? Like even if you could do that, right? And it could go off and build that, right? You still have to build um tens of thousands. Yeah, like think you know, think about the resources that go into building massive applications like Chrome or uh Clickhouse, right? Now multiply that by just the breadth of the ecosystem, right? You know, you think, okay, well, like I'm not gonna build my own Chrome browser. But like um, you know, people run headless Chrome for like browser testing and stuff like that. And, you know, we see uh, you know, test environments become attack paths because a lot of test environments these days have access to cloud uh resources and stuff like that. So um the just the raw compute, even if you had AI give you what we have, right? Like the compute costs of doing all of that just for yourself is potentially very large. And that doesn't even get into all of the like, okay, now partner with scanner vendors. So they recognize your homegrown distribution. Or you know, maybe maybe you vibecode up your own uh container scanner, right? But now, okay, you you have uh you're sort of grading your own homework, right? Uh and so you know, is that giving you a reliable signal? It's not as sort of proven out a capability as you know, battle-tested scanners like GRIPE or or Sneak or you know, yeah, I mean, there's so many. Um, so yeah. But like if you're using something like um, you know, Google's container scanner, they probably are not going to add support for your you know random homegrown distribution. Um, so I and I'm guessing the majority of uh commercial scanners or even open source scanners are probably gonna be like, whoa, whoa, whoa, whoa, whoa, you want me to add support for uh a distribution that is only used by you? Uh no, fork the project, you know, or build your own uh database uh of vulnerabilities and use that uh instead of trying to upstream those kinds of things. So and those partnership relationships that can be easy if it's open source because you can just add support yourself. But um that depends on the maintainer's willingness to accept it, right? Um and if it's only used by you, they might not. Uh and if it's a closed source scanner, those conversations often take um, you know, a very long time uh to work out the the details of. So I it's another aspect of it that's um very human. Uh, but um but yeah, I AI doesn't necessarily accelerate those kinds of conversations uh, you know, at all.

SPEAKER_00 38:06

Yeah, no, I I love that idea, and I kind of wanted to uh actually jump topics a bit, Matt, and kind of just focus a little bit more on the entrepreneurship side and you know, your uh general advice.

Startup Reality Check On Go To Market

SPEAKER_00 38:18

I was curious because you've described yourself as a serial entrepreneur inside big companies. Kind of we go back to your career at Google, Microsoft before starting ChainGuard. Um what's the biggest thing that running a company of your own revealed um the entrepreneurship at Google and Microsoft couldn't?

SPEAKER_01 38:39

Uh uh personally, I would probably say just exposure to how uh go to market works. Um so go to market uh at at least the chain guard encompasses right, sales, um uh marketing, you know, post you know, pre-sales, post-sales, right? Um, so you know, sales engineers, uh, and you know, customer support, uh like all of that, um, you know, when working at a company like um you know Microsoft or Google or VMware, right? That these are massive companies, right? Like I uh I think that I've probably interacted with more peep salespeople from Google or Microsoft uh from the outside than I ever dealt with the entire time I was there. Um but I I think that um I think that that this is sort of doubly true of companies like Microsoft and Google, which are so established as brands that in many cases customers just come to them, right? Uh for example, like when we were getting started, we didn't talk to a single salesperson at Google. Um we when we started using Google Workspace or GCP, um, our our early business just sort of fell in their lap. Uh, you know, they didn't have sales folks sort of reaching out to us because they were a known brand and we we came to them, right? Um that doesn't really help startups at all, right? It doesn't happen at startups, right? You people customers don't just come up to you and be like, oh yeah, Chain Guard, what's that? I'll I'll buy your product. Um uh you you really have to work at it. You have to get you know create brand awareness. And I think really one of the hardest things at the very earliest stages, if you can't uh if if you aren't a known entity where where customers are coming to you, how do you test your early product ideas with customers and find product market fit? Right. And so I think for that uh I give a sort of bonus uh answer, which is I I think I completely underestimated the importance of the professional network that my co-founders and I had built up over the years. Um and coming from companies like Microsoft and Google, our networks had spread to some pretty amazing places. And I'm I'm consistently shocked how often, you know, some some random person I work with is now, you know, um in in uh you know some some decadent position or somewhere else within one of the companies that we're you know trying to uh talk to. Um and it's uh you know, silver lining for me, I it's been a great opportunity uh and excuse to like reconnect with old colleagues and catch up with them. Uh but um but yeah, I mean, I I think uh the value of the networks that uh you know I and others built out over those years was useful and not just for you know selling to them, but also recruiting, right? Um, you know, we've we've pulled a fair number of uh really talented folks uh from uh you know that we've overlapped with in previous lives.

