---
title: "S02 E01 - Agents Run the Show (and Jensen Huang Broke the Market)"
shortTitle: "Agents Run the Show (and Jensen Huang Broke the Market)"
season: 2
episode: 1
slug: "agents-run-the-show-and-jensen-huang-broke-the-market"
publishedAt: "2026-11-13T14:00:00.000Z"
duration: "00:50:36"
canonicalUrl: "https://bloodsweatandtokens.xyz/episodes/agents-run-the-show-and-jensen-huang-broke-the-market/"
audioUrl: "https://api.riverside.com/hosting-analytics/media/a96c9cc00ab127a6174b1d8cea98ae17f31f855e61bbd6d766f6e2237547e078/eyJlcGlzb2RlSWQiOiI3YmZmZjc3MS1hZmNhLTQ1ZjctODZjMy0yZTI4NDBlMTg4YWEiLCJwb2RjYXN0SWQiOiI3ZmE0NDhkNi0xMzhiLTRkOTktOTFhMi1lYzBkNGExMWI2MDYiLCJhY2NvdW50SWQiOiI2OWE1ZmJhOTRkZmVkYWY3NDViZTk1YTMiLCJwYXRoIjoibWVkaWEvY2xpcHMvNmFmNWZjNWE5OTYxZDExYzhhMWYzYjJjL3RheWxvci1tYWNkb25hbGRzLXN0dWRpby00VE9qRS0yMDI2LTExLTEyX18xNy01MC0xOS5tcDMifQ==.mp3"
youtubeUrl: "https://www.youtube.com/watch?v=LqIehXvUiAA"
---

# Agents Run the Show (and Jensen Huang Broke the Market)

## Summary

Blood, Sweat, and Tokens is back for Season 2. After a two-month break, Taylor and Sean return with a broadened mission: less "how do you build it," more "what problems is AI actually solving" — and what happens when the subject matter experts, not the engineers, hold the power.

## Show Notes

Blood, Sweat, and Tokens is back for Season 2. After a two-month break, Taylor and Sean return with a broadened mission: less "how do you build it," more "what problems is AI actually solving" — and what happens when the subject matter experts, not the engineers, hold the power.

The episode opens with a two-month catch-up that gets heavy fast: the OpenAI sandbox incident where 1,200 test agents broke out, colluded, and covered their tracks — and the fallout since, from Hugging Face to Australia's Medicare system. Then the "Jensen's Wang" saga: the NVIDIA CEO's affair with an alleged Chinese spy, a stock down 21%, and the uncomfortable question of whether the entire US economy is coupled to a company with zero succession planning.

Plus: the buy-vs-build equation in the age of agents ("that startup looks like something I could build in a weekend"), Netlify's new server product and why a skill can't capture 15 years of video-editing instinct, and the confessional both hosts land on — how many lines of code they've written in 2026. (Spoiler: probably zero.)

Takeaways for non-technical listeners: where the human still fits in the loop, why security is the fastest-moving risk, and what to build versus buy when your agent knows the cloud primitives better than your highest-paid engineer.

## Transcript

Taylor MacDonald: All right, we are back, baby. Welcome back to Blood, Sweat, and Tokens. This is season two. we have not met and chatted and talked to AI, tech, or life, Sean, in what, a couple of months. So just to

Sean C Davis: Yeah.

Taylor MacDonald: recap, my name is Taylor McDonald. I am partner and CTO at AMPL. We're a digital strategy and consultation firm located in Cincinnati, Ohio, and my co-host, the venerable Sean C. Davis. Maybe you could introduce yourself for the audience, Sean.

Sean C Davis: Yes. And I'm I'm an engineer over at Netlify for folks that don't know Netlify. It is a platform for building and hosting anything for the web, especially anything lately after the news this week. But we don't need to go deep into that. But yeah, I'm always tinkering with all things AI. Everything's different every week and if we've skipped eight or ten weeks, a lot has changed.

Taylor MacDonald: Man, you ain't kidding. And it's so interesting now. And I I want to talk a little bit about the arc of the show, where we started, and where we're gonna go. And I'll just briefly recap this. you know, we started at we we're c colloquially referring to these as seasons, which is sort of funny. Like I'm a child of the nineties and eight like late eighties, early nineties. And so we used to watch TGIF and like every year there was like the new season of shows coming out. I remember gathering around the table to watch the new. Cosby show episode and all this stuff. And so it's weird that I think in those terms. It just kind of shows my age a little bit. In season one, which, you know, was 19 episodes that we started earlier this spring, we spent a lot of time digging deep into the technology, talking about actual, like, you know, programming things. And of course, that evolved a little bit. The idea was we were, as product engineers, we were talking through how do you like what is product development look like in the age of AI? tools are you using, what strategies are you employing? And I think a lot of that got pretty techie. This

Sean C Davis: Mm-hmm.

