B (2:42)

The Citrini sell off is here. Citrini popcorn.

A (2:44)

Despite tracking to close to billion. Isn't it a billion monthly orders. Something insane.

B (2:53)

My experience with it was you had mentioned to me that, oh, it's the current thing. And I was pretty offline. And then by the time I actually started refreshing the timeline, I was like, oh, I'm clearly stuck on some search feature because I'm only seeing Citrine posts and posts reacting to it and reacting to the reaction. And it had done a full news cycle, both a backlash and then a backlash to the backlash. And Tae Kim's getting in there fighting and people are posting rebuttals and someone posted a fully AI generated just like turn it up another notch version which is hilarious. He's like, if that wasn't extreme enough, I got something even more extreme for you. People were having a lot of fun with it. Sunday is for the posters. I guess people are having fun. What should we read through of the actual Citrini article? Because it is sort of long and we do have someone from Citrini coming on the show so we can maybe go through some of the reactions. I mean the futures were red last night. The market is down broad.

C (3:54)

Yeah.

A (3:54)

And it's interesting, a lot of people were saying, guys, there's no way that futures are red just because of this Citrini essay. That is obviously science fiction.

B (4:04)

Yeah.

A (4:04)

And then it turns out Bloomberg this morning came out and actually stated that.

B (4:10)

Reported it.

A (4:10)

That it's the Citrini selloff.

B (4:12)

Yeah. Yeah. But I mean there were also the tariff things. The, the tariff news was sort of digested on Friday, Saturday. So there was totally like that could be possible. There's a lot of other stuff going on. Joe Weisenthal has a post here. The Citrini sell off. The quote from the terminal. Software payment stocks slide after citrini post on AI risk. DoorDash and American Express led declines in software and payment stocks on Monday. And we were debating like PayPal had gotten beat up and there was a question about what is the strategy going forward. PayPal's always been an interesting situation because like absolutely goaded founding team. But they're all disinterested in jumping back in and turning it around.

A (4:57)

Yeah. You have Max with a firm who over a long enough time horizon will probably compete in every.

B (5:04)

Yeah. And it's hard because Even if you went back, how do you build a position? How do you get control of that company? But it does seem like there's some value there, there's some stickiness, and it'll be interesting to see. Does it go private? Does the new management come in? I mean, they already have some new management, so I'm not exactly sure where it goes, but it feels like the payment rail thing will be sticky for a while. Young Macro had a take quote tweeting Citrini Good and interesting piece, but a necessary caveat is that it's essentially a hypothetical conditioned on severe institutional failure rather than some sort of macro inevitability. As the pie itself notes, one of its two failure modes, liquidity stress and capital impairment, includes a liquidity component that the Fed can address quickly with liquidity facilities and asset purchases, repo lines, quantitative easing as seen in recent episodes of banking stress. Losses from impaired assets won't disappear, but can in principle be moved where the broadest shoulders are. It's an odd situation because you're seeing all of this capital evaporate from the public tech markets, but the labs aren't public yet, so they can't fully absorb it. Like they sort of can, but it's much more opaque. And it's not like most the average investor can't just read this piece and be like, okay, I believe it. I am worried about software stocks, so I'm going to rotate into a basket of anthropic OpenAI and SpaceX. They don't have that option yet. They might by the end of the year.

A (6:34)

You can buy Nvidia, you can buy Google. There's plenty of ways.

B (6:38)

Yeah, totally, totally. Just look at Leopold Aschenbrenner's allocations and copy trade that I suppose plenty in this hypothetical economy, much more than before, he says there will be plenty of broad shoulders. Couldn't they just tax Anthropic into paying all your pensions? In this hypothetical scenario, this would be done through regulatory or fiscal mechanism. The second failure mode, an aggregate demand shortfall for massive unemployment, can be addressed through fiscal policy transfers, wage subsidies, et cetera. The piece argues that this may be constrained by falling tax revenues, but in a deflationary low inflation environment, the treasury can run large deficits and the Fed can buy as much of that debt as it needs to. Leaving aside that we'd expect the tax structure to change, the political process is unlikely to be as meaningful as a bottleneck as the piece claims, as the hypothetical fiscal hawks pointing to unsustainable deficits would not have Much of a point given the economy. And this hypothetical will have much higher potential output. Also, that obviously wouldn't poll well. The first principle is intuition obviously tells you something is awry. When we're told that people will want at least as many real things as before and the economy will have the means to produce more real things than before, but the people won't be getting the amount of real things that they want or need. This is because that typically requires major policy institutional frictions or delay in translating capacity into purchasing power. If you suddenly have two loaves of bread in your house instead of one in your house, and you weren't starving with one, you probably shouldn't starve when there's two. But if by some hypothetical, through the complexity of the novel two bread loaf production process, you suddenly get tangled and can no longer access the cupboard, then it's quite possible you will starve. I thought that was a good take. It is interesting how this moved the markets way more than AI 2027. Like it's sort of the same piece and it has a lot of the same sort of extrapolations based on AI progress. A lot of same.

A (8:43)

But like AI 2027 probably like the market reaction was like, invest more in the labs at the time because it was kind of aimed for more of a west coast audience in general.

B (8:56)

Yeah. I just mean if you had taken AI 2027 and you had just asked like, I own this basket of public company stocks, what should I be doing? And AI 2027 is your backbone, you would probably sell a lot like you're selling right now. Like the AI 2027 situational awareness PDFs. Like, those are the Leopold Aschenbrenner philosophy that is now being reflected in the market. But you know, he didn't go long any of these stocks. Anyway, sorry, really quickly.

C (9:28)

I wonder how much that is just because of like, who is actually writing it. Like the total people are like, you know, very San Francisco coded, like vaguely ea, maybe safetiest people. Where this is like, okay, they're like

B (9:39)

a financial research firm and it's written with a financial audience in mind and it speaks in that language. So it's been somewhat translated. Interesting that it took almost a year for it to be translated in this way. But then, yes, it's. It's interesting for a couple reasons because it didn't. This was not the takeaway. And I remember in the Vanity Fair piece I had some funny quote in there being like, the market should be moving off of what happens on the door Cash podcast And it seems sort of silly, but I think that we are now seeing the downstream ramifications of that.

A (10:16)

So, Jordy, I did think it'd be helpful to kind of provide a summary of the essay. I think the original essay, it feels like they use quite a lot of AI to write it. I'm going to use AI to summarize it before you.

B (10:27)

There we go.

A (10:28)

So anyway, so this scenario, you should have got to this at the beginning. But by late 2025, agentic tools become vastly better at coding and complex tasks. Obviously that was. Yeah, it's historical. Firms found they could use AI to replicate work normally done by humans, radically cutting labor costs. Productivity looks great on paper. GDP and productivity metrics soared because AI output counted in the official numbers. But most of that value didn't translate into real consumer spending. So like, businesses are spending money, but they're spending it on data centers. And that is not as opposed to labor, where if you give somebody money, they'll buy a house, they'll do home improvement, they'll buy cars, they'll put their kids in school.

B (11:06)

Consumption.

A (11:07)

Consumption. Right. So they're calling this ghost GDP economic output that doesn't actually circulate in the real economy. And then they identify an emerging negative feedback loop, which is companies lay off white collar workers and reinvest savings into more AI. Displaced workers have less spending power, consumer demand weakens, especially for discretionary goods. Companies facing weaker demand invested even more in AI to maintain margins. And this creates a negative feedback loop with no natural break. The next step is market and credit stress. So they talk about private credit having lent to a bunch of these different SaaS, companies that are now being threatened. Defaults climbing as this sort of like perceived recurring revenue ends up not being fully recurring. And next intermediation and friction collapses. There's a whole segment talking about how like a lot of value capture in the world is actually just humans, like not wanting to deal with frictions. It's like not switching car insurance because even though you know you're paying more than you should, it's just kind of a hassle. But if you could have an AI agent go and do that, then maybe you do that more often and that pulls out some potential earnings from the system. The other thing they talk about is all these different tech hubs and how many prime mortgages there are that might not be so prime if there's a layoff and somebody ends up having to switch career paths or something like that. Generally talking about unemployment surging, consumer spending collapsing, severe drawdown in the stock market and then even our sort of normal economic indicators hiding the sort of overall weakness. And anyways, takeaway, AI being great and powerful may not equal all the markets ripping, but.

B (13:04)

Well, two different things.

A (13:06)

Market ripping, individual individual companies and median income. Yes.

B (13:12)

Like they're two, they're three wildly different things. You can see asset prices rise massively based on future promise of GDP growth. If it's guaranteed that GDP growth is going to happen 10 years from now, the market will price that in today. And then if all of that GDP growth goes to one person, you're not going to see median incomes rise. You're not going to see like broad prosperity in America. Yeah, I have more on that. But let me tell you about Okta. Okta helps you assign every agent a trusted identity. So you get the power of AI without the risk. Secure every agent, secure any agent. So it's worth really quickly.

A (13:44)

Yeah, worth maybe noting like the kind of companies they call out in the piece by name. So poor all these companies. Let's check in with ServiceNow, which is, which is one of the companies that was most heavily. Yep. Down for almost four and a half percent today. So they talk about a bunch of the SAS tools they call out Monday.com, asana. Zapier. Zapier is kind of funny because in some ways like it was just like work automation before work automation was even that cool. Yeah, I was using, using Zapier 10 years ago. It does feel like they're kind of in a, in a decent position given that like they're already their core businesses like help people automate different workflows. But. And then doordash, we already talked about that. MasterCard and Visa, they talk about suffering revenue pressure because an AI agent would just opt for stablecoins, which feels like again, something that many people on the show have come on and made the case for. Why agents will leverage stablecoins or prefer stablecoins seems, seems like a possibility, but unlikely that that will just, you know, all payment volume will shift over there overnight. And then amex. Amex they called out specifically because of their consumer base being just generally weakened by labor displacement. And then a bunch of others. Travel booking, insurance, real estate, tax, et cetera. Travel booking I thought was funny because most travel agents don't actually take fees from the consumer. They take fees from the side of the hotel, the airline, whatever. And so the idea that you'll just immediately get like every AI agent or every individual will just immediately. I didn't fully process that one. And anyways, back to you. I did think that Jon Lober, I wanted to go through his piece because I thought he had one of the better rebuttals to the piece of work.

B (15:54)

Yeah, I thought that was good. I think the thing that just keeps sticking out to me is like, and I was debating with Sagar and Jedi about this as well. He was telling me like, AI is the only thing holding up the economy. I was like, no, AI is actually doing very little for the economy right now. It's doing a lot for the markets, it's doing a lot for the future. But in terms of the actual economic impact of AI, it's very low. And we just know that because you add up the actual AI revenues from the AI labs and you're talking about 30, $40 billion and okay, maybe there's like a 5x multiple on that. And so you're generating $200 billion of GDP on top of those tokens. But like, that's just not that much in the grand scheme of the actual America's gdp. And so there's this disconnect between like the market, which is pricing future gdp, future cash flows, future value creation. Then you have what is actually driving GDP today. And then you have like the actual workforce and what, what Americans do. And so there's this, there's this odd disconnect and I keep coming back to the Tyler Cowan, like slow takeoff philosophy around like what's actually holding up the American economy. It's like health care jobs, and there's a lot of jobs that are, they feel very AI resistant. I don't know, maybe something changes, but like, it just feels like, like the number of people that are software developers, less than 1% of America, the number of people that work at tech companies broadly is less than 10%. And so even if there's some massive reallocation there and then you go into, even in white collar, if everything shifts, the rest of world is hit. And there's just a lot of other dynamics that feel like you can see crazy gyration in the markets and you can see really quick reallocation of 10% of capital, billion of dollars flowing around. But that doesn't immediately translate to what is happening in the real economy. There's always this disconnect.

A (18:00)

I'm laughing a little bit because we've seen so many short seller reports over the last few years where somebody accuses a company of really, really bad, potentially illegal behavior and the stock will move down like half a percent. And then somebody writes like kind of a cool piece of science fiction. But like easily can poke a million holes in it and then it every, you know, it sends like all these mega caps really funny down. Let's go through this piece from John.

B (18:29)

Before we do, let me tell you about Cisco Critical infrastructure for the AI era unlock seamless real time experiences and new value with Cisco.