SPEAKER_00 41:38

So well then Matt, I have to ask you because this is really interesting to me, because um it seems like a lot of very successful founders like yourself come from very credible backgrounds. Um, you know, this other company called Coding Security, they do vulnerability management, you know, their founders are ex-abnormal security, and uh it seems like a lot of successful founders are either previous founders or they have just such credible backgrounds. So then for people um like like my roommate or just in general, um, every everybody wants to start a company nowadays, I think. Um so when you look back at your time building Chain Guard, I guess what words of advice would you especially give to the younger audience? What should they prioritize and what what what takes it to the top 1%, just given

Advice For Entrepreneurs: Listen And Focus

SPEAKER_00 42:23

that you know they don't have that credible network that you said you just mentioned?

SPEAKER_01 42:30

Uh yeah, so um I I think that the most important thing is to listen, listen to customers, listen to prospects, right? Like I'm if you can get the time of day from someone to give you feedback on your product, listen to it. Um, I think the key thing is you need to find and solve a real problem that they have. Um, you need to resist the urge to build cool things that nobody's asking for. They may be cool. Uh you know, I I can nerd out and build cool engineering things all day with with folks. But like it's it's it's a dangerous temptation because it doesn't necessarily, you know, you end up with the um proverbial like uh uh solution in search of a problem, right? Uh you need to focus on real problems that they have. Um I and I honestly, I think that um it took it, I mean, it took us a little bit of searching to help folks actually tackle their biggest problem, right? So as a company, when we were starting, uh we we knew the supply chain space was um uh ripe for improvement. You know, solar ones had just happened, log4j happened shortly after we we started as a company. And um we we started out, you know, I my co founders Dan and Kim had started projects like Salsa and Sigstore, which help uh folks um to um You know, secure their supply chain provenance and sign things so that you know the chain of custody, um, sort of like in a CSI show, right? Folks are signing the envelope uh, you know, containing the evidence, and then you know, they verify it before they unseal it and examine it. So like we we started out um with this objective of helping folks to secure their supply chain. And you know, what we found talking to folks and and what was running in their environment was, you know, even if we helped them, for every hop that they controlled, they were really just sort of a final link in their own supply chain, right? Everything that was coming in, that 90%, the iceberg below the water, of all the code they weren't writing, the Python base images that they were getting, the Python libraries they were pulling in, they didn't have a safe way of consuming those. They had they couldn't verify signatures on that. Nobody was signing anything, right? And so uh what we found was uh not only that the one of the biggest problems they had, and well, I would say the one of the biggest problems they had wasn't necessarily that they didn't know where that stuff came from. It was even if they knew where it came from, like those things weren't being fully patched, right? Those things weren't meeting their needs, right? We had customers going for FedRamp compliance, and you know, they had thousands in some cases, but like I said, uh we've we've had customers come to us with millions of vulnerabilities running across their fleet, right? Yeah. Um and they were building on top of existing distributions and leveraging off-the-shelf packages. And so uh we we knew that by going all the way back to source, not only could we give them a trusted chain of provenance and fill that gap of knowing where your software's coming from, but by applying all of the patches, we could give them a better source for all of the open source that they were consuming. Um, and in doing so, you know, eliminate CVEs. That pain point that I was talking about, finding a real pain point that customers have, you know, the CVE fatigue that folks feel is very, very real. And it's a real problem that they have. I honestly, I part of me can't believe nobody solved this before we did. Um, because it is a solvable problem. And I I think the irony is that um uh a lot of folks when we first started to do this, they didn't believe us. Like they were like, this isn't possible. Like it can't be possible because nobody has done it. Um and some of my favorite anecdotes sort of stem back to that. I think when Sneak added support for our distribution, um, they were like, huh, we think we got it working, but we can't tell. It's not showing us anything. Can you give us an image that has vulnerabilities so that we can try it out? And I was like, huh, I never thought anyone would ask us for that, but okay. And it we ended up building out a whole repo with um uh images that exhibit different behaviors that scanners should support. Um but my other favorite, um, and this was a team I helped bootstrap at Google, the container scanning team, uh, they actually delayed launching our um support for our images because they had to make the UI better. Because it turns out the UI didn't look very good if the images had no vulnerabilities because they just hadn't seen it before. And so I think that um the key thing, right? Find a problem that folks have and uh you know, sort of once you find that real problem that you can solve, right? You need to have conviction and relentlessly focus on solving that problem better than anyone else until you have a repeatable process for growing and growing your inventory or whatever, uh, and a repeatable process for selling it to customers, how you articulate it, how you talk about what you do. Um one of our investors is Sequoia, and our board member from Sequoia Bogamil um uh says um he said to us at once at a board meeting, by doing one thing well, you earn the right to do more things, right? And I, you know, I I see this when I talk about focus, right? Like you need to focus on the things that you uh need want to do well. Uh and it is by executing well on that and focusing on that one thing that earns you the right to do more things, right? Um it is, you know, you look at Google, right? Google, I mean, to this day, the vast majority of their revenue, ads, right? They they are the biggest ad company in the world. Um and you know, they they solved the hell out of that. But that funded everything else that they've done, right? Like, you know, building out cloud, right? That for many years did not make any money. It burned money. What money? The money coming from ads, right? And so I think, you know, this this point stands, right? By doing one thing really well, you earn the right to do more things. And you know, the the um uh success of that one thing you do well can help fund uh doing more things. It it earns you the right to experiment in additional areas. Um, and so I I tell folks this this anecdote constantly, and I even have some of my own riffs on it, right?