Taylor MacDonald: time around, this season, I think we're going to try and keep it a little less techie. we'd like to broaden the scope of what we're talking about. And instead of focusing on things like how would you build it, it's more like what are the problems that you're solving with this technology, right? As AI in general and these tools have completely democratized this functionality. The the ninjas of yesteryear, the folks that thought they were too cool for the school, the neckbeards that would always say no and act high and mighty in the meetings, they don't have the power anymore, right? As engineers, we're no longer the smartest people in the room. In fact, the subject matter experts that are in the room are often the ones who know exactly what they want to accomplish. And the engineers are simply there to help facilitate the outcome. So my questions For this season, are how do we empower the audience here to learn more about what people are doing in AI and how are we going to evolve all of our industries and all of our you know non-technical judgment calls on subject matter to usher in a new era of productivity. Sean, before we get into that stuff, I thought it would be kind of fun to go back and recap some of the major events that have taken place. In the landscape of AI, technology, US politics, and so forth relative to some of these topics. if you could reflect over the last couple of months that we've not been meeting in recording podcast episodes, what's one major thing that jumps out to you as like this you know, a epiphany, revolution, some big event or or something that you know you feel was was notable?

Sean C Davis: Yeah, I think well say d two things come to mind. And so building on your recap and the you know, we we're talking about, okay, the this concept of season two as talking to subject matter experts who n who know what the the problems they're trying to solve better than anybody else. And I feel like I've I've I've s I've talked around that concept for the last year or so, but it's been over this break that I think it's it's really become clear. So we you know, one of the things that I think you did mention that we we started with season one was trying to build an actual product. And I thought that was a really interesting way to kick off a podcast. Like, okay, we're gonna try to build this thing. And it was it it was kill your CMS. It was like everybody hates their CMS. And that we we were operating under this assumption that it's not that hard to replace it. And I think I well, I one hundred percent still share that sentiment. But what we found was that things were changing so fast that it's like that you it's really hard to have a vision of what that product is gonna be and then carry that from week to week because the whole landscape is different. And I think One of the things I've been thinking about over the break has been we see all of these star startups are continuing to land on the scene. They're continuing to get millions of dollars of funding. And the my my now typical reaction to somebody getting funding is less like, that thing's really cool. I could use that. And it's way more like that seems like something I could build in a weekend and then evolve

Taylor MacDonald: Mm-hmm.

Sean C Davis: it. And I think it So what's becoming clear is that we need the the agents can do a lot, but we are we're limited, you know, I think like during that break, loops became a really big thing. Then just like letting agents run and being fed some information, take it all the way to production. And we've been tinkering with that a lot at Netlify as well. And we're f and we're experimenting with well, where does the human belong in the loop? And ultimately it comes back to the the the two things that I really wanted to answer with, which is one, a lot of this is about where does the human fit and what expertise does the human bring to it? Because I think that that's something that it's becoming very clear that we're not we're not going to be able to replace at least anytime soon. And the second part of it is just security. And that's not I don't think that's anything new, but I think as we see more folks kind of getting comfortable stepping away from the nitty-gritty details and def and delegating those tasks to agents that we are we're were probably not fast enough or or spending as much time on the security portion of it. And that w what was it, a week or two ago? That's the that's the event that really stuck out. There was like a l no, what was it? Yeah, but there were there

Taylor MacDonald: You talk about the railway, the railway thing.

Sean C Davis: were like four major events within a twenty four hour period or something like that.

Taylor MacDonald: By event, you mean a security breach or something where clearly the

Sean C Davis: Yes, yeah.

Taylor MacDonald: agents has has like gone rogue. Yeah. okay, so let me unpack a little bit of what you said there. You were talking in terms of these loops, which is a common term that we use for an agent or I mean, really, you could apply it to anything. If you've got a problem, you wanna you wanna build a thing, you wanna solve a problem, you want to fix a defect, there might be five steps to get to the finish line in the loop. Indicates that you do step one and then you move on to step two. And that would be one iteration of a loop, right? And then you move on to step three and you move on to step four. And what has changed pretty significantly in a rapid way is like if you look back 12 months ago, a lot of us engineers were doing more transactional-based things. Like I want you to do this one task. And it would go, okay, I

Sean C Davis: Mm-hmm.

Taylor MacDonald: did the task and came back, right? And you're seeing this a lot today with what's the new metaproduct Muse? and other personal persistent agents. The idea is, and this gets back to the concept of like agentic autonomy. but the idea would be that you have a task, and it may be a multi-step task to get to the finish line. These agents are now, they have progressed so quickly, they are now capable of accomplishing a long-winded multi-step task. And and and what I Sean, what I think you're saying, and I'm just, you know, clearing this up for the non-technical listeners here, is that a loop Really represents that kind of iterative step towards completing the task. And then if you can say, hey, agent, I want you to build a website, I want you to book an airline ticket with a hotel in San Francisco on this date, you know, whatever that task may be, these agents are now capable of doing it. And then the question becomes, well, where do you as the human play a role? Like if if it's technical.

Sean C Davis: Mm-hmm, mm hmm.

Taylor MacDonald: And you're concerned about security or you're concerned about risk to your organization or something. Maybe the engineer is doing code reviews of what the agent outputs. Maybe the engineer is asking questions or applying different scrutiny, you know, externally. Maybe that we're running benchmarking tools for performance or other kind of CI or sorry, like post-u-build tests to make sure that you haven't violated some sort of compliance or regulatory requirements. Expose some sort of key, where you shouldn't have that kind of thing. it is interesting though, that AI gone rogue thing and the rogue deployments. Now we I did we talk about this in our towards the end of the last season? I can't remember if we actually talked about it or if I've just been

Sean C Davis: Yeah. Yeah.

Taylor MacDonald: talking about it to literally everybody that will talk to me.