A (18:38)

John Loeber wrote a great piece very quickly after this called Contra Citrini and he says popular markets commentator Citrini recently published a compelling and popular piece of AI doomer fiction, admittedly with some small probability of occurring. But I'm old enough to have seen many cycles of economic doom saying I want to present a critique of Citrini's work and show a much likelier, more positive view of the future. One never underestimate institutional momentum. In 2007 people thought the US was geopolitically done under peak oil. In 2008 they thought the US dollar was just shy of collapse. In 2014 they thought AMD and video were done. Then came ChatGPT and they thought Google was done every time. Existing institutions with momentum have proven themselves far more durable than onlookers thought when worried about institutional turnover and rapid labor displacement. It's very funny that Citrini writes, even places we thought insulated by the value of human relationships proved fragile. Real estate, where buyers had tolerated 5 to 6% commissions for decades because of information asymmetry between agent and consumer, people have been calling for the end of the real estate broker for 20 years. You don't need superintelligence for this. All you need is Zillow or Redfin or Opendoor. That's exactly this example actually shows the very opposite of Citrini's point. We have the type of labor that most people consider obsolete. And yet market inertia and regulatory capture have made the real estate broker far more resilient than anyone would have bet a decade ago. My wife and I bought a house a few months back. The transaction required us to have an agent, ostensibly for the above reasons. Our buyer's agent made about 50,000 on the deal for about 10 hours of form filling and party coordination that I could have done myself. This market will eventually be efficient and prices labor fairly, but it takes a long time to get there. I know a lot about inertia and change management. I built and sold a company that focused on moving insurance brokerages from service to software, and the main thing I learned is the iron rule of dealing with human reality. Everything is always more complicated and takes much longer than you think it will. Even if you already know about the iron rule. That doesn't mean that a meaningful change in the world won't happen, but that the change will be more gradual, giving us the time to respond and adjust. Second point. Software has infinite demand for labor. The software sector has been struggling in recent months as investors fear that companies like Monday Salesforce Asana can now be easily replicated and that the value of their backend systems is indefensible. Citrini and others talk of AI coding as a spell of the end of jobs at SaaS. Companies are 1 the products become obsolete, 0 margin and 2 the jobs themselves disappear. What everyone seems to be missing is this. These products effing suck. That's his opinion. I can say this because I've actually spent hundreds of thousands of dollars on these products. Sure, maybe AI enables competition to replicate their products, but more importantly, AI enables competition to deliver better products. It's no surprise to see the stocks drop an uncompetitive sticky lock in sector filled with another swear word and becoming competitive again. And my own personal call out here is even like, until we see a round of layoffs at a company that is 5000 software engineers at once, it's hard to believe that AI is replacing software engineers versus just making them a lot more productive. If somebody's a lot more productive,

B (21:59)

you'll

A (22:00)

pay at least the equivalent amount to maintain them.

B (22:02)

Yeah, interesting.

A (22:06)

More generally, it is uncontroversial that virtually all current software is garbage. Everything I use and pay for is littered with bugs. Some software is so broken that I can't even pay for it. I have not been able to send a wire using Citibank's online banking in three years.

B (22:18)

This was my pushback against Rune. Rune was like, was like, oh, like Codex is so good. Like, you know, you can just vibe code everything. It's amazing. And I was like, why is the United app bad?

C (22:28)

The United app is good.

B (22:30)

And then he was like, it actually is good. And then I used it and it's like, not that bad. But the point holds. There is some bad software out there. I will die on this hill. But yes, hopefully it's going to get better anyways.

A (22:43)

There's a deep and important truth. Even if we get something like the software singularity, the level of demand for labor here is practically infinite. Famously, it is the last few percent of completion that take the most work. And by that token, virtually every software product could probably scale up its complexity and features by something like 100x before beginning to saturate demand. 3. This was probably the best point from his Response reindustrialization. There will be some labor displacement of course. Driving stands out. Many types of white collar work, as Citrini suggests, will undergo some gyration as some jobs disappear and others change meaningfully. AI may be the straw that breaks the camel's backs for jobs like the real estate broker where the job had actually already disappeared a long time ago but the pay was still there. The saving grace here is that the US is in the US we have virtually limitless capacity and need for re industrialization. You may have heard about bringing back manufacturing, but it's more than that. We are large. We largely no longer know how to create and don't have the facilities for making the core building blocks of modern life. Batteries, motors, small semis, the whole electric stack is something we we are almost entirely dependent on China and other countries for

B (23:49)

barely make fertilizer. Once you start looking at the physical world, you see a virtually endless scope for work on job creating nation benefiting fundamental infrastructure work that is politically bipartisan. I like that. Where does he close? He says and beyond. The outcome of industrial megaprojects is of course that we move toward abundance. These abundance build. America will once again be more independent and make things at large scale and low cost. Transcending material scarcity is the key in the long run. If we do lose almost all white collar jobs to AI, we have to be able to provide with a continued high quality of life. Part of this will get automatic we get automatically. Just because AI taking margins to zero means that those consumer products will become equivalently cheap. This is a deflationary effect. My view is that different parts of the economy will take off at varying speeds and virtually all the areas are slower than a piece like Citrini's might suggest. To be clear, I'm extremely bullish on AI and expect that one day my labor too will be obsolete. But it's going to take a while to get there. And that time gives us the opportunity to make good policy on that front. Preventing a market meltdown the way Citrini imagines is actually pretty easy. And the federal government's response during COVID showed how proactive and aggressive it is willing to be. I'd expect large scale stimulus to kick in quickly once needed. It slightly irks me to say that it won't be efficient, but that's also not the point. The point is material prosperity for people in the course of their lives. Broad consumer well being that legitimizes the state and carries forth the social contract not satisfying the accounting metrics or economic norms of the past. If we are nimble and responsive to the slow but sure technical revolution, then we will be fine. That's a good response.

A (25:33)

Rise calls out. Out of every example they could have chosen, they went with DoorDash. The barrier to entry for launching a delivery app is not and has never been software. It's just distribution, restaurant adoption, user adoption, and of course, driver adoption. It would be really funny to be using like the Vibe coded version where somebody's like, yeah, I just launched a delivery app and your food will be here in four hours.

B (25:59)

Yeah, it does feel like if I

A (26:01)

was Tony, I'd be flying to find Citrini, having to work face to face opening up a can of I know where you're going.

B (26:13)

Yeah, it feels like, okay, so you built like you Vibe code a profitless, like open source delivery app that anyone can use and you assume that it would get adoption just because it's more economically efficient, like it's cheaper for all parties, so they will join. But DoorDash is actually like a three party transaction, so you need to market to all three and it's really, really hard to break through right now. And maybe it would just go viral and everyone went on. It just feels tough. And then he was saying that like, well, in this future, it's like all three parties are using agents that are perfectly rational and hunting around for the best opportunity so things can shift faster. And I believe that to some extent, but it just feels like still a little bit farther away because of adoption and actual.

A (27:08)

Yeah, DoorDash has modes. That is the big. That, that is the simple trick that all Vibe coders hate.

B (27:17)

Yes. Yeah, I mean, it's like there's a lot of capital that went into building the network and the app, but also a lot of capital that went into marketing and onboarding the pool of. In the marketplace. Like the actual liquidity.

A (27:31)

Yeah, the only, the only scenarios that I can see the sort of like Vibe coded delivery or sharing economy app working is at a local level, but there's already a bunch of competition there. Like, I have a guy that when I want a ride to the airport, I call him, he picks me up. I don't necessarily use Uber because I like having the same guy as my buddy now. I like, I like going to the airport with him. Yeah, right.

B (27:57)

So, yeah, I was thinking about Amazon Basics and like, that hasn't destroyed every company, every brand. Like, and why is that? Like, will that. Is that a good analogy that like, okay, the big AI labs will have Amazon basics for SaaS, but people will probably still want certain brands. There will be certain people that are locked in. Okay, yes, I know the Amazon Basics paper towels are cheaper, but I just happen to like this particular brand that's a little bit more tailored for me.

A (28:28)

But Amazon Basics was like, hey, where you buy paper towels from this brand? Normally we're going to sell you the same product with art, with our logo. Yes, yeah, Effectively the same product.

B (28:40)

Yeah.

A (28:40)

And I think the AI disruption that is much more real is like you have entirely new paradigms for software, an entirely new relationship with software. And it's not just like, oh, you know, somebody built the exact same version of Salesforce. It's like somebody built an app that automatically sets your schedule every day and you're not even, not even thinking about like, oh, I need to be monitoring

B (29:05)

this dashboard or yeah, no, no, I agree. I think the. But I think the Amazon Basics of Salesforce, it probably is not that big of a business opportunity because the whole value prop is that it's lower margin. And so Amazon Basics is not driving Amazon's market cap.

A (29:21)

Well, yeah, Amazon has solved the distribution. They're like, we have the customer. But when you have a lower margin profile and you don't have the customer yet, naturally means you can't spend as much money to acquire customers and build out sales and distribution and all that stuff. That's very different situation. It's not like people are just going to like the SaaS supermarket.

B (29:42)

I think his pushback would be like, they will effectively because they will go to an agent that says, I need to accomplish this job. And it will say, okay, well, for that job, I need a tool. There's a legacy SaaS provider and there's a Neo SaaS provider that's Vibe coded and I could actually build my own version too. And it will pick the most economically efficient across that frontier. And so that's where he's coming away with some downward pressure, which I think is like reasonable. It's just again, I just keep coming back to timelines here. Before we move on, let me tell you about consul. Consul builds AI agents that automate 70% of IT HR and finance support, giving employees instant resolution for access requests and password resets. Let's continue.

A (30:25)

Very funny post from Dime Square Holdings. You think this is crazy, but just wait until next weekend when I publish my substack article. You should freak out and kill yourself right now.

B (30:36)

People are really having fun with this. What was this next one?

A (30:39)

Wow. Dimes is on a roll average 2026 AI macro research. Now I am legend this may be the most terrifying novel you will ever read.

B (30:49)

Isn't that just the zombie apocalypse movie? That might be the most market moving piece ever written, says Clouseau Investments. That seems accurate. I mean, there are other factors going on, but this does really feel like remarkably impactful. And I don't know, maybe it's a buying opportunity, maybe it's a warning to everyone else, but it certainly broke through. I mean these articles is really, really underrated. I noticed like I heard about Something Big is Happening on a car podcast. I got that forwarded to me in an email. These pieces of AI is a thing that you need to be be paying attention to are breaking through in a way that AI 2027 did not. And like the you haven't seen Ilya and Dwarkesh like that. Actually that's a joke because really, that did not break through. But Something Big Is Happening did break through to the tune of 100 million views. Which actually means like everyone read it beyond it broke containment. Like it truly, truly went big. And that's certainly market moving. And the Citrini post did too. We can go through his agentic commerce thing, but it's a little wonky. Let's see what venture anthropologist had to say. So Trini is completely wrong about the impact of AI on the economy, but his article does correctly show that various forms of AI doomerism will become incredibly popular in 2026. Yeah, people are pushing back and someone was saying that this is similar to Karl Marx's critique of capitalism. One of the this is from Mohit. One of the often slept on benefits of attending the University of Chicago is that they make you read Marx as part of the core curriculum. Which is why this article gave me flashbacks of taking Sosc114 as a freshman. Marx, writing during the Industrial Revolution, predicted capitalism would periodically devour itself. Firms replace labor with machinery to boost profits, but competition diffuses, the technology drives prices to marginal cost, and the gains get competed away. This was the collapse of profits. Meanwhile, displaced workers lose purchasing power, hollowing out the demand for the whole system depends on production rises, but no one can afford to buy what's produced. The contradiction between production and realization. Citrini's piece describes this exact dynamic, then declares there's no natural break, but it's the most Marxist. It's the most Marxist piece of financial analysis written in years, and it makes the same errors Marx did. Shumpter offered the obvious rebuttal 80 years ago. Creative destruction doesn't just destroy, it creates industries. We can't yet conceive of everyone in the replies is already making this point and I think they're right. But the sharper rebuttal is Hayex Prices are the break Citrini says doesn't exist. WHO funds $200 billion a quarter in capex when equities are down 40%, private credit marks are in the 50s and consumer demand has collapsed? Cost of capital rises, incremental build out becomes uneconomical, capital gets destroyed and reallocated. Citrini also unknowingly describes Marx's proletarianization of the of the petit bourgeoisie. The 180kpm driving Uber is textbook, but the article claims this collapses consumer demand and that's where it breaks the top decile drives 50 plus percent of their spending and their wealth is in equities, not W2 income. They're long the hyperscalers posting records in Citrini's own model. Blue collar is insulated because AI replaces cognitive labor, not physical. The professional middle class gets crushed, but aggregate demand doesn't. The spending class is the capital owning class. The K shaped recovery they fear actually stabilizes the demand base they say is collapsing in the stable aggregate demand. The petty bourgeoisie finds a way to reinvent invent itself. I think the Citrini piece is excellent and worth reading, but history has repeatedly shown that periods of transformative productivity gains ultimately accrue to the consumer through lower prices, deflation, more leisure and higher quality of life. Marx's error wasn't diagnosing the disruption, it was underestimating the system's ability to adapt. Very good. Let me tell you about graphite Code review for the age of AI. Graphite helps teams like GitHub ship higher quality software faster. Really going heavy on the goat emojis today. I like it.

A (35:10)

It's Monday.

B (35:11)

Everyone is reading the latest Atrina piece thinking it's an institutional research piece, when in fact what they are reading is a marketing piece of fiction meant to go viral. And viral it did. Go lesson lesson there. I wonder how broadly the article virality like long form has not been going viral on X or Twitter for a decade. Like even before the link ban and stuff. Like it was really really hard for articles and links to really go big. They did go big when Twitter started because there wasn't that much content. So people would write and then they'd bring that and then they discuss that. But I can't remember a really big debate erupting around an article.

A (35:57)

Maybe a little bit, but Steve asked does Tyler clap for each ad Reader. Is that a soundbite? It is real. He claps.

C (36:03)

Yeah, I clap for every single, single one.

B (36:05)

We'll give you a chance to do it again. FIN AI, the number one AI agent for customer service. If you want AI to handle your customer support, go to FIN AI. We did get some feedback that the clapping can be a little loud during the ad reads. So, you know, Tyler, constructive criticism there. Anyway, should we continue with the Citrini back and forth? We do have someone from Citrini coming on.

A (36:28)

Yeah, we can. We can keep going.