SPEAKER_00 49:13

Um, so and I guess building off of that point, um mainly wanted

Career Advice For An AI Job Market

SPEAKER_00 49:20

to also hear your thoughts. And I was talking to you a few friends um from my pad podcasts, and I think a lot of them are pretty scared of how fast AI is moving, and um mainly its effect on the younger generation. I think sometimes I feel like a lot of people have given up on the job market and they kind of want to go start a company because of that. Um, so when you know Gen Z are hearing a lot about AI potentially replacing entry-level jobs and making the job market more competitive, what advice would you give to college graduates about skills they should focus on, how to adapt with AI and roles they should pursue in this environment if you were to go back in time and you're 22 again?

SPEAKER_01 50:06

Um So uh I in terms of skills and entry-level jobs, I think the I think the single most important skill for all engineers right now, not just entry-level folks, um, is to embrace and learn how to use agentic coding tools. Um this includes learning how to use them to research topics, to learn new things, right? Um, you know, I I use Claude to, you know, analyze code and you know, help teach me things all the time. When I was learning about like MCP, for instance, I was um using Claude to, you know, uh not just experiment, but I was using it to uh learn, right? Uh and analyze things and show me things and you know, where it gets them wrong, it can get them wrong, right? Like I would sort of pressure test what I was hearing from it and um, you know, make sure it, you know, sort of test your own knowledge, right? But um it isn't really that I think that the current tools will be around forever. It's sort of quite the contrary. Um, as much as I love tools like Claude Code, um, I think we have yet to see the final form of agentic tooling. Um, but the fundamentals of how agentic tools work and how to get good results out of them is probably the single most important skill uh for engineers, uh really possibly for everyone to learn right now. Um I recently described this to uh you know someone else's almost an existential career conversation, even for seasoned engineers, right? Yeah. And I, you know, I think if if you want a silver lining for entry-level engineers, uh it's that even we seasoned engineers don't have that much of a head start on you, right? Like, I mean, this stuff is brand new. Um, and if you learn to use it exceptionally well, um, I think, you know, uh in some in some places you may run circles around even seasoned engineers that may not be taking um you know the the um um the potential of AI and agentic tooling seriously enough.

SPEAKER_00 52:18

Fantastic. And so mainly just the whole idea of embracing it, learning how to use utilize it, and more of it's your it's your best friend when it comes to the work that you do.

SPEAKER_01 52:30

Yeah, and I mean to the point about learning about it, don't I I would I would say don't start from a place of vibing and not looking at it at all. I would say look at what it does, learn from it, right? I mean, I think it can massively accelerate your own learning and your own growth into being a more seasoned engineer as well.

SPEAKER_00 52:50

Absolutely. Um,

Final Takeaways And Thanks

SPEAKER_00 52:52

well, Matt, thank you. Uh this this was um I I'm super glad we had this conversation. And I feel like I learned a lot. I know our audience learned a lot and um really loved in depth on just security in general. So really appreciate the conversation and thank you again for joining the show. Yeah, thanks for having me.