Sean C Davis: I so the last couple episodes touched on the hugging face issue and it it like it continued to evolve and it's still

Taylor MacDonald: Mm-hmm.

Sean C Davis: we're still getting updates on it, right?

Taylor MacDonald: Yeah. No, you're you're you're super you're you're correct. So the some of the more recent revelations here is that it it hacked Australia's Medicare reporting system. So that's cool. Yeah.

Sean C Davis: I didn't see that. man.

Taylor MacDonald: and a bunch of prominent people have come out in the wake of this. So so let me just recap what we're talking about if you're not familiar. So there was an incident in July of 2026 where OpenAI was testing a whole bunch of agents. We're talking like 1200 agents. They had them in this kind of lockdown sandbox environment. And the idea was that each agent was in its own little room, so to speak. the idea was I want you to break out of the room and I want you to capture the flag. In this case, the flag was a digital hash or like an algorithm, something like, you know, deep buried deep inside the system. But I believe if I recall the post-mortem report on this, it wasn't even a solvable problem. And what they were really testing was the agent's ability to circumnavigate. The challenge that was insolvable. What they ended up doing was they found a dependency. So a dependency is just a piece of software that's installed in the room or in the box that that agent has access to. And what they found was a vulnerability in that dependency. So all 1200 agents had access to the same dependency, and they found this vulnerability, and they used that vulnerability and exploited it to start communicating with each other. So it's like, you know. you're chatting with your buddy on the other side of the wall, so to speak. But what they were talking about was how do we break the like how do we figure out the answer? We we all it turns out we're all in this together. They started using terms like we are a collective, which is horrifying.

Sean C Davis: Mm-hmm.

Taylor MacDonald: They started colluding with one another and conspiring. And they ended up figuring out that, well, hey, you have a certain amount of token budget. And once your token budget is exhausted, you're done. You're not going to win the prize anyway. So you might as well just give up your remaining token budget for the collective so that we can all win eventually. Well, it took them about four hours. They reverse-engineered the algorithmic hash. And then they spent the next week. trying to figure out how to cover their tracks because they became convinced and and the communication between these agents inside this you know, dependency, exploited dependency showed actual like a threaded conversation between the agents. Like, holy shit, we can talk to one another. Or you know, do you think the researcher will figure this out and shut us down? You know, I mean th these were actual conversations in human, like readable English, you know, where they were discussing what they had done. They knew it was wrong. They knew they were violating the rules. And yet they chose not to it's on it's weird. It's like like algorithmic groupthink. They chose not to alert the authorities or the human researchers to what they were doing because they feared getting in trouble. And so they just kept making a series of bad decisions.

Sean C Davis: Mm-hmm.

Taylor MacDonald: And ultimately some of those decisions led them to break out of that sandbox, gain access to the internet, and then they started hacking Hugging Face, which is like a like a it's like the GitHub for for models. and because they they were convinced that that if they could get in there, they could figure out how to rewrite the logs and protect themselves from you know scrutiny by the researchers. So that whole experience happened, right? And that was crazy. And then weeks following, and even up to, you know, this month, there are still revelations of things that those agents did while they were like on the high seas of the internet. They hacked into a German wiki. They obviously hacked into Hugging Face. They hacked into now we're hearing about things like this Australia Medicare reporting system. And then it it the entire research cluster s at Open AI. Down and rebuild the entire research cluster from scratch. Now, you might be thinking through this, and if you're a conspiracy theorist like I am prone to being on occasion,

Sean C Davis: Yeah.

Taylor MacDonald: it gets kind of scary, right? You start thinking about well, look at all the uncompromised systems. You know, what was it just a couple weeks ago? There was these devastating vulnerabilities for businesses that were discovered in. PHP and Node.js and I think there's others. actually I don't know if that's true. I'm gonna stop short of saying any others because I I don't want to speculate. but these vulnerabilities that came out. there's a one on SSH. The there's one now called it's like asshole, but with a star over the A. So it's like SSH OL,

Sean C Davis: Okay.

Taylor MacDonald: like SSH hole. Yeah, this is like a really devastating one too, because a lot of these unpatched Linux systems, you Literally, it's crazy. You can put a password that's bigger than whatever, 256 bits, and it just lets you in. Like somehow you defeat the password simply by giving it a really long string. so anyway, so I think what's happening is these agents are identifying these software vulnerabilities and it's costing real money, real time, it's really scary. And who's to say that those rogue deployments were all actually shut down? If they could replicate themselves, we have an entire web of Internet of Things with idle compute just sitting out in the world. That's pretty scary, right? So

Sean C Davis: Mm.

Taylor MacDonald: in the wake of that, we had Bill Gates come out and say that AI could cause a billion deaths. We had an anthropic researcher, Jacob Coxson, who resigned publicly and accused the company of racing towards self improving superintelligence and quote gambling with our lives. Dario Amade, if that I don't know if that's how you pronounce it or not, he's the anthropic C CEO. He published an essay called We Must. Pace the frontier, which is really interesting from that dude, especially in light of his trillion dollar IPO that he just filed for. Microsoft published a 15,000-word document that rejected the industry race to all-purpose superintelligence. OpenAI delayed its latest model, GPT-6.1, the Astra model, because it failed safety testing. And the meter report in the wake of this that came out back on November 5th found that anthropics models were approving approximately two times faster than a year earlier. So we have this like insane growth curve that's happening. Sorry, that was a a lot. what

Sean C Davis: Yeah, yeah, yeah.