B (36:30)

The thing about the stream piece is that that is internally inconsistent is where does all the surplus go? Okay. We become impoverished and aggregate demand collapses. What financial asset is spared from that? A lot of people are asking that. And yeah, I mean, the answer is, is commodities. You want to be in like. Like gold and silicon and whatever is at the bottom of the stack and then whatever has the deepest moat. And then you want to be at the. In the AI companies as well.

A (36:55)

Sean says he does as well. Thank you, Sean.

D (36:58)

Thank you.

A (36:58)

Let's switch gears to something that is certainly more important than the Citrini piece. A street legal modded garbage truck.

B (37:06)

What?

A (37:07)

Pratt and Whitney J3 jet engine.

B (37:09)

Okay, let's see it. Is this real for AI?

A (37:12)

Let's pull this up. You don't even know anymore.

B (37:15)

Wait, I've seen this Pratt and Whitney before. Hermeus, I think, bought one. I have no idea if that's real, but that's remarkable if true. Yeah, no, you can. Those are expensive. But I do think that they will sell those to you.

A (37:30)

Feels like a six foot flame at the end of your vehicle. Is not street legal.

B (37:38)

I would agree. I would agree.

A (37:40)

Anyways, horrors coming out of Mexico. Yeah, Yesterday. Really sad situation. Our very own Joe Wiesenthal had been in the Puerto Vallarta area, had just left. I think he got out of there an hour, an hour before the chaos erupted. So very grateful.

B (37:58)

I hope everyone who's down there is safe.

A (38:00)

R Marriott on Reddit. Somebody said Weston Puerto Vallarta won't honor late checkout with streets closed. I am platinum elite. Over 1,000 lifetime Marriott nights.

B (38:14)

Wait, I thought that was a joke. I didn't realize somebody actually posted.

A (38:17)

TV is on fire due to the cartels setting fires and buses. Cars and buses on fire all over the city. The airport is closed and Ubers and taxis are not running. I asked for a 4pm checkout, which I'm entitled to based on availability. They won't extend past 2pm and said we would have to use the hospitality suite. We are supposed to be leaving for Bucheiros this afternoon, but that isn't looking very good. Worst Bonvoy property I have ever experienced. I don't think anyone will be checking in today, so there's no reason to not at least to at least not extend us to 4pm this is so.

B (38:52)

Does this person just not understand the scale of what's happening? Maybe you could break it down for anyone who's living under a data center. Like, what actually happened in Mexico, because it was not just buyers in a few cars. This was like a military operation. Correct.

A (39:07)

Because my wife texted me yesterday afternoon while I'm on X monitoring.

B (39:14)

You're monitoring the situation.

A (39:15)

All the open source intel.

B (39:16)

You texted me. You texted me. And I was like, oh, what's up, Joe?

A (39:20)

Wow. I've expected something like this for a long time given the tensions down there. And then I'm just watching this and my wife texts me, our friends want to go to Mexico in April. Can we go a long weekend? And I was like, are you joking? Anyways, a really, really sad situation. Basically, the leader of the cartel cjng, which is like effectively his paramilitary group. Any. Like, if you looked at any photo or video of them over the last 10, 20 years, they look like they're special forces.

B (39:58)

Yeah.

A (39:59)

I think the story is that many of them actually did train at some point with US Special Forces and then flipped.

B (40:05)

Or. Or the. Probably the Mexican military.

A (40:07)

Yeah, no, they were. But the. But the US Special forces have trained Mexican military.

B (40:12)

Okay.

A (40:12)

And so these guys are like. Like, they have their own version.

B (40:15)

Yeah, it's not a larp. Like, oh, they just like, picked up something. They watch like a video on YouTube.

A (40:20)

Yeah, I mean, I'm sure. I'm sure some of them are not a teacher, but in general, this is like a paramilitary organization. It's like one of the largest, like, private armies in the world. Probably the largest private army in the world. And so El mentioned their main guy gets taken out and then they respond by starting to just like blow up, like roads. They. They took over an airport. I guess they just start causing mass chaos.

B (40:47)

Yeah.

A (40:47)

Because their leader, this person in Puerto Vallarta. If you look at any video of Puerto Vallarta, if you just went outside yesterday and looked around, there's like fires rising up everywhere. It looks like. It literally looks like a war zone. So for somebody to be hitting Reddit at this moment and being frustrated, it's like, hey, maybe just, you know, the State Department put out like, almost exactly when this person was posting a security alert saying due to ongoing security operations and road blockages and criminal activity, US citizens in the following location should shelter in place until further notice. And like you're getting a shelter in place warning and you're. You're mad about your Marriott points. But again, hopefully, hopefully things settle down.

B (41:33)

I agree. I agree.

A (41:35)

South.

B (41:35)

Well, I want to move on to some nostalgia. We are going to get Tyler Cosgrove up to speed on what it was like to live in the 90s and the early 2000s. First, I'm going to tell you about MongoDB. What's the only thing faster than it? The AI market your business on MongoDB. Don't just build AI own the data platform that powers it. I grew up in an era before mongodb. I think of my life as pre mongodb. It's crazy. Back then you had to store files and text files. Now MySQL existed, but the 90s and early 2000s were iconic. And we got a get Tyler Cosgrove up to speed on what it was like during the heyday. These were the vibes, the blockbuster. So much consumer electronics. Like everything had a. You had a different device for everything. A Walkman, a Game Boy, an Xbox. You can still go see a monster truck rally.

C (42:39)

I had an Xbox.

B (42:39)

You had an Xbox?

C (42:40)

Yeah, we have mod retros right here.

B (42:41)

Okay. Okay. So you're maybe up to speed. Well, there's another one that talks about the liquid metal object design. And I found this reel very informative to show how technology in the digital world actually shaped the physical world. So we can play this liquid metal object design. I feel like this is overdue for a comeback. Pretty close.

A (43:04)

Floating around robots, Blobism, Blob minimalism and biomorphic design.

B (43:09)

Biomorphic design.

A (43:10)

Easy to see that industrial designers were

B (43:13)

trying to push past this because everything was really blocky in the 80s.

A (43:16)

Futuristic vision for things like sportswear, watches, music players.

B (43:20)

Clear. Clear case.

A (43:21)

This is a prime example of when technology influences form.

D (43:25)

CAD A modeling software introduced what is called nerves. Non uniform rational B splines.

A (43:32)

What this allowed is for industrial designers to create mathematically smooth curves to be calculated with precision. It enabled people to make.

B (43:40)

So before you had to just like use blocks basically and like you'd get like a sphere and that was it. You could do like a sphere and a cube, but you couldn't really do whatever shape.

A (43:49)

And back then they were rocking this kind of hardware when they were saying that all retail stores, yes, globally, will be wiped out within the next five to 10 years.

B (44:00)

Yeah, you had to be there. Yeah. I mean, reflecting on the dot com boom, I think is particularly interesting right now. I mean, I do like the takeaway from the dot com boom. When most people pull up the dot com boom, they're just like, oh, it's a bubble and everything's going to zero. And that's not quite the lesson because the Internet was still actually the most powerful force for economic growth and change. And it did radically change society. It just did so over two decades instead of like one year as was predicted. So there's an article in the New York Times that's sort of comparing the dot com boom to the AI boom. And it's a piece by David Streitfeld at the New York Times. He says people loved the dot com boom. The AI boom, not so much. And I buy that. Generally like the tenor around the dot com era. Yes, there was a lot of froth, but in general, people were like, oh, this is like sort of interesting and cool. I gotta play with this, like tinker. They just missed it more as a toy than like true doom. Yes, there was Y2K and people were worried about that, but the stats weren't quite the same. So right now there was just.

A (45:12)

Were you like, aware of Y2K?

B (45:14)

Extremely aware. Extremely aware.

A (45:17)

How did you process it? Because my parents were trying to explain to like a five year old.

B (45:25)

Yeah, yeah.

A (45:26)

No, I was like, I could generally.

B (45:29)

Yeah, yeah. I remember the turn of the millennium. I was overseas on a vacation and there was a lot of fear. Yeah, well, you got to get out. Everything's collapsing. You got to seek refuge. No, no, there was like a fear that like, okay, like something crazy might happen. But in general, my parents were like dialed in enough that they had been through the boom and bust of like, something bad will happen to actually understand that, like. No, the fixes were in place and there were a bunch of interesting fixes. So if you're not familiar with Y2K, basically the idea was computers were programmed to store dates as two digit numbers. So you would just say it's 86, then it's 95, 96, 97, 99. What happens when you get to 2000? It just says 0, 0. And all of a sudden all your interest calculations for your bank account freak out. You have negative money, the whole financial system collapses. Anything that's planned, Right. All of this was like the fear of what might happen, Tyler.

C (46:31)

Yeah, I mean, that doesn't make any sense.

B (46:33)

Right? Why?

A (46:33)

Well, like, easy for you to say, Tyler. You weren't born for.

B (46:39)

Oh, my gosh.

C (46:39)

The numbers are resetting all of a sudden. Like, you can see, it's like a calendar.

B (46:43)

You can.

A (46:43)

You had to be there, Tyler.

B (46:44)

Yes.

A (46:44)

You had to be there.

C (46:45)

Didn't no one have foresight? No, nothing happened.

B (46:48)

No, no, yeah, people did have foresight. And so they started implementing changes. And the changes cost a ton of money. I think the total bill for Y2K systems updates, because if you had hard coded, like our dates in our systems are, we're a bank and we store dates in two digits. You gotta go and change that. And that's a couple days of war rewriting some code. They didn't have cloud code back then. And so, yes, it wound up being something like hundreds of billions of dollars were spent in the lead up to Y2K to prevent the gong for the Y2K paydays. A lot of people made a lot of money. But there were also a whole bunch of interesting rules. And hotshot over here, I'm going to give him a popular quiz. Do you know how to calculate a.

A (47:38)

If you leap here, if you get. If you get this wrong, there's going to be consequences.

B (47:42)

Okay, okay. No, no, no, no, no. Stop, stop, stop. Do you know how to.

A (47:44)

Consequences?

B (47:46)

Oh, no, no, no.

A (47:50)

I told you there'd be consequences, Tyler.

B (47:54)

All right. Okay.

A (47:55)

That was the most Zoomer thing I've ever seen in my life.

B (47:58)

He just has to go. He's. He's brain dead.

C (48:00)

Wait, how to calculate leap year?

B (48:01)

Yes. Like, like, like, like, when do leap years happen?

C (48:06)

I don't know. Like the actual calculation. Isn't it every four years?

B (48:09)

Every four years. Except every hundred years. Except every thousand years. So they toggle back and forth. And so in. In 1900 was not a leap year. And so if you. If you didn't know any of the rules, you would just think, oh, every four. Any of the special rules, you would just be like, every four years is a leap year. It's a leap year. 2000. But if you knew the hundred year rule, you would be like, oh, actually, it's not a leap year, so I need to hard code the system that is not a leap year. And you can fact check this too, because I'm not sure, I'm just riffing here. It might be wildly wrong. But the thousand cancels out the hundred, and you wind up with just a normal year. And so if you did nothing, you win.

C (48:56)

Yeah, but. Okay, so Gregorian calendar goes into place, 1582.

B (49:00)

Yeah.

C (49:00)

So you have like 400 years to figure this out.

B (49:02)

Yes, yes, and we did. But it cost us $100 billion. As is all technological change. But Y2K was like, it was very millenarian. People were dooming about the apocalypse. But these were like fringe sort of cult types. The same thing happened with 2012. I don't know if you remember 2012, apocalypse stuff. Y2K was the same thing, but it was not widespread. AI doom is truly widespread. There's a study in YouGov. More than 30% of Americans are concerned that AI could end human life on Earth. Like that is a wildly high number compared to how many people believed 2012 was gonna be the end or 2000 was gonna be the end. Like I would be shocked if either of those dates were single digit percentages. Most people were like, yeah, okay, like I might need to like print out my bank statements. A lot of people are doing that. Like print out your bank statements before Y2K because like then you'll have a backup and you'll be able to go in and say like, no, I actually have $10,000 in my bank and I don't have negative 9 million because it's not the year 1000 right now. But you know, everyone got through that. The other interesting thing is that there's this disconnect between the doom AI is going to kill everyone and then what is the impact of AI? There's this new research paper from the National Bureau of Economic Research, nber. They polled a whole bunch of firms in America, a whole bunch of companies, and they said, are you getting a benefit from AI? Tyler, what percent do you think said, yeah, AI is helping out?

C (50:37)

It's probably pretty low.

B (50:38)

It's extremely low.

C (50:39)

20%.

B (50:40)

It's exactly 20%.

A (50:41)

Nailed it.

B (50:42)

80% said that AI was having no impact on their productivity or employment. We gotta get those numbers up, folks. There are clearly good ways to use AI to benefit your company and your people.

A (50:55)

I mean it could also be those surveys. You never know who, you never know might be somebody who's just not. Who's gonna answer the survey quickly. And they don't realize it.

B (51:03)

But I mean, just think about it. Like there are lots of companies where you have to be HIPAA compliant. Maybe they just don't have a HIPAA compliant LLM available. And so they're just like, yeah, I literally can't use it. Or like, or like all LLMs are blocked on my local work network because the IT department's still figuring out how we roll it out. Like these things happen all over and it affects like lots and lots of people. I mean, if you're just a cashier at Walmart, is AI helping you? Like, it's just not. Right.