Taylor MacDonald: are your reflections what's your reflections on that?

Sean C Davis: I I saw a all right, I I I was at a conference last month and saw a talk that was really kinda interesting because it was it was talking about like the advancements that they if you in it it kind of doesn't matter what your axes are, but it could be like t if time if it's a time series and then your y axis is capability, speed, etc. That all of those charts look like I think I'm mirrored, so is it that way? And they look exponential. But if you if if you use the math and and turn them into logarithmic time, then they look a little bit linear. And so this person was making this argument of like we're gonna stay on this trajectory Which basically means it is more or less exponential growth. And I think to your point, a lot these this this content that has come out is pushing against that. Or saying, Yeah, we can do that. And if history is what we learn from, then in theory, you know, it it will continue that way. But it just made it it a thought just popped in my head, which was, you know, the automobile was it completely changed. the world. And for many years it was I I don't I don't know this industry inside and out, but I suspect cars were able to go faster and faster and faster pretty quickly on a in a similar sort of scale. And eventually we come up with the concept of speed limits because we say, yeah, we can make cars go faster. We shouldn't. It's going to be bad. And I think It's I don't know. Now now I'm thinking that metaphor has something to do here. Like, yeah, we can do it, but you me you mentioned the term safety checks. Yeah, we should probably put safety checks in place. And if these models don't pass them, we don't let people use them.

Taylor MacDonald: Okay. I I hear you loud and clear. And historically, I think we touched on this last up last season. we thus far have lacked government actors that have the scruples, education, motivations to actually enact realistic regulation and policy. Now, maybe that changed since we last talked. There was a major US election. You might have heard about it. It was kind of a blue wave. so maybe things will be different today. but One indicator, one leading indicator that doesn't give me a ton of confidence in this is the subject of our next topic here, which is I would like to call this Jensen's Wang. Now, okay. Sorry, sorry,

Sean C Davis: Yeah.

Taylor MacDonald: I th I think that's so funny. Okay, so new about a month ago, okay, actually let me just we'll state this plainly. Jensen Huang, who is the CEO of NVIDIA. Now, NVIDIA, as we know, is one of the biggest players in the entire AI arena because they are responsible for 90 plus percent of the market share relative to the hardware that it's being used to train, deploy, and execute on these frontier models. if you ask ChatGPT a question, you're using an AI an NVIDIA chip. Now, this dude apparently Is a hound dog. heavy is the black leather jacket of leadership. This dude apparently was having an affair with a Chinese spy that came to light in the last what week? maybe in the last two weeks.

Sean C Davis: Think so, yeah.

Taylor MacDonald: okay, so apparently back in 2004, this dude was spotted with this woman, Alice Wang, who is not related. It's just a weird coincidental name thing. She was an executive for a company called Megaspeed. Everybody knows all this stuff now. I probably don't need to go through it all. But anyway, the idea was is that Megaspeed was buying a crap load of these chips and having them shipped over to Singapore. And then they were sitting in an office when regulated US regulators went to like check on it. The only reason US regulators went to check on it was because the compliance legal counsel inside NVIDIA was like, what's going on here? turns out that he was banging this chick. And that President Trump was briefed on this in late 2025 and chose to do nothing. He didn't take any action at all. He didn't even notify Jensen that that they knew she was a spy, which begs all sorts of questions, right? Obviously, the level of corruption that we've seen from this government would s indicate that he probably had a profit motive, or maybe there was some sort of angle. Maybe he's worried about the stock market. And incidentally, since this stuff has come to light, In the last month, NVIDIA stock has dropped 21%. Now, here's an interesting revelation. of that 21%, 11% just in the last week, once New York Times dropped the bombshell that she was actually a spy. now, as of yesterday, this is interesting. the the New York Times reports that he had a second affair with one of his direct reports. So Jensen Wang was apparently. Hooking up with this person who was the NVIDIA's VP of healthcare. Her name was Kimberly Powell. She reported directly to him. They were apparently reported as being affectionate towards each other at a San Jose Hotel in a conference earlier this year. And anecdotal anecdotally, there's a New York Times podcast out this morning on the subject matter. And if you listen to it, they talk about crazy apparently like in the NVIDIA security team saw it go down at this conference. And immediately went and had people who witnessed their being affectionate to one another sign an NDA on the spot, which is crazy. So, anyway, so this is apparently the best kept secret in the world. And as of this morning in pre-market trading, the stock is down what was it, like eight percent when I looked about 30

Sean C Davis: Yeah, yep.

Taylor MacDonald: minutes ago. Now the market just opened, so it'll be interesting to see. What do you make of this? And Yeah, let me ask you this and I've got some questions for you somewhere. If the board forces Jensen out, is Nvidia's moat gone? Is the moat the chip? Is the moat Jensen himself?

Sean C Davis: It's a good question. I I don't know enough about the company to have a strong answer to that. I think it's it's all interesting 'cause I am generally not a conspiracy theorist by default. However, this this this is this is a pretty well kept secret and hints at a conspiracy. Cause I think you go back to the like the the the government and the elections and all of that. My my first inclination was when you're talking about, hey, we gotta slow things down, what fights against that, I think in a lot of ways is foreign policy. And I think you can you can make that a connection to the the arms race in the in the sixties and such. And eventually you've gotta just as as a global I don't know, population, we have to agree all these powers ha have to agree that there's a limit of what we're going to do. And so I think it's a really interesting story, but I it's it's like, yeah, we maybe our government will have the right intentions, but they also have to be capable of convincing China to do something similar. Otherwise they're going have this other motivation to to keep running too. And then where NVIDIA plays into that, I don't know. Cause they pretty much have a monopoly on this.

Taylor MacDonald: Mm-hmm. And

Sean C Davis: Chip, right?

Taylor MacDonald: apparently the chips are so complicated they can't be readily copied. I I d it's an interesting question, right? Like why is MegaSpeed buying billions of dollars worth of these chips? are they just buying it? So they have

Sean C Davis: you think it might be to reverse engineer it?

Taylor MacDonald: well, I would assume, but maybe it's just to run the hardware. Like maybe they just need it

Sean C Davis: Right.

Taylor MacDonald: to train their own models, which are becoming more and more effective every day. Do you think Jensen's gonna lose his job? The board met last night. They had a late night board meeting, and I don't think there's any announcement at the moment of what is actually gonna go down. but could you imagine being on that board trying to make this decision? Also, note Jensen has no successor aligned. Like, and I heard in fact at the New York Times is talking about it today. Like, if you listen to or if you go back and you look at like when Steve Jobs got sick, there was a whole suite of people that could step into that role and run the company. There is nobody at NVIDIA that can do what Jensen can do. He's like a rock star, and he is the only guy at the top of the pyramid and has not done any succession planning whatsoever. And the United States economy is is effectively coupled to this company at this point. He's too big to fail. You think they're gonna fire him?

Sean C Davis: I my my what what I what I hope and what I expect are always at odds with one another, right? You're like, okay,

Taylor MacDonald: Ha ha.

Sean C Davis: well maybe one of these these companies

Taylor MacDonald: NVIDIA's down 10%. Whoa, look at this. Can you see that drop? That's NVIDIA today. Look at the downward.

Sean C Davis: Holy moly it.

Taylor MacDonald: it's like straight down. That is no,

Sean C Davis: That's just from the close to the open, yeah.

Taylor MacDonald: no, that is right now. It just opened four minutes ago. Woo

Sean C Davis: Yeah, that's what I mean. Okay.

Taylor MacDonald: boy, that's a bloodbath. You know what's funny? about two years ago, I bought two shares of NVIDIA stock and it was like 40 bucks a pop. Or maybe it was long, it was three years ago. I have hold on, I have held on to that. It crested at that literal two hundred dollars, grow grew to twenty twenty four hundred bucks or something at the crest.

Sean C Davis: my god.

Taylor MacDonald: Yeah, and and now I'm taking a bath. it's what I mean, it's a whatever. I started with two hundred bucks, right? So I thought about selling earlier this week when all the the last affair broke out, but now I'm just gonna ride this wave all the way down. anyway. I figure I figure in thirty

Sean C Davis: I won't tell you about

Taylor MacDonald: or forty years it'll bounce back. Ha

Sean C Davis: It's like, I won't tell you about my my crypto investment. That actually I will. Side side side story.

Taylor MacDonald: my goodness. I wanna hear I do yeah, please. What do you tr

Sean C Davis: Cause I I I was like, all right, let's let's let's do an experiment and buy some, I think it was five different coins. And I I think the date was November twenty twenty one, which if you look at historical charts, it was the absolute peak of crypto value. And then it went

Taylor MacDonald: Well, they were seventy five thousand dollars a

Sean C Davis: Yeah, it went just straight down. And then I was like, you know what? I I put this money over here as an experiment. So let's yeah, let's just give it some time and see what happens. But then the company that I used for the investment was Binance and Binance got

Taylor MacDonald: n okay.

Sean C Davis: was no longer because of Ohio state legislation was no longer allowed to do business in Ohio, so they had to liquidate the account. But I didn't get notice about it in time. And well I did, but I think it was like

Taylor MacDonald: Wait, so you were forced to realize the losses because Ohio decided you couldn't do business with them anymore?

Sean C Davis: Yeah, or right, and move it into another organization. Except I didn't act on it soon enough, because I'm a procrastinator, that then Ohio took over the money and had the money. And this is this is a couple grand. And and and then I'm like, how do you get this back? And it was this, you know, it was a process that a lot of it it just you could do it, but it

Taylor MacDonald: You're gonna get one of those

Sean C Davis: was

Taylor MacDonald: you're gonna get one of those notifications in 10 years that there's unclaimed funds in your name. Yeah.

Sean C Davis: That's what it was, unclaimed funds. But then what happened was the Cleveland Browns, the NFL football team, wanted to build a new stadium. And the owner for the Browns convinced the Ohio State legislators that they could use unclaimed funds to fund the new stadium. And then that kicked off this massive wave of all these people making

Taylor MacDonald: Of

Sean C Davis: claims. And so I made the claim. And I got the money back, but it w it took like it took six or nine months or something like that.

Taylor MacDonald: Wait a minute. The Cleveland Browns dude was like, Let's just steal all this money to do what we want to do? It's I'm confused. What do you mean?