C (51:28)

Yeah, but I'm sure there's still like they use services that under, you know, like behind a bunch of layers, there actually is AI going on. Right.

B (51:35)

I agree. I agree.

C (51:36)

So like, maybe they're not interfacing with LLMs directly, but at some level, like there are LLMs running totally total.

B (51:43)

Even if you zoom out to AI just being like machine learning, it's like, okay, so the Walmart person at the checkout counter is not seeing a benefit, but Walmart definitely has a recommendation system on their website.

A (51:52)

Yeah.

C (51:53)

Or like the Walmart software is being updated faster than it usually would be. Isn't that.

A (51:58)

That's AI?

B (51:59)

You believe that that's happening?

A (52:01)

It could be.

B (52:03)

You're willing to bet it all on that? You're willing to bet it all on Walmart going through a dramatic digital transformation right now. You don't think it's slow? You don't think they're still like, let's figure this thing out?

C (52:13)

I think they're probably used to be.

B (52:16)

You think they're past the pitch deck phase? Maybe. Hopefully. Yeah.

A (52:19)

And we got to talk about. We should, we should talk about Rufus. Do you guys hear about Rufus?

B (52:23)

No. What happened?

A (52:24)

Rufus is going crazy.

B (52:25)

Okay. While you pull that up, let me tell you about public.com investing for those that take it seriously. Stocks, options, bonds, crypto treasuries and more with great customer service.

A (52:35)

I'm trying to pull up the.

B (52:37)

I have some more. My, my takeaway was that the average American believes that they are in Terminator judgment day, but they still have to go to Cyberdyne Systems and do their fake email job and just right up until the bombs drop. That, that's the general tenor around AI. Like the, the vibes are rough. But if we go back to the dot com bubble and try and understand what's different, there's some interesting stuff that we can learn. So first, there was definitely a vibe around permanent high growth and a new economy. There was this economists, analysts, executives, they were arguing that productivity would permanently accelerate and recessions would largely disappear and the business cycle would be broken by information networks that moved at the speed of light. Before the SaaS apocalypse, there was what you referred to earlier, the retail apocalypse. The most extreme formulation. Formulation was total physical retail extinction within 10 years. So within 10 years, they predicted by 2009 there would not be a single retail store anywhere in America. This was the prediction. This is the Prediction. This is the prediction directionally. Directionally accurate for sure. For sure. So shopping malls would become obsolete. All brands would be commoditized by cheap online alternatives. Some of this happened. Amazon Basics is popular. Temu flooded America. Shopping malls are struggling. But Walmart's a trillion dollar company, Nike's worth 90 billion, and Rick Caruso has seemed to sort of figure out a way to make malls work in LA at least. There were also a ton of other crazy.com proclamations. Revenue doesn't matter. Only eyeballs matter. All media will permanently be free because file sharing and products like Napster simply cannot be stopped. And so every piece of media will be free forever. That obviously didn't happen. And offices will disappear entirely. Digital currencies will replace fiat money at its core. The most extreme claim was the Internet was a civilizational phase change equivalent to the printing press or electricity. And most importantly this transformation would happen in five years, not 50 years. And so time compression was the biggest forecasting error here. Not every dot com prediction, like nearly every dot com prediction had some directionally correct element to it. But various breaks were applied either voluntarily or involuntarily and things slowed down. Media companies sued file sharing companies for example. Financial markets pulled back, companies adjusted their strategies and retreated to Internet proof modes. And so protests and political movements also had another role to play as a break. There was this really interesting anti tech protest in the late 90s, the Battle of Seattle. So over four days 40,000 protesters rallied against the World Trade Organization to put push back against Internet driven capitalism. There were 600 protesters who were arrested at the Battle of Seattle. And they were arguing that the Internet was linked to corporate consolidation, outsourcing and labor distribution placement. Like all relatively true things hard to disprove, but the timelines are what matter of course. And so the Battle of Seattle didn't result in any specific dramatic curtailing of Internet adoption, but it did raise the political salience of international trade relations and was clearly in the back of policymakers minds when they set sectors targeted tariffs and domestic preference procurement rules over the next decade. And so I was thinking about this in the context of the New Brunswick Data center protest. So the actual this, this data center that got canceled in New Jersey, by comparison to AI like it's tiny, it's

A (56:19)

generous to call it a data center.

B (56:21)

Yeah, it's more of like a data point.

A (56:25)

Edge computing.

B (56:26)

Yeah. So it's 25,000 square feet. The current like meta large data center campus is 500,000 square feet. So 5% of the size. And so this data center, we don't know who is actually going to buy the capacity, where it was going to go, but you can think of it much more like delivering you Netflix faster than training the next AI model. But it worked. They got the data center canceled. And so this is going to be a data point in the minds of AI policymakers, decision makers, leaders for a long time. And I think that that will, that will affect things. So, you know, the Internet rollout continued even during the bubble, and the bubble popping and pushback and all sorts of different things. AI will continue as well. But I think it's important to like, refocus the conversation on actual impact. Like the 20% needs to go up and people need to say, yes, this is helpful, and then mitigate the negative externalities before they turn into problems, like real problems for average Americans. The energy issue was foreseeable, it was predictable. And maybe that's what we need to be forecasting. The next turn of AI 2027 or AI 2028 should be like, here are all the problems that we're going to bump into along the way. Let's go mitigate those now. Because all the hyperscalers could have been subsidizing electrical build outs like years ago for sure. So anyway, let me tell you about Gemini 3.1 Pro. Gemini 3.1 Pro is here with a full. With a more capable baseline. It's great for super complex tasks like visualizing difficult concepts, synthesizing data into a single view, or bringing creative projects to life.

A (58:15)

We need a moment of silence for international business machines.

B (58:21)

What happened?

A (58:22)

Falls over 10%. Actually 11% now after anthropic announces that Claude can streamline COBOL code.

B (58:30)

Oh, no. There we go.

A (58:32)

Wild, wild times. The example that I was talking about earlier about Amazon's Rufus, I was going

B (58:38)

to say Anthropic announces they're going to launch an international business machine. We are an international business.

A (58:46)

Mike Isaac was reporting. No Financial Times, he was just commenting. Amazon's internal AI coding assistant decided the engineer's existing code was inadequate, so the bot deleted it to start from scratch. That resulted in taking down a part of AWS for 13 hours. And it was not the first time it happened.

B (59:05)

I love it.

A (59:06)

Sometimes the best course of action is to delete and recreate.

B (59:09)

Delete everything. Sometimes that's what you gotta do. Lots of people are having fun with the data center protest. I think it should be taken seriously. But apparently the New York Times ran an article in 1887 that says, peasants destroy a balloon. Is this a real this is a real article.

A (59:28)

Peasants, 1887.

B (59:31)

1887.

A (59:33)

You can actually find it on the New York Times website. October 25, 1887. The Russian peasantry appear to be sunk in ignorance and superstition. During the recent eclipse of the sun, three famous Russian savants descended. I'm trying to read. This is in the Times Machine. It's very, very.

B (59:52)

Oh, it's actually like. It's a scan. Right. It's not text.

A (59:55)

Yeah, Incredibly hard to read. But the peasants did destroy the balloon.

B (60:02)

They destroyed it.

A (60:02)

They got it.

B (60:03)

Somebody asked, would you live next to a date center? Not a data center, but a center for dates.

A (60:09)

Dates are underrated.

B (60:10)

Yeah, dates are good. Healthy, delicious. Anyway, 11 labs build intelligent, real time conversational agents reimagine human technology interaction with 11 labs. Let's go over to the horse section of the show. There's some big horse news going on. The moment you've been waiting for in the Financial Times. A horse walks into a lab.

A (60:33)

See, peasants destroying a balloon in 1887. Is setting a waymo on fire in 2025.

B (60:38)

Yes, but. Okay, so the interesting thing about the Waymo fires was that we live in LA where the Waymo fires happened. And if you were on the Internet, it looked like Los Angeles was burning to the ground. And Tyler went to the Philharmonic. The Philharmonic, which was directly like a block away from where the main protest was. And we were like, whoa, man. Like, this seems pretty dangerous from what we're seeing online. And it was fine. Right? You just pulled right in.

C (61:08)

Yeah, I saw someone holding a flag and that was it. I didn't see. I didn't even see the fires at all.

B (61:13)

And so a lot of these. The scale of these protests is hard to pick up on because things can go really viral. And you can have a protest that's. That's a couple hundred people. And if it's in one block and the photographer is good about lining it up and you're not seeing like a helicopter shot of like tons of people in the street, it can actually be sort of small. I remember that video of the data center protest where he runs outside and he's like, we did it, we did it. Like, it seems huge, but I actually think there were only like a couple hundred people there. And you comp that to the World Trade Center Organization Battle of Seattle. It was 40,000 people. They arrested 600 people. Like, that's pretty significant. And so I guess I'm not saying, like, we're still early for protests, but it is important to understand the scale of what's happening in the real world and the actual impact of that. And you need to be charting this because if they're getting bigger, they need to be addressed more. And even if they're small, people have good points, so they should be listened to. And, and you know, and the solution should be brought to the populace before it gets to a vote. Like, if you could, you could tell that that New Brunswick debate would go way differently if the hyperscalers were there saying, like, hey, good news, like, we've done so much forward thinking here that your energy prices are going to go down. Like, people would be like, oh, okay, cool. And we're making it beautiful. It's going to be building a park, and we're building a park underground.

A (62:40)

We're putting grass on the roof.

B (62:43)

Exactly. Yeah. There's like five easy tricks to get data centers approved all over the country, but everyone's been putting them on the low priority pile. But they're certainly going to be more important over the next couple of years.

A (62:59)

Why don't you read us?

B (63:01)

Yes. A horse walks into a lab. It's a December afternoon at the Campo Argento de Paris Polo at Palermo in the northern suburbs of Buenos Aires. The sun is shining in a sky of clear Argentine blue. The jacarandas. This is too noisy. I'm gonna put this down. The jacaranda trees are in bloom. You're sitting in the stands overlooking an immaculate green lawn six times the size of a football pitch. A military band with brass trumpets and drums, red epaulettes and shiny black jack boots has just marched away. Argentina's president, Javier Milei, famous for his Elvis sideburns and economic chainsaw, has taken his seat above the center line. Eight players center on to the pitch. Tanned arms, taut muscles, hair curling over the collars of their polo shirts. It's the first semi semifinal of the Argentine Open, the most prestigious tournament in the polo world. The one the players really want to win. This year, the stakes are higher than usual. It may be the last open for Adolfo Cambasio, the world's number one player for more than two decades and the sport's goat, the greatest of all time. Cambasio has changed the way polo works, not only through his skill and tactical genius, but as a result of a bet he made nearly 20 years ago. He bought into the idea of cloning ponies a decade before his rivals. This year, many of the ponies he will ride in the Open will be clones, identical twins of his favorite horses from years gone by. The players line up four against Four. The ponies waiting, ears pricked, poised for action. The whistle blows. The game begins. The players streak up the pitch, stick swinging, using their ponies to ride off their opponents, to block them from getting to the ball. They gallop, turn, stop, turn on a sixpence and start speeding in the opposite direction. They bounce on the ball of a small head of the stick. They bounce the ball on the small head of a stick, hit backhands under the pony's necks. If there's a break, they gallop to one end and leap onto a fresh pony before charging back into the fray. You can tell, sitting high in the stands, the ponies get the game. They can anticipate what's going to happen. They're enjoying themselves. That, in and of itself, is remarkable. In the millennia since humans thought to tame the wild horses that roam the steeps and plains, we have bent them to our will for our species. Horses have charged into battle, dragged plows through rocky fields and carriages, through the slop of medieval cities. The real slop problem, medieval cities.

A (65:33)

Now, lee versus slop.

B (65:35)

Man has taught them that a hard white ball needs to get to the end of a large field and through the gap between two poles. In return, we feed them, tend to their shoes and teeth. We give them massages and march them up and down hills to make sure they are fit enough to play. We polish their coats, plate their tails and bandage their legs. And to the best, give them a chance at life through cloning. And another, and another. But it seems that isn't enough. Now two men, a scientist and entrepreneur, are going beyond making copies of an original. They are engineering polo ponies to make them even faster, in the hope that in a game of high stakes and slim margins, it will give them the edge to win. For one pony Polo pireza, the circle of life began on an estancia. Estancia near the town of Coronel Suarez, a six and a half hour drive southwest of Buenos Aires. She was born on December 12, 1988, a slight mare raised on a diet of weeping love grass, the silvery fronds that grow in the red soil of the pampas. It was apparent from an early age that she had what it took to make a great polo pony. She played in her first Open final, aged only five, ridden by Pepe Higai, one of four brothers who competed at the highest level. Four brothers, all polo legend.

A (66:50)

That's elite.