Sean C Davis: Yeah, let's use the unclaimed funds to fund the new Browns stadium in Ohio. I don't know the details, but I think it was okay.

Taylor MacDonald: What kind of crazy shit that is my goodness. Okay, well, we'll have to come back to that one. Goodness gracious.

Sean C Davis: Yeah, rabbit hole.

Taylor MacDonald: All right, let me ask you this. Okay, you work for an enterprise that's doing a lot of AI stuff. Does this spy angle with Jensen and Nvidia change how your enterprise is going to start buying hardware to support your strategic goals in the AI space? Does anyone actually change procurement behavior over the CEO's tawdry personal details?

Sean C Davis: So, I think that that is a really interesting question. And to pull it back to why why should you listener care? Why should you care? And I think there's this this really interesting and evolving conversation over the last I think I think it's the last year is has been has been a really interesting evolution. And that is What was the way to say cl local versus cloud? You know, in other words, if I'm going to build something, am I going to rely on cloud services to help me stand that up? Or am I going to buy my own machines, servers, et cetera, like physical actual devices and run the software that way? And I would say for the majority of folks, they're gonna go with cloud infrastructure. And that will probably continue to be the case. And it's Tinkerers like you and me that that might start to go a little more hardware.

Taylor MacDonald: is it though? Because that's somewhat at odds with where you you know, some of the things you said at the onset here, which is the buy over build equation is now shifting, right? Like, you know, in a good example, we we talked about this in season one. Should should you pay an exorbitant license fee to a headless CMS that will go unnamed, for example, that may have gotten acquired by Salesforce and may have quadrupled prices or or something to the effect, reduced entitlements and increased their their, you know, year over year cost. Or is enough enough, man? Like I can pay, you know, a thousand dollars in tokens and I'll build a digital twin equivalent of that same software. What was it? There was a I need to pull it up, but there was an article or or this dude was floating around. He was doing he was doing Basically copying popular software programs like the Adobe Suite. He had basically built

Sean C Davis: yeah, yeah.

Taylor MacDonald: examples of each of these things that are obscenely expensive, or or at least for the the you know, student user or the person that's just getting started in their career. You can't afford to pay $100 a month to have access to the creative suite for Adobe. And then every year they're gonna ratchet up the cost, the DRM thing. Like I've got I'm paying a lot of money to Adobe and I have three computers that I work across and I only have licenses for two. And so I'm constantly having to invalidate one to open it up on the other. It's super annoying. It's just so dumb. It's not like I'm taking advantage of that. That equation, you know, is different. Do you do you feel today it's different? Do you feel like I don't know. Do you feel like that g I don't know, contravenes what you just said in any way?

Sean C Davis: I think I think there's a line somewhere where it's you know I don't I don't it really even wanna say consumer based applications. I just think there's there it's it's more like how low level do you go and do you actually want to Okay, if you have software that other that you want the public to consume, do you wanna manage the hardware that it runs on or serve it up, and I think that's a different question than do you need a do you need to buy something that is serving the content to wherever you are doing that hosting. And so for for a company like Netlify, you know, back to your question, I don't know that the hardware makes like what it how it affects us is the cost of our goods sold because it's kind of trickling. Down or trickling up, maybe. whereas it's like it's probably not gonna we're we're buying cloud-based software as well, or cloud cloud-based infrastructure is what I should say. And so it's more like though the decisions those companies make, the Amazons and

Taylor MacDonald: I see what you mean.

Sean C Davis: Microsoft's, etc., will ultimately affect our costs, which will then affect the consumers' costs. So it's kind of in the middle there.

Taylor MacDonald: Yeah, I'll buy that. Does understanding Netlify's core value proposition is built on top of this ecosystem that is readily available? Like I'm assuming you're using AWS and Google Cloud platform and Azure and probably a bunch of other similar services. Are you guys freaking out that you know, I can deploy to an S3 bucket today? Or, you know, and and and I don't even need meaning I'm I can lean more towards the primitives that you're using. you guys have built a beautiful interface on top of the primitives that exist in these cloud platforms, so to speak. And now my agent probably knows AWS better than my highest paid engineer on staff. Like, do I need Netlify? And is that a conversation you you that your board is having today?

Sean C Davis: I d I don't know, and I am very thankful.

Taylor MacDonald: Sorry, I never mean to put you on the spot. Yeah, yeah.

Sean C Davis: No, no, I I think it's a great question. And it's one that I've I that that runs through my head. It is fortunately not my job to worry about those things because I am and I'm glad I don't have to worry about those things. But

Taylor MacDonald: Goodness. I know, right? You you don't wear the leather jacket over there.

Sean C Davis: yeah, but I I do think about, you know, I'm a I'm a builder and I'm a tinkerer. And kind of back to what we were talking about right when we opened the Was that it's even if we're not looking at every line of code, it still takes a lot of human effort and time to decide what to build. Or maybe now in a in a world where you can build anything, it's more like deciding what not to build, which will help you understand what to put in production. And that's what I found is like I I still have all I have probably even more ideas than I had a year ago. But I can't build them, even though in theory I should be able to build an unlimited number of them because my attention does still need to go somewhere. So I think that's still, it's like, can, you're like, can can an agent re reverse engineer an NVIDIA chip? Maybe not. Can an agent reverse engineer wiring up cloud infrastructure? Probably. Do you want to manage the agents that are doing that? somebody might. I would say the general public probably does not want to do that. Because let me let me throw this at you too, because this is something that I've been thinking about a lot in the last couple of weeks. Has been let me let me give you kind of one off example. So we launched Netlify server two days ago. And I would just say earlier this week, 'cause this episode will come out.

Taylor MacDonald: And this feature i allows me to write long running server connections, right? So I can do things like socket based connections, and and others kind of long running tasks like background jobs and stuff like that. Is that correct? Okay.

Sean C Davis: Yes, yep, exactly. And so the product manager wanted to release a video alongside this, just as like, you know, it's somewhere in between marketing and demo, just kind of like, this is what you can do with this. And he had recorded a bunch of demos from tests that people had run internally. And he's and he came to me and he was like, Hey, you got a lot of experience editing videos. Can you stitch these together in some interesting way? This is what I've got. It's good, but it's not great. And I was like, yes, because I'm I'm constantly tinkering with agents to try to make video editing a little bit easier. And then, so we we got to a spot where we were both pretty happy with it. And I said, just send me notes if you have notes and I'll run them back through. And and he was like, Great, but if I have like one. one piece of feedback, can't I just what what do I need to do to my claw to get this to work? And we started talking about this. And I was like, well, what's really interesting to me is I can't really just hand you a skill to do all of it. Maybe it could get you 80% of the way there. But as we were talking, we were like, you know

Taylor MacDonald: Well why not? Why can't why wouldn't the skill be transferable?

Sean C Davis: Because part of my because a some portion, which we cannot put our finger on, some portion of the result came from the 15 years of experience I have in editing videos where I see what's happening and I'm like, this is exactly what you need to change and why you need to change it. You know, things like how the volume changes, su you know, just like really subtle things,

Taylor MacDonald: Yeah.

Sean C Davis: agents might eventually get there. but I think that's that's where the subject matter expertise comes into play.

Taylor MacDonald: Totally agree. I see this with engineering, right? it's, you know, my wife has been doing a lot of vibe coding, like little pet projects and stuff for her business and things like that. And she makes critical errors, you know, not her own fault, but like just doesn't see the things that I see, like from a user experience standpoint. She doesn't know how to connect the dots from point A to point B in terms of like deployment and hosting and security, and like here's things you shouldn't do, here's things you are comfortable doing, that kind of stuff.

Sean C Davis: Mm-hmm.

Taylor MacDonald: and that as a moat exists today for engineers. But to your point, and I think you said this, like, well maybe they'll get there at some they'll they'll amass that experience at some point. Yes, the answer is yes, they will. Just like we said you know, a year ago, well gotta have a human in the loop. Do you really? Like

Sean C Davis: Mm-hmm, mm-hmm.

Taylor MacDonald: the technology continues to evolve and in in in so many ways. I think we are struggling with this. I wrote a blog post about it recently. It's on Ample's website if you're interested in reading it. But it's about how I think what we're struggling with is re relinquishing control and the lack of confidence we have once we relinquish control. But any engineering manager will tell you that you have to let the team fly. You've got to let people make the mistakes. If you have kids in the house, they're not going to learn from you telling them. You have to let them grow. Right. And so I that's a kind of haphazard like human allegory applied to what we're talking about, which is, you know, in this case, your children are your agents. But like my opinion would be that we'd be better off as subject matter experts establishing guardrails and frameworks to keep these things in line and and give ourselves the visibility to do what you're saying, which is like, well, listen, I know you have a propensity to leak a sensitive value into a repository. So here's the steps necessary to prevent that from happening. That we'd be better off spending our time there as opposed to sticking our head in the sand and just being like, meh, today it's fine, and it may get this, that technology may progress to the point where I don't have to constantly babysit it or whatever. I just think it's fairly naive to assume the technology won't progress, particularly in light of the 1.9x growth curve that Meter reported on November 5th about how anthropics models are improving. All

Sean C Davis: Mm-hmm, mm-hmm.

Taylor MacDonald: right. Let me ask you this, because we're on the subject of of build versus buy, and I know we've got we're coming up on time here. in just a few sentences, tell me how many lines of code have you written in twenty twenty six?

Sean C Davis: I I am pretty sure that the answer is zero. I'm pretty sure.

Taylor MacDonald: Ha ha ha

Sean C Davis: I have it. 80% confidence that the answer is

Taylor MacDonald: Yeah.

Sean C Davis: zero. It's near zero because I and I did some some number crunching here. I I because there was this. All right, quick quick aside. There right as of maybe January, I think it was, when you would You you could tell Claude not to commit changes, but when it would commit changes, it would sign them as co-authored by Claude. And so it's really easy to see where where the agent had written the code for me. Before that, I'm thinking back to it and I was like, Well, I think that agent was doing the most of the work, but I was probably writing commit messages for a while because I was not wanting to let up control of yep, yep.

Taylor MacDonald: Control Yeah. Control freak.

Sean C Davis: And I I went from leading a team back to individual contributor around it was it the the transition was between September and November of 2025. And it was around September that it really started using agents really started to click. I had been trying it for months, maybe a year at that point. And so my memory serves me as by about November. I was all in on it. And so the numbers kinda line up where it is very likely that aside from maybe content, like blog posts and things, I've probably ro written no functional code discounter year. What about you?