B (66:50)

Polo remains a macho sport. The top players are male, though there is an open tournament for women, too. The grooms who canter the spare ponies up the side of the pitch wearing the gauchos traditional floppy beret. The Bonia are typically male. The Open has only had multiple has only ever had male umpires, but the ponies are usually female. The players favor mares, citing their intelligence and grit. Polo Puerza was one of the great polo mares of her generation. She was a bright bay the color of autumn conquerors, with black legs, a white star between her eyes and another splash of white on her nose. She played at the top level for 14 years, winning the cup for best pony at the Open, among many other awards. And speaking of ponies, the Purple Llamas Vanta Automate Compliance and Security. Vanta is the leading AI trust management platform. That mayor was a natural polo player. One in a million, he said. Guy said in a video made to celebrate Polo Puerto induction into the Polo Pony hall of Fame. She had an impressive heart. Also, she was a mayor who never got tired. Polo Puerza retired in 2004. Before she died, samples of her DNA were banked in liquid nitrogen in the laboratories of Kyrion, a biotechnology company co founded by Gabriel Viceroy and based in a science park in the town of Pilar, 35 miles northwest of Palermo's polo fields. With his neatly trimmed beard and white lab coat, Veschera, 46, would not be mistaken for a polo player. When I first emailed him, he apologized for not responding sooner. He was competing in the World Indoor Archery Championships. Oh, made sense. His company is named after Chiron, the mythical archer. Half man, half horse. It's the morning of the Open semifinal and we're sitting around a large table in the cool of the conference room at Kirin, just a few meters away from the vats of nitrogen housing Polo Puertas Puerto remaining cells. There are photographs of horses on the wall. Each clone. Each is a clone Vichera was responsible for creating, each a genetic replica of a famous polo pony. One photograph. The newest is of five bright bay foals covered in baby fluff, their legs still too long for their bodies. These are the gene edited polo puerzas, he says. The ones you are going to meet. I stare at their photos. There is no clue in the faces of their unusual conception, no way to differentiate them apart from their similarities to each other and to the videos of Polo Puerza that I have watched. What do you think? Would you go to a polo match if all the ponies were genetically modified and cloned? Or do you want a natty league? Natty ponies only.

A (69:36)

I'm in favor you're in favor of cloning? Yeah. I mean, why not push the sport to the limit?

D (69:41)

Hmm.

B (69:43)

I feel like it's kind of like not in the spirit of just like a couple guys getting on horses that they just like found. Round them up, you know, it's.

A (69:50)

Well, they should. They should have the sort of like Wild league.

B (69:54)

Yeah.

A (69:54)

Where you can just go find.

B (69:56)

I think this.

A (69:56)

Find a wild horse. Go. Go play.

B (69:58)

This definitely creates an opportunity for the Wild league.

A (70:01)

Yeah, sure.

B (70:02)

This is kind of the free range league. What do they call it?

E (70:05)

The.

B (70:07)

The enhanced games. Yeah, this is the enhanced games for polo for sure.

F (70:11)

Anyway.

A (70:12)

Tyler, are you a polo accelerationalist?

B (70:14)

Sorry, Your business is AI? Their business is securing it. CrowdStrike secures AI and stops breaches. Tyler, are you a polo purist?

C (70:23)

I think I'm in favor of the cloning.

B (70:25)

You're a polo accelerationist?

A (70:26)

Yeah.

B (70:27)

P ack, as they say.

A (70:29)

Yeah.

B (70:29)

Well, good news for you. There's lots of cloned ponies coming your way to a polo field near you. What did this post say? If you run every day, you'll be ready for any situation that calls for extreme cowardice. Was this deleted?

A (70:48)

I think so. I think it's probably the running community. Zach, Pog Rob came in and said you're about to be running for me.

B (70:56)

Yeah. Let me tell you about Applov and profitable advertising made Easy with Axon AI. Get access to over 1 billion daily active users and grow your business.

A (71:05)

Today there is a game called Data center on Steam which lets you build and manage your own data center. This is low key genius. The best way to educate people on a new trait. Hyperscalers should lean, should learn a thing or two about edutainment.

B (71:20)

Edutainment. This is fantastic.

A (71:21)

Tyler, Somebody was saying it's not out yet.

C (71:23)

It's coming out March 31st.

B (71:25)

Oh, okay. Mark your calendars.

C (71:27)

Yeah, I'm going to have to grind this.

B (71:29)

Productivity is going to fall.

A (71:31)

Somebody was saying it's like it could easily be an Ender's game scenario where it's just.

B (71:37)

Those racks weren't simulated. Those were real NVL 72s ender. Yeah. I love that this is all a ploy. The humanoids are already deployed. They just need you to wire everything up. This is amazing. And it feels like the game mechanics feel just from this video, remarkably deep. Like you're not just walking around a data center doing cabling the entire time. You're also deciding tax treatment and what software runs and getting probably kubernetes installed or something like that Slurm. Very, very fun. I love these one off games.

A (72:16)

Apparently there's another game called just called Insider Trading coming to Steam.

B (72:24)

If you're good at insider trading, you're going to love this game. Steam has a game called Insider Trading Get Ready. It's a roguelike deck builder that lets you literally pump and then crash the market. This is going to be wildly, wildly popular. No. Depends a lot on the actual mechanics of the game, but hilarious and says a lot about the society. But I think it's. I don't know. I'll give it a try. I wonder if it will have microtransactions. That's the big question. Or if it's here for the love of the sport, Love of the game. But these roguelike deck builders are fantastic. Balatro went mega viral a couple years ago. Really, really fun game. Just crazy poker. Basically it's like Poker Rules but with a whole bunch of crazy modifications that allow you to just do insane things and sort of turns it into a completely different game.

A (73:18)

Ryan says someone make a TVPN intern simulator.

B (73:21)

That'd be good.

C (73:22)

I'm still waiting.

A (73:22)

We would actually. Yeah. Now that Tyler's like promoted and just, you know, a real, a real deal employee, which happened after like two weeks last year. But that kind of would be kind of fun for all of us to kind of like relive the days of, of intern summer.

B (73:41)

Yeah. Whoever has the highest score gets hired.

A (73:43)

Yeah.

B (73:45)

I do wonder. We've seen this. I heard this story about like everyone talks about the death of AAA games right now. Have you heard about this? So it used to be, you know, GTA 5, Halo 3, BioShock. Like there were these big games that would sell for 50, 60 bucks. They would sell a lot of copies. But then they weren't. What do they call them? Permanent service games, online service games or something? Perpetual, like Fortnite is a game that has endless updates and monetizes forever. And same thing with Counter Strike, League of Legends. There's a few others that have wound up generating a ton of money for these companies because once they get them up, they're like Ecosystems, Roblox, great example. Versus if you're doing like BioShock and you make a bunch of money, then you have to do BioShock 2 if you want more money from those customers. And then you have to do BioShock 3. And at each point people are like, well, I didn't actually finish BioShock 1, so I'm sort of out of the market for BioShock 2. And your Tam just gets smaller and smaller while your development costs get higher and higher. And there's been a whole spate of AAA perpetual service that. Why am I blanking on the term? There's online service games where they've come out and they said like, okay, we've seen what Counter Strike has done. We've seen what League of Legends has done. We want that for our company. Free to play. Yeah, Free to play is the model. But there's something around the word service that's the gaming lingo. But there's been a lot of flops recently. Like a lot of companies have spent a ton of money on these online service games that they hope will become the next League of Legends or the next Counter Strike two. And then they just flop and they're shut down in like a couple months and it's a huge loss. At the same time, there's been a whole bunch of developers that have gone sort of the indie route and done really, really well. Live service games. Thank you, Bobby. Cosmic. They're called live service games. Nailed it. And there was this interesting story of this developer that spent like three years working on this live service game and it like completely flopped. And then like in the free, in his free time made like this game called Peak for in like three months and it went like super viral and did really well. And I'm excited to see like, like do. When do we see the actual acceleration in Vibe coding? Do we get more of these like meme type games that have like really interesting mechanics? Does it actually free up the developers to come up with not just interesting viral hooks? Like Data Center Simulator is funny enough to get us to click on it. But the mechanic actually has to be good too.

A (76:29)

A million people will think it's funny. 10,000 will try it.

B (76:33)

Yes.

A (76:33)

How many people actually play for more than 10 minutes?

B (76:38)

And the key to playing for more than 10 minutes is not the graphics will be taken care of. We already have Unreal Engine. The engine will work like you're not going to be falling through the floor. It's not going to be buggy and you'll be able to generate assets and actually make the thing look like it. But if you can come up with some sort of novel reward mechanism progression system that's interesting, that shows, okay, I'm learning and I'm having fun and I'm reengaged. I think that will. Dan says Peak was that developer's peak. Nominal determinism hits again. Yeah, you never want to launch a product called Peak anyway really quickly. Let Me tell you about Lambda Lambda is the super intelligence cloud building AI supercomputers for training and inference that scale from one GPU to hundreds of thousands.

A (77:22)

Signal says serious question. How do you make someone with absolutely zero gaming experience CEO of a very prominent and important gaming platform? Asha was announced as the new EVP and CEO of Microsoft Gaming after a multi year run over on the enterprise AI side of the business. A lot of people had opinions on this.

B (77:47)

Well, I have an opinion. Signal's question is, how do you make someone with absolutely zero gaming experience CEO of a very prominent and important gaming platform? You make them lock in and spend three months gaming. And so that should be the first task. It should say, okay, you have no meetings. Microsoft Teams is. Clippy's got shut down for you.

A (78:09)

Clippy's got it.

B (78:10)

Yeah, we're handling everything. Your job is to speedrun every major Xbox game Fable, all the Call of Duty series. You're going to play all the Halo games. You're going to get good, you're going to rank and you're going to learn to speed run and you're going to really, really lock in and establish true credibility that can't be faked. And then we will announce you. That's the hack. People are pouring one out for Phil Spencer who was at Microsoft for 38 years and his profile picture is just the Xbox X because he's a legend. And Palmer Luckey quoted it and put an F in the chat because the world you grow up in no longer exists apparently. What else is going on here? Okay, so someone asked Asha, what's your favorite game? And Smash, JT says okay, I'll play your game, you rogue. Chrono Trigger forever goaded. Final Fantasy 7 Goldeneye 007 Chrono Trigger will forever be number one. I never played Chrono Trigger. I did play Final Fantasy VII and GoldenEye and Asha said, Great list. I did my top three. And another reply, Halo Valheim, which I believe is newer and has had much less sticking power in 007. It's been a long time since I played Chrono Trigger. Have you done every ending? Thanks for all the detail. I appreciate it a ton. And is the question that Chrono Trigger doesn't have multiple endings? I actually don't know. Okay, so I don't know what's going on here, but Xbox CEO accused of using AI for replies saying she played Chrono Trigger in her reply to Smash. She would have been six years old. You could play it later, which would be very young age to play it. But Maybe she loves JRPGs and picked it up later. It is a curious thought. She must be a huge gamer. Or this is AI. This doesn't read like AI. I don't know. What does Palmer Lucky say? He says Crown Trigger is my favorite game of all time and I was only three when it came out. True. Yeah, good point. Also, I don't know Chrono Trigger. I played Final Fantasy vii. I don't actually. I think there are multiple endings. Like I would not remember. I don't know. Anyway, what else is in the timeline? We should tell everyone about the Linear lineup for today because we have four guests joining us. We have Ala from Citrini, we have Will Brown from Prime Intellect, then Michelle

A (80:44)

coming and a lot of. Not Citrini. He just co authored the piece.

B (80:48)

He co authored the piece and he also wrote a part one that's a very good read that was released before the mega viral essay. And then Mike's coming on from also capital at 150. So linear, of course, is the system for modern software development. 70% of enterprise workspaces on Linear are using agents and you should be too. So.

A (81:13)

Yeah, Nick says hot take doesn't matter if CEOs gamer. Strauss Zelnick has said it's perfectly. A CEO's job is to attract, retain and motivate the best talent in the business and then get out of their way.

B (81:22)

Yep.

A (81:24)

New Xbox CEO Asha doesn't need to be a gamer to run a company. She simply needs to do what the CEO's job of running a gaming company is supposed to do, which is to hire talent and allow game studios to make their creative vision come to reality. The thing I think they maybe could have done better with the announcement is like, talk about what the rest of the management team looks like. Because if you position Asha as this like, elite operator who's going to like, really, like there's a way, like if she's like managing a team of people that are gamers and love gaming and she's like working with them to figure out how to make the platforms better and better, that's more compelling than bringing somebody in.

B (82:10)

Yeah. I'm trying to think of other industries where the CEO doesn't use the product.

A (82:16)

Part of this is like. Industries where the CEO doesn't use a product. I mean, think about almost every category of enterprise software.

B (82:29)

I think they all dog food the

A (82:31)

product, but not on a personal level. Like their teams might.

B (82:35)

Yeah, I guess that's right. Yeah, I was trying to think of like, are there Like, I mean, it's like Rick Rubin doesn't know how to play instruments, but he does listen to the music. And I feel like most of the big, like, Hollywood agents or power players, like, maybe they didn't know how to use a film camera, but they watched movies, I believe. Like, I don't know if there's, like, someone out there who's just like, yeah, like, I've never seen Saving Private Ryan, but it made me a lot of money because I greenlit it because, like, I knew.

A (83:08)

Yeah. One thing's for sure. Xbox is not in founder mode.

B (83:12)

Yep.

A (83:13)

And will never be. But would it be?

B (83:18)

Should it be? I don't know.

A (83:19)

Well, how do you. The. The guy. The guy who started Xbox. What is the guy? That's. It's. Seamus Blackley is, like, credited with creating and designing the original Xbox. I think he's now in the. In the chocolate business.

B (83:37)

Yeah. I mean, I definitely think the CEO of a video game company can just provide a fantastic environment for creative individuals. And also, I mean, Xbox is a hardware company. It's also a live streaming company. It's also a studio where you just have studio heads that go in greenlight projects. It's not all directly related.

A (84:01)

Handel says Larry Ellison using Oracle on a nice Sunday morning.

B (84:04)

Let's go. Good job. That's correct. I don't know. He probably does store a lot of data in Oracle. Who knows?

A (84:13)

Yeah. Aaron says the CEO necessarily is not in the product daily.

B (84:17)

Yeah.

A (84:18)

Who knows? We'll wait to see. It's. It already happened.

B (84:23)

Yeah. Well, yeah, I mean, we'll see.

A (84:26)

We're working on. We're. We got in touch with her on Friday. Find a time for her to jump on the show.

B (84:31)

Yeah. PS6 might be delayed because of the memory stuff. There's also. I don't know.

A (84:36)

Yeah. They should just delay the next Xbox and let. Let Asha just game for three months, like you said. Six months.

B (84:45)

Yeah. The really interesting thing on the hardware side is a lot of people were freaking out over the weekend playing with ChatGimmy AI from Talas. We had the founder on the show. He has baked llama 38B onto silicon and so it runs at 16,000 tokens per second. So you ask it your typical LLM query and it just boom, loads the page. It's all done. There's no token streaming in. You're just at the bottom of the page. It's actually sort of jarring because then you have to scroll back up. But it's clearly, like, incredible. And this is coming and we've experienced it with Codex 5.3 Cerebras or Spark. Is that what they call it? And there's a few others proc. And so.

A (85:32)

And it doesn't have web search.

B (85:34)

Yeah, yeah.

A (85:35)

So if you ask it what is tvpn? It says Gwen Butcher's Pizza Network.

B (85:41)

We gave you a different answer this time. Wow. It's really hallucinating anyway that I think that the system on a chip Cerebras, the wafer scale, super fast inference is

A (85:54)

going to be very amazing for labs. It really is just a next token predictor. It says TVPN could also stand for the Black Pine Network. This is not a well known term or organization but it could be a fictional or made up name.

B (86:09)

It's having fun.

A (86:09)

It's having fun.

B (86:10)

But I think that there's a very interesting play where the gaming systems basically bake a style transfer diffusion module onto silicon and put it on the chip. This is what Nvidia did with dlss. Dynamic something super sourcing, super sampling, deep learning, super sampling DLSS. So if you have a Nvidia, what is it? GeForce like 4090, 3090. There's a section of the chip that's trained to take a 1080p video game and up res it in real time to 4K. And so you can run. If your hardware can only run the game at 720p 60 frames a second it will up res all of those frames. It's not perfect but it gives you a sharper image. It's basically just AI sharpening. That's happening. You could imagine a model that is trained to turn the images that are generated from a video game from Unreal Engine into something that is actually photo real. Like make it like a movie. That prompt that we've seen happen and you are like wow, that actually looks like a movie. You could run that in real time at 60 frames a second and be playing a video game that looks truly photoreal because the actual game engine graphics have totally plateaued and there doesn't really feel like they're just going to jump to cinema quality anytime soon. But if you use AI to do the last step, I feel like that could be really good. What do you think Tyler?

C (87:41)

Yeah, you could also do like a Genie 3 type model baked down.

A (87:44)

Right.

C (87:45)

So like interactive video.

B (87:46)

Oh yeah, yeah, yeah. It's really slow and limited right now but if you bake that down you can play that.

C (87:51)

Yeah.

B (87:52)

I still think there's a lot of work to be done on Genie 3.

C (87:55)

Yeah. Maybe it's like one or two more models.

B (87:56)

Yeah, like clearly that those are like llama 2 level right now. But yes, yes, yes, I completely agree. Anyway, let me tell you about Restream 1 livestream 30 plus destinations. If you want to multi stream go to restream.com Dan's Gaming says my theory is that Phil and Sarah did not want to shove AI into everything at Xbox. They're forced to retire and resign. Microsoft is replacing them with someone with a strong background in AI and no experience in gaming. This is just getting insane. I don't know, I have a very, I'm completely white pilled on AI in gaming as I just said. Like I think AI in gaming can be really, really great. I mean there's a ton of games where the developer would love to have NPC dialogue that they don't have to sit there and write, okay, this townsperson's going to offer you five coins for your sweet. It's like no, just like be an npc, you have coins, act agentically and then you go up and you're exchanging with the townsperson your sword for your coins or whatever and you have a much more natural interaction. That feels really great. I don't know. There's a million bull cases for AI in gaming. In my opinion, it seems like the hardest. It seems like one of the easiest things to sort of justify we have Mr. Shah. So let's tell you about Figma Ship the best version, not the first one with Figma, including introducing Claude code to figma. Explore more options and push ideas further. And without further ado, we'll bring in our first guest of the show, Ayla. How are you doing?

A (89:23)

What's going on?

F (89:25)

Doing great.

A (89:26)

How are you guys doing great. Is this your first time triggering a global sell off?

F (89:35)

The first time so far, but I'm just the messenger. Just the way I look at it, we've got a lot of opportunities and a lot of scary things coming down the pike.

B (89:42)

Okay, so yeah, take us through the thought process, like how long had this been simmering? What was the actual process of putting together this report? And then what do you want people to take away from it? And then maybe we can go into some of the reactions and your reactions to those reactions.

F (89:58)

Absolutely. The process ultimately is that I've been building an AI for 15 years and I've been an investor for 20. And so especially the last six months as I've just been using agentic coding myself and my teams have adopted it. It's just been a step change function and how much we can get done. And just thinking through, hey, how is this going to, we're early, we're a startup, you know, we're going to be at the leading edge of how people are adopting things. You know, assume the corporate world is a year or two years away. It's going to be pretty profound. And I think the underlying thing, you know, as sort of an amateur macro economist, is we're just not producing white collar jobs to begin with. I hadn't actually seen the extent of that until I kind of looked at, you know, specifically what we call like the information sector, so different parts of kind of technology. Those jobs are down 8% from the peak in 2022 already. And so those are the places where people are adopting the most aggressively already. And we know, you know, every week there's firings out of like big tech.

B (90:53)

Yeah.

F (90:54)

And so in that world, what happens when the technology that big tech's been using for a while has gotten a lot better and now your average corporate starts using it as well? It can get quite scary. And so we wanted to kind of think through the implications of that and the piece.

A (91:09)

But how much of those layoffs do you think are. We've talked about a bunch of those layoffs on the show. They're usually attributed to AI, but if you dig under the hood, it's like they just wanted to kind of resize or get more efficient or they're reprioritizing resources and not actually because they just launched some new agent and suddenly everything's changed. Hey, we don't need these thousand engineers anymore.

F (91:37)

So I think, you know, those are all great corporate euphemisms. And of course that's how they're going to say it. But I think the way I would think about this is it's not necessarily like agentic powers happened and now everyone's going to get fired. You know, agents and LLMs broadly are just sort of on the tech tree as a continuum from software. And so software has been making companies more efficient for decades. And you know, that has caused a lot of downstream effects. And now that software has just become much more intelligent. And so in that sense, I think, you know, companies that are efficient have been doing a form of this for a really long time. And we think about, you know, the age starting now in 26 is just something that's going to accelerate that.

B (92:16)

Okay, so yeah, what else was key in the thesis or maybe potentially overlooked that you think people should be really focusing on?

F (92:28)

I think the problem a first thing, the most important thing is Just the labor market dynamics. We've just been in a really weak labor market for a while, and that's before these things roll out. But then you put that together with the fact that we just have a very structural environment where what is the thing that drives our entire economy, it's wages. Most of those wages that are ultimately driving all the discretionary spending is coming from the white collar worker. And the problem with that is that we're now entering this place where you made all these assumptions on like loaning money to all these companies, to mortgages and everything else. Like white collar economy is our economy. If you all of a sudden just take a leg out of that economy, it has a contagion effect into basically every asset in the world. And so that I think is the part that people haven't thought about, because when people were making these loans, no one ever consumed the world in which. Wow, okay, now white collar jobs are in sort of permanent decline. Right. If that's at 2% a year, then I think we can skate through. But if it's at 4 or 5% a year, then we need action a lot more quickly.

B (93:34)

Is the white collar economy actually the full economy or is it more just like the stock market? Because it feels like white collar workers are disproportionately allocated to assets versus consumption. And you see things like, you know, like there's a lot of health in more blue collar sectors. Health care is growing. And then you also see dynamics like just, you know, like we've seen like jitters in the, in the consumer market for a long time. And then we just see the health of the American consumer just continue and continue and continue. And it feels like it's maybe driven by something like lower level. And there's always this disconnect in my mind between like the economy and the market.

F (94:19)

It's a great question. I think the issue here is that it's all just one labor market. And right now blue collar is doing better because there are not firings there. Yeah, I don't think, you know, I think robots are probably 18, like 24 to 36 months behind other forms of LLMs that are, you know, just diffusing through society. But the problem is, let's just say it's one labor market ultimately. And if there's no more white, if the white collar jobs are going away, let's say, you know, in our scenario we talk about 5% of folks might get fired in a couple of years. Those 5%, if there aren't white collar jobs for them to Relocate into, then they're going to have to move into the gig economy and the blue collar labor force. And so that puts pressure on the entire labor market, not just the white collar one. And to answer your other question, health care is growing, education is growing. The reason those things are growing ultimately and we did some work in our piece to try and isolate white collar that is not government driven. And so the government continues to spend more. That's why health care is growing. They're the biggest payer in, in health care. They're, they're guaranteeing all the loans in the, the education industry. And so those, those sectors continue to grow because government spending grows. But that's again, it gets very circular if government spending is coming primarily from taxes and primarily payroll taxes because the average worker pays a lot more in taxes, you know, per dollar than the average corporate does. And so some corporates make a lot more money. Workers payroll taxes go down more. Then there is a bit of a contagion effect into bonds as well there too.

A (95:42)

On Saturday, John and I were going back and forth about some of the really wild predictions around the impact of the Internet that were being made in the 90s. There was clicks replace bricks. People were predicting total die off. Well, I mean, to be fair, to be like, I'll just finish. They were expecting a total die off of all brick and mortar stores in five to 10 years, which was like widely discussed prediction. It was like, why would you ever go to a store to buy something if you could just get it online sent to you directly? Yeah, and I think a couple others. So like not as relevant to your piece, but people were predicting like permanent high growth, the end of business cycles. There was the like media disintermediation narrative which was like the Napster era. Everyone was gonna get all media for free forever. Newspapers would die off. Record labels.

F (96:43)

Aren't you guys the media disintermediation narrative?

A (96:46)

Yeah, we are.

B (96:46)

But yeah, it's all about timelines.

A (96:48)

20 years later and CNBC is still a much, much bigger business than all business media, at least in our world.

F (96:58)

But like newspapers, like magazines, completely gone.

E (97:02)

Right?

F (97:03)

All of that has moved to the Internet.

B (97:04)

Totally, totally. It's just like, sure, but like 5% employment shock in a unique order. Way different. I mean like a 5% unemployment shock is completely different if it happens over a quarter than if it happens over two decades. Right. Like these are just fundamentally way different things.

A (97:18)

Yeah, the other thing, yeah, the last thing I would say is like there was like this concept of like frictionless capitalism, meaning that like middlemen would be eliminated because you could just go directly to the source and that would push pricing pressure down. My question, and I know you guys are not writing your piece saying like, you know, this. We, we believe we will stake our entire reputation on, on this sort of narrative. But what do you think? What, what, how much did you pay attention to like the 90s, early 2000s Internet predictions? What do you think they got wrong? Why is this time different in terms of how a new technology will diffuse to the economy?

F (98:02)

I think the difference is if you just plot what's happening to technology, it's all just going exponential. These are all just continuous timelines of like, we have microcomputers, we have the Internet, we have mobile phones. And today, you know, we have very powerful AI. And so I think most of the predictions that you ticked off there, it's kind of interesting. I would, you know, just looking at them today, you know, I would say they couldn't really happen until you had proper AI because like, if you have the ability to just freely, you know, commerce, like have commerce the way you do today doesn't work if you still have to do all the work ultimately, like you have to go and you have to log and think about the friction, amount of friction there is in buying a product for most people today. Right. You still have to go to the website, you have to put your credit card in. It's all work. We only have gotten to kind of the tech required for those predictions I think this year, and that's why this is the year that I think it really begins, because now it is completely seamless. You just, and no one's really doing this yet, but it's going to happen. I think, you know, in the next six months is just tell your agent, you know, tell Gemini, tell ChatGPT, go buy these things, it has your credit card. And now that world that they were describing is truly going to come to pass.

B (99:10)

Yeah. What about the canary in the coal mine analogy? I was looking at unemployment statistics in India and the Philippines and it doesn't seem to be doom and gloom over there. I don't know. I didn't dig in super far. But would you at least expect that the unemployment rate would spike overseas before it spikes in America? Or do you think this all happens simultaneously? Simultaneously.

F (99:39)

It's a tricky question. I think ultimately white collar work is a lot more of our economy than it is the economy of India and the Philippines. And they are much sort of like more immature economies that are growing through investment and things like that. But certainly, I think we called it out. The consulting sectors in India are certainly going to be challenged in other places as well. But the reality is like the timing, timing is everything in the markets, clearly. But the trick here is if you're a corporate and you are hard pressed to get AI into your organization today, ChatGPT and OpenAI will send you a forward deployed engineer. If you have billion dollars in budget, right? If you have a $10 million budget, they're not going to. And so who are those folks turning to? They can't usually do it themselves. And so they are going to the outsourced providers, the accentures of the world. And so I think those businesses are likely going to be in a lot of trouble over the medium term, but they probably will have a big bump from people really putting that AI into their organizations first. And so it's a bit of a tricky timeline there.