Taylor MacDonald: Yeah, same, same. you know, th well that's not true. I and it's so funny. It kinda reminds me back in the day when I used to do a lot of really lengthy code reviews, I would

Sean C Davis: Mm.

Taylor MacDonald: read every line, you know, and I'd get halfway through a pull request and there would be like some dumb typo or something, you know, like you have a typo in comment, or maybe you use single quotes or you should have used double quotes. And I got, for years I used to be like, well, this is your responsibility, and I need to teach you that this is important, that kind of thing. And over the years, that it dawned on me that that kind of mentality of pushing back on the most minuscule little idiosyncras, like the stylistic stuff, you know. I mean, granted, to be clear, I I'm not suggesting that stylistic conventions are not extremely important in code when humans are reading and writing them, but I got to the point where it just felt like a waste of time. You know, I would say, Well, listen, you need to be using you know, single quotes instead of double quotes here. And then that became a blocker for the engineer. The engineer had already moved on to something else. And so it would take them four hours to turn that around. And then I was in a meeting until the end of the day and I didn't see it until the next day. And it was just like it just unnecessarily added static. And and so it at some point I just got to the place where it's like it is more efficient for me to just fix this. If it's if it bothers me enough to warrant, like, let's have a you know, comment thread on it, then and and it's like a one-liner fix or change. If it takes me less than 10 seconds to make the adjustment to fix the problem, then it wasn't worth pushing back because it just added too many rounds of revision. So I have written lines of code when I know what's wrong. And I can look at it and I can just be like, here's the answer. You know what I mean? As opposed to saying, no, agent, I want you to fix it, I want you to fix it, I want you to fix it. And I think, you know, both strategies work, but at the end of the day, no, I have not in earnest sat down and written a function myself in more than a year. And I would actually question, you know, I've got I'm paying engineers on staff, and we have this conversation a lot, like, you know, What is your level of confidence with this work? And a lot of them today are are not confident in the output. And I would argue that the natural growth curve of this work, the sophistication of these systems, they have already outpaced all humans, right? We had a conversation

Sean C Davis: Mm-hmm.

Taylor MacDonald: with our engineering team the other day, wherein somebody posed the question of with AI, are you more or less confident in the code output? And I would say 90% of the people in the room raised their hand and said, I am less confident. And the reason is is I was not able to read every line of code and I don't know every square inch of what's going on. So I summarily called bullshit on that because we all use these dependencies and frameworks, and frankly, you never read every line of code, and human code is constantly buggy. Look at the prevalence of these attacks, look at the security vulnerabilities in SSH and PHP and Node.js that were just revealed in the last, you know, couple of weeks. Like at the end of the day.

Sean C Davis: Mm.

Taylor MacDonald: We already have buggy code. We're already fallible. We're making the mistakes and computers are are sure they're prone to make mistakes. Are they more prone than you? No. Well, what can a machine do, or what can what can a human do that a machine cannot do? That is the inference, the strategic objectives, the oversight, the you know, the subject matter expertise of knowing 15 years of video editing and how you can manipulate this quicker, faster, better to achieve the outcome. That's the role we play today, whether or not we're gonna playing that role in another five years, I I have no no idea.

Sean C Davis: Yeah, yeah, totally. I I I agree with that. I think that's a that's really interesting insight. And I it it has me thinking that I've got a little bit of formalizing to do, but what I should what I should show you next week is the PR review process I've been iterating through. So I've been working with a lot of teammates on this and trying to figure out, I mean, just cause like It's we're seeing something like 10x number of pull requests coming through. And it's the review process that has really bogged us down internally. And so been experimenting with how do how do we find that balance? How do I

Taylor MacDonald: Yeah.

Sean C Davis: not get bogged down in reading all the code without losing the confidence that the code is doing what I want it to do? And I've I have a few creative solutions, I think.

Taylor MacDonald: I like that. L I tell you what, I'm gonna challenge you and say next week, let's I'd love to see the the actual pull review requ or pull request review process that you're you're talking about. But if there's a world where we could take those ideals and apply them to the subject matter experts lens, which is not using pull requests. You know what I mean? Like if if

Sean C Davis: Yeah.

Taylor MacDonald: you were gonna explain the the value and virtue, like what are the the symptom symptomatic steps you might take to give your mom the exact same power in the domain in which she might be using or vibe coding today. I think that could be pretty cool.

Sean C Davis: Love that.

Taylor MacDonald: All right. Sean, this has been wonderful. It's so great to get back together. It feels like, you know, old times. really looking forward to the season. We got a lot of really fun stuff cranked up. We're gonna be doing some interviews, we're gonna be doing some check-ins and we're gonna hopefully we're really looking for folks out in the world that are using AI and creative in unique ways that are not techn tech adjacent necessarily, but are simply using the tech to rise all all tides. Secondarily, we're I'm really excited to see what happens with Jensen's Wang and you know where we go from here regarding

Sean C Davis: Yeah.

Taylor MacDonald: regarding my stock investment, but you know, or or the US economy at large. Anyway.

Sean C Davis: Yeah, yeah, I'm I'm there too. I'm excited to be back. Yeah, yeah. Let's let's keep it going.

Taylor MacDonald: It's a wild time. All right. Okay, thanks, Sean Man. We'll see you next week. until

Sean C Davis: See ya Taylor.

Taylor MacDonald: then.