B (100:39)

What moats do you think hold beyond this? Because I think a lot of people latched onto the DoorDash example as something that they thought had a moat. And in the post you sort of underline how that could maybe not be as durable as a moat as people thought. But in the long case, like what, what moats do exist? Like do network effects stay do complex coordination, intellectual property? Like what, what doesn't break down?

F (101:12)

You know, real brand value where people are choosing you over other things because of the brand and the status signaling across brands matters a ton. Sure, network effects are more powerful than ever I think in this world. So things like Meta really have a lot of lot to sort of gain in that sense. But I think things that look like their network effect businesses, but in fact are just the ones that are doing the hard work of aggregating demand and supply, I think will be more challenged. And so DoorDash is a good example there. It's not necessarily the biggest risk versus some of the other things, but I just was in a thread with Gavin Baker talking about this. But the problem for DoorDash and Uber and folks like that is right now they're doing two jobs. They're doing the job of aggregating demand and the job of aggregating supply. They're both hard jobs, but the demand side is the harder side. And we think the world of the future, there are lots of folks in, let's say food delivery. Instacart wants to get a bunch of market share and grubhub wants to get a bunch of market share. Let's say the agents are the ones doing the buying. It's 2028 and 40% of the sales are through agents. You just tell Gemini, hey, order me some noodles. In that world. Instead of it's going to go to each and every provider. And right now there are four providers that do that. But now it's very easy. If I'm building a startup in this space. Previously I had to get all the drivers on board, get all the restaurants on board and acquire customers. Now Gemini and ChatGPT are acquiring the customers for me and all I have to do is get this, get the supply side going so it makes it much easier for new entrants to come in and for existing second, third, fourth tier players can really say, I'm going to relax my margins, try to get more top line. And so you think that whatever the 15% vig is that DoorDash gets today, maybe it's more than that. Some of that I would think Gemini and ChatGPT are going to ask for themselves. Wherever I send the traffic, I'm going to get a piece of that and then some of that's going to go back to the consumer.

B (103:05)

Yeah, it feels like was this the most stretched or controversial prediction?

F (103:13)

It seems like it was certainly the one that got, you know, getting the most chatter. And I think we did it for a reason. We wanted to be a little provocative in thinking it through because, you know, it's an amazing business and they're gaining a market share. But the fundamental idea that you're. Because what did the lock in? Right. Like the drivers have lock in on doordash or on Uber? Not really right there. You know, most, most drivers are doing Lyft and Uber so they're, they're not locked in. The real lock in the real business value, the franchise value of an Uber or DoorDash is the customer lock in because the customer gets comfortable, they've got everything saved. They want to hit a couple of buttons. They don't, they don't price shop. Agents are happy to price shop as much as possible. And so if you take that away, then it's a real problem for businesses that are ultimately built on customer lock in.

B (103:54)

Yeah, yeah.

A (103:55)

I don't know. I think the interviews that we've had with the Lyft, I mean, you know, again, take, take it with a grain of salt. They have a narrative that is important to their business. But like, if you ask these people what is the greatest challenge, it is managing, managing the supply side, it is not the demand side is not where they're saying like, hey, like this is really what we need to solve it's like, hey, as we get more drivers on the platform, revenue naturally, naturally goes up. And so I just, I'm just hard pressed to imagine a world in which, you know, somebody think, think about if somebody in my town, which is like 15,000 people, like Vibe, codes a delivery, a delivery app, and I go into ChatGPT or with another agent and I say like, I want food. It's like the agent wants to get the best possible service. I would imagine the agent to route to the platform with the supply that is going to be able to deliver in the shortest possible time horizon. And imagining a world where there's like this Vibe coded small team operating that just happens to aggregate as much supply, which is just increases the likelihood that my order will be delivered on the best possible timeline, which is going to be the number one factor for customer satisfaction. I just don't see how solving the front end kind of demand piece actually makes a better consumer experience, which I assume the agent would optimize for on behalf of the user.

F (105:32)

So let's consider what actually happens here, right? You make the order on DoorDash. DoorDash sends it to the restaurant. The restaurant essentially. Sometimes they use their own driver, sometimes they send the drivers from doordash. But now imagine the agent can take you directly to the restaurant site and place the order directly with the restaurant. And you can keep half the savings and the agent can keep half the savings.

B (105:59)

Right, but where's the driver?

A (106:01)

Where's the driver coming from?

B (106:02)

Because I feel like I understand, I understand the customer demand side. Like you start with an LLM or an agent who shops around for you. So maybe that's solved. Maybe it'll find you just via SEO and you can just put out like, we only take a 5% cut instead of 15% and the agent picks up it.

A (106:20)

You.

B (106:20)

I understand getting all the restaurants on board because you email them and say, hey, it's 5% instead of 15%, they're sure we'll turn it on. But for the drivers, how do you actually reach out to them and get them on the platform? And how does AI lower that cost? Because right now I think about like, what was the driver marketing budget over the last decade at uber or at DoorDash? And it's probably like in the billions of dollars. And so I feel like just to generate that much liquidity, I have to invest that much to onboard all those drivers, build awareness. Maybe it just goes viral because they're like, hey, I can make more money here. But that feels hard.

F (107:01)

I think it's going to take time. But I think there are a bunch of smaller sort of driver aggregation networks that exist today that are not the ones that we know about. For instance, I started a business called Thistle and we do delivery of healthy foods to your door. We, we split it between half of them are own employee drivers and the other half, you know, I think we have like, like 500 or 700 drivers that we just use a third party service to provide. So I think there are a lot more of these businesses. All of those businesses now will also just have huge opportunities to kind of take market share. Ultimately what we're saying is the friction in doing commerce is going way down. Places where there are rents, the prices can go down. But ultimately this is just an opportunity for more and more entrepreneurs to kind of build businesses for the new world.

B (107:42)

Yeah, I think it's interesting because we're here debating this somewhat temporary thing because self driving cars, robotics changes all of that in a huge way. But we use the term sloppable for companies that can be vibe coded away and clankable for companies that can be disrupted by robotics. And I've always put the delivery services more in the clankable category than the slope category. So I was shocked to see what

A (108:09)

are the, what would you have spent more time on if you knew you were going to get 50 million views and the markets would react in the way that they have?

F (108:22)

I would have finished writing the third piece where I talk about solutions which I have not gotten yet.

B (108:27)

A lot of people are demanding solutions. You just hit me with a ton of problems. That's funny. Do you think that there's any. There's this question about like in my mind, like yes, Google and Nvidia are public, but anthropic OpenAI and XAI through SpaceX are not public. They're sort of like this massive, you know, multiple hundred billion dollar sell off in the public markets that sort of should, if you believe your thesis, that should sort of funnel to the labs. I would imagine if when I read it like there's a lot of doom and gloom about companies that are out there, but it's a lot of bull. It's a lot of bull case for AI labs, but that can't happen in one day because like rounds happen every once in a while. They're private, there's all these different things. But do you think that the world would change when the big labs get out in the public markets?

F (109:23)

I think it's absolutely going to change. I have a strong suspicion that anthropic is going to go, you know, in the next three to six months. They just have so much momentum and there's a lot of value being first. P and L also just looks a lot better than anyone else. So I would think that gets public and it's going to be pretty interesting if it happens. Certainly labs are ultimately, they seem like we're very well positioned to win. I would wonder over the medium term, like, you know, what happens with some of the Chinese models and whatnot, if people actually want just something that's more local and something that they own. But it does seem like the most likely outcome is, is going to be that the existing incumbents are going to get the most share. And I think Google is particularly well positioned since they already own all of those customers today and they can finance losses from inference a lot longer than everyone else. But I think ultimately, like there's a world in which the labs are the biggest winners here. There's also a world in which like you end up with just a lot more competition and people trade and change. But the thing that seems very clear to me that the absolute, like there's no way they won't be the hugest winners here is going to be the underlying tech, meaning the semiconductors. So everything.

B (110:25)

You could go even deeper. You could go into like commodities and like copper and energy and oil and natural gas and stuff. And people have.

F (110:32)

Yes.

A (110:33)

Did you see the.

B (110:35)

Yeah.

A (110:35)

Did you see the. Some of the criticism was that the essay was very Marxist.

B (110:41)

Oh yeah.

A (110:42)

He said Marx, writing during the Industrial Revolution, predicted capitalism would periodically devour itself. Firms replace labor with machinery to boost profits, but competition diffuses. The technology drives prices to marginal costs and the gains get competed away. Meanwhile, displaced workers lose purchasing power, hollowing out the demand. The whole system depends on production rises, but no one can afford to buy what's produced. The contradiction between production and realization. Citrini's piece describes this exact dynamic, then declares there's no natural break. It's the most Marxist piece. Financial analysis, not my word.

B (111:15)

I don't think that critique and makes

A (111:18)

the same errors Marx did. Yeah, creative destruction doesn't just destroy, it creates industries we can't yet conceive of.

B (111:26)

Yeah, that's interesting. I mean, maybe that's going in the solutions.

A (111:28)

Is that going into solutions?

F (111:30)

So let me address it a few ways. Marx was a really smart dude. He got a lot of things right. Very early Marxist can mean communist. Marxist can also mean just understanding how capital, nature interact. And in that sense, yes, it is Marxist. He had. He was very Insightful. But I think the thing that, that we're missing here is that it's there, there's, there's the economic layer, but ultimately it's the political layer that matters. And, you know, we're in a world where we've, we've had two parties and both parties, you know, economically have a little bit of difference, but not a huge amount of difference. And so we kind of bicker. But in a world in which jobs are going away really fast, I think there's going to be a much stronger alignment for just the laboring class overall to say, hey, we need to fix this problem. It's a very fixable problem. What we're actually expounding here is that gdp, if done properly, will absolutely explode. We're getting way more efficient. We've built a machine dot, we build machine intelligence, but we have to structure our society such that as those things happen very, hopefully very slowly, we do the right thing from a taxation perspective to say the winners should win, but if that's what's causing the displacement, let's make the pie a little bit bigger for everyone. And that I think ultimately should be something that appeals to a lot of folks in the AI complex because if we don't, then something like this is likely to happen and AI progress will slow down because we'll have an economic crisis and we're not going to finance nearly as much of it as we are otherwise would.

A (112:59)

So do you think the future is what anthropic head of sales position in France. The company will be spending €530,000 per year. The government will get €340,000 and the employee will get 190,000. Is that, is that the level of taxation you think we're headed for?

F (113:20)

I think when we're at, you know, France's level of government spending, then, you know, the math probably means roughly that I would say that, you know, government spending would be at France's level, I'm guessing, like, you know, five, seven years from now if this, if this scenario kind of comes to pass. And so I think we'll head there over time, but I think it's less a question of the percent of spending and how much goes to the employee versus goes to the government. And ultimately what is the size of the total pie. So the bet here is that the pie, if done properly, can just increase multiples of what it is today. And thus, you know, it's just a win, win.

B (113:54)

One question. I mean, it sounds like you're working on potential solutions post which I'm very excited to read. Thank you. I'm interested to know your reflection on the messaging that's coming from the leaders of the AI labs because they've outlined many sort of low probability but potentially negative scenarios. We have the white collar work number. We've had many of these comments from lab leaders, but I rarely hear them follow it up with and the answer is print, print, print or interest rates will be will save us or unemployment insurance or UBI like all of those like solutions that I think people. It's funny because people are quoting your post being like, this is easily solved with this solution. It's like, okay, well that's great if we all agree and I think you might with some of the quotes. People are all over the place. But I'm wondering about your reflection on the messaging from the labs around solutions versus pure focus on problems.

F (114:58)

I think it's a really interesting question and very interesting setup in that the labs on the one hand want to get the word out there. And so Dario especially has been the loudest here. There's a really good Axios article from last year, May, where he's kind of sound of the alarm bells. People aren't really, he's like saying people are not listening. Obviously a lot has changed since then, but they can't go so far as to say like, hey, if you put the pieces together, then this is how it's going to play out. I think it's too sort of damaging to sort of their reputations and like, you know, their ability to fundraise and things like that. And so I think it's other folks like ourselves that kind of have that duty to go and really start thinking that through. I think it seems like anthropic is pretty engaged should that conversation really start happening. And I think this is the year it needs to really start happening. And so I think they all kind of get it. And so it's just a question of like, how do we as a society start moving in that direction?

B (115:50)

Yeah, I think, you know, obviously there's. I'm still processing part of the piece. I agree with some of it, I disagree with some of it. But what's really underrated is just like how useful this process of writing an article for a particular audience is. Like, I disagreed with a lot of something big is happening, but it hit with a very different audience than machines of Loving Grace or the adolescence of AI or of Machine intelligence. And there's pieces that are written for AI insiders, leaders, researchers. Then there's the broader tech community. Then there's like everyday people and you clearly hit the nail on the head with like speaking to the financial community and we see that in the markets. Not amazing results, but maybe it's, maybe it's worthwhile because we will get really great solutions and a better conversation around it. So I think in due time this discussion needed to be had. So thank you.

A (116:56)

What's an industry or job of the future that you could see emerging?

F (117:02)

I think again if we solve this like everything related to sort of leisure is going to absolutely zoom and that those are going to be the biggest growth industries of the future. Right. Like what do humans want to do?

A (117:13)

Total victory.

B (117:14)

Watch polo. Watch a cloned horse play polo. For sure. Yeah.

F (117:21)

So you know, imagine humans have like the entire day to just enjoy themselves instead of having to work.

B (117:27)

Now that is something I've been promised for 100 years. So I'm deeply skeptical. But this time is different. I want it to be different. Let's bring on the leisure boom. I'm here for it.

A (117:37)

I'm here on anything in your solutions doc around reindustrialization. The frustration that so many people in tech that have been building in hardware in the real world or trying to recruit people that are getting offers from social media companies or Now Labs or SaaS companies. One of the problems for America in the last 20 years was that if you just wanted to make $100 million, you probably were much more likely to do that building enterprise software than building critical infrastructure or anything in the real world. So is kind of new infrastructure and re industrialization a potential landing point for people that had the 180k a year PM job that might be going away?

F (118:32)

It's a great question. I think there's certainly going to be a lot more opportunity in those sectors and I think we've done some pretty smart policy things that are moving us in that direction. But we're also just in a lot of ways so far behind China there and doesn't I affect those jobs both for you know, on the re industrialization side just like it does for writing code. And so that's where I think it will get trickier. I think over as a country we're going to spend an awful lot more on that and I think we're going to, we're going to catch up but we're not, it's not clear that's going to be through just creating a bunch of additional jobs versus you know, the ultimate thing we're seeing with AI period is just high agency people who really know how to use the Tools can just do the work of many, many people.

B (119:13)

Yeah.

F (119:13)

And I think that trend applies in every industry to some extent.

B (119:17)

Yeah. What an exciting time. Thank you so much for taking the time.

A (119:21)

When's the next piece dropping?

F (119:24)

Hopefully by the end of the week. But don't, don't, don't hold me to that.

A (119:29)

When you know that then it could be hard for the follow up to get as much reach as this one. That's kind of the way these things go. But now, now the pressure's on to really pay attention.

B (119:38)

Just don't, don't have any. You'll be fine. We're excited to read it and we'll talk to you soon.

A (119:44)

Yeah, great to meet you.

B (119:45)

Have a great rest of your day. Thanks so much. Let me tell you about Turbo Puffer, serverless vector and full text search built from first principles and object storage. Fast 10x cheaper and extremely scalable. And I'm also going to tell you about Gusto, the unified platform for payroll, benefits and HR built to evolve with small and medium sized businesses. And without further ado, we have Will Brown from Prime Intellect in the tvp.

E (120:09)

How's it going?

B (120:10)

Welcome to the show, Will. How are you doing? It's been too long.

E (120:13)

I'm doing great. I'm doing great. It's a, it's great to be back. I think this is the fourth something like that I looked at. There was a list at some point of the record and some people are like, have been on.

B (120:23)

Some people have been on, on something like.

A (120:26)

I'm looking forward to the 400th.

B (120:28)

It's great.

A (120:29)

It's been a lot of fun. What are your, your old buddies at Morgan Stanley thinking about the current thing in tech, the, the 2028 intelligence crisis? Have you gotten any messages?

E (120:41)

That's a great question. I have not had the full deep.

A (120:44)

They're too busy. They're too busy hitting the sell button.

B (120:47)

No, everyone in Morgan Stanley is too quickly too busy setting up Mac Minis to run open claw. That's what's happening because they all just read something big is happening.

E (120:56)

Right. And so like I think there's definitely a lot of opinions on, on all sides and I feel like, like that to me the piece was pretty cool. I don't necessarily agree with it, but I think it was effective at getting people to have more interesting conversations than for example, recent, other maybe viral pieces about how everything's going crazy. And it seems like the conversation ended up getting into the weeds of monetary policy and how people are going to react and how hard Is it to vibe put a doordash clone? And these sorts of things I think are actually the sorts of conversations that are good for more people to be having. Whether or not a certain prediction is right. I think it just generally as stuff is getting crazier, I feel like this is the sort of stuff that allows the rest of the world to kind of hear about from their friends a little more grounded discussion about what could happen.

B (121:46)

Yeah, well, give us an update from prime minelect. What's going on in your world.

E (121:52)

Yeah, yeah. So there's a few things that I think are interesting as well as I want to talk about just today given some other stuff that's happening on the timeline. But so couple weeks ago we released a training platform to make it really easy for people to do RL on top of leading open source models with their own environments. And we've tried to really make the make it an agent native experience where you're kind of like there's a some model people have been kind of tweeting out their experiences with. The term people have been using is vibrl, which is we're kind of now at the point where the infra to manage the training is kind of in place and you can do it without thinking about the hardware and the GPUs where the models still kind of struggle, but you can really focus on the environment and designing your tasks and specifying what you want and having. Turning existing data that you already have into training recipes. And so we're kind of at the point now where like this is pretty accessible for people to kind of go train models. And it's been pretty cool kind of seeing people have fun with it.

B (122:47)

Yeah. Concretize some of the like actual applications. I imagine this works best if everything flows through, through text, flows through CLI tools. Like because when, when I think like okay, great, I'm going to set up an RL environment and automate one of my workflows and I'm like, well, I'll need to open Adobe Premiere which has a license and then I'll need to go to YouTube and download some videos. I'm just thinking about like editing a short video we have.

E (123:12)

Of course, yes, some of that stuff. Definitely. Like there's definitely a range of like simple to complicated.

B (123:17)

Yeah.

E (123:17)

But I think there is a lot of sweet spots where it's like the it's coding tool use and interacting with kind of app simulators which are the sweet spots of a lot of these like focuses for people doing training in the labs anyways. Rather than like full fledged Photoshop. There's a great tweet actually from figured the ramp guys from Ramp Labs showing off how they'd been using it for some stuff. It was the team behind Ramp Sheets. And so you could kind of imagine like the sorts of things there where like you don't actually have to build the whole thing, but you can like. And this is where I think the coding agent stuff is really useful is there's a lot of ways you can kind of have the right simulation of the application that is like it isn't necessarily the full backend, but it's enough to be able to capture the task. And the coding agents are good enough that with Human in the loop using like it's all CLI data. So you go into your terminal, you, we have our cli, the prime cli and you use that to kind of like set up your skills and get your coding agent configured and your agent's md. And so we've tried to like make that really smooth, but then you just are kind of like talking to your agent about, hey, my data's here, my app code's over here. Let's put it all in the right place and kick off some runs.

B (124:24)

Yeah, where are we on the path to personalized rl? I'm thinking back to like the RL that went into RLHF around ChatGPT GPT 4. And I remember it was like they had maybe tens of thousands of contractors sort of grading responses, giving thumbs up, thumbs down, giving feedback, varying levels, like a really large scale generalized process. If I'm running a medium sized company, is this something that I can pull from logs of what's happening in the business? Should I be firing up a data labeling company to help me generate more data? Because in the long term I would love just a screen recorder. Watch me what I do. And then it's rl' ing and then it's getting better and all of a sudden it can just do what I do with just like one prompt.

E (125:14)

Yeah, it's pretty close. It's not that sci fi, like if you have stuff especially let's focus. If we focus on like text or image input, full screen recording gets a little tricky.

B (125:25)

Sure.

E (125:26)

But if it's like text or image input that kind of comes from like agent logs and it's like human inputs text and images to an agent log and you have these log logs and you're trying to synthesize these logs. I think the trickiest part is refining criteria about what counts as good for Rescoring another tribe. But a lot of times the criteria are either pretty general across tasks or you can infer a lot of them from a user's response or just from the initial prompt. And from what we've seen, especially for a lot of very concrete problem solving use cases more so than, let's say, creative writing, but for things where it's like there's a right answer and it's not too hard to see if the model got the right answer from doing an agent trace, either from the humans response or let's say that if companies want to have their humans label as part of using the product by default, yeah, this is doable. And the RL recipes are kind of stable and scalable enough that it doesn't always work, but it works reliably enough that it. I think the barrier to entry and the cost are just like at a point where it's cool to see that this is now a thing people can go do and we're seeing a lot of people have success with it.

B (126:32)

Yeah. How are you thinking about the debate between MCP and cli? Peter was going back and forth and it was something that I was wondering. Even when MCP came out, it seemed really cool, but at the same time it felt like, well, the front end and I remember going to the front end and inspect element and see what's coming across and oh, there's an HTML request right there. Let's reverse engineer that.

E (126:58)

Yeah, it's all kind of the same thing. It's sending requests. And so I think people realize models were good enough at coding that the skills are essentially it's doing the same thing, but it's just you have more flexibility to. I think the area where MCP makes the most sense is when you really want fine grain, like auth stuff going on where there's like kind of credentials and you want to be able to kind of notice that it's being done and have the user approve certain requests or not approve others. That's where the formalism of the tool call is really useful, as opposed to it just being code that has an API token. But from the perspective of capabilities, skills are nice. MCP has its areas where it makes sense, but it's really just models using computers. Whether it's MCP or code or skill files and reading docs, it's like models are pretty good at reading stuff and if it has instructions on how to do a thing, they can kind of just do the thing. And they can do that for a while enough that it's Useful.

B (127:54)

Yeah. How intermediated do you think this product will be? And what I mean is that like let's just use some toy example like widgets company would benefit from a custom fine tuned model or RL model but they don't go to you. There's actually a company that's an intermediary that is providing like a SaaS product that then is fine tuned on like anonymized industry data or they went and generated RL or they'll even come to the company and say hey, we'll Hansel well prime cli. You're not going to have to know what that is. You just give us the data and we'll act as like your customer. How do you think that plays out in the market? Market?

E (128:38)

I mean it's definitely going to happen across the spectrum. I think the people who we'd work, we work the most directly with are the ones who are a little more like AI native and the ones who are going to work with kind of because I think we're really like building for developers as our kind of target audience but not necessarily researchers. So I think like the people who are like following the benchmarks and reading about the new model releases and building with cloud code and the agent frameworks, that's really like our target audience. People who like think about evals and prompting versus people who don't think about that. So we do actually work with a lot of the big data companies where there's. I think maybe the one interesting story is there's a lot of market demand for. Because everyone's building environments and selling them to the labs. But you can see a lot of these companies want to know that their environments are good. And so using RL as part of this process is the way that you evaluate the quality and be able to prove like hey, we got the good stuff. Yeah, because it actually improves capabilities. And so there is this whole economy of companies that really specialize on building environments and working with data. And I imagine this does become a big part of like the way that this stuff is consumed by end companies is through people with that kind of expertise at the data level.

A (129:50)

Jordan, talk about what the Chinese labs.

B (129:53)

I was going to ask the exact

A (129:54)

same thing in terms of distilling American models. Talk about kind of the scale.

B (130:02)

I've seen some rave reviews. I've genuinely seen some rave Reviews of Kim EK2 and then at the same time I've also seen like hey, it kind of fell flat on its face when I pushed it beyond a toy example. So yeah, what's real and what are you experiencing?

E (130:17)

Yeah, so I think they're definitely a couple months behind. Like they're not at the 4.6 at the codecs 5.3 level. They're pretty close to what we had before for that. And I think that's kind of where it's been and it feels like this is tightening. But I think at least where I get much excited is like they're good enough that going the extra mile with customization is a differentiator where you can take a model that's already almost frontier and make it the best model in the world at your thing pretty easily and pretty quickly. And so I think that is, even if you have to do this every three months, it's always a capability's race. But I think if this pipeline, if this process of like taking your data and improving the latest model becomes really easy and repeatable, then it's like you can get a lot of value out of doing that. And I think that's the sort of thing that's going to be in a lot of people's toolkits. In terms of the open source models generally. I think there was some interesting debate on the timeline today that I dove into for a little bit around anthropic and deep SEQ and doing distillation. And I think it feels like there's kind of two things. There's the kind of geopolitical element, there's the kind of terms of service of like, oh, they're doing bot farms, they're scraping, that's not allowed. Then there's also the idea of distillation more broadly of is it. And the two. I totally get the first two, but I think the thing where I was kind of like trying to push back a bit was like, I mean, everything on GitHub is someone typing a prompt to Claude and submitting it to Claude code and then they're going to review the pr, then they're going to merge it and this is like perfect training data. And so the Internet is just getting flooded with perfect CLAUDE distillation training data.

B (131:52)

Interesting. Yeah.

E (131:53)

And there's not much you can do about that. And so it's like, is distillation really the hill we want to die on?

B (131:59)

Okay, yeah. I guess the secondary question is like, put all of that aside and then just ask the question of like, is there some ticking time bomb with using a distilled model where you run into some wall or you have some problem in performance down the road? And so yeah, you're doing well in benchmarks but it is a less effective. And is that like actually problematic from a business perspective or is it just like okay yeah like I'm getting three months behind but it's three, you know, three times cheaper. So I'm fine with that trade off versus like I thought I was using something great and then it, it blew up on me.

E (132:39)

Right. So it depends a lot on your application. So they think there's certain things that like the models are already like more than good enough and these are like kind of more commodity like extraction or summarization or labeling use cases where like you kind of just want to optimize for cost. In some cases you want to optimize for speed. If you want to optimize for performance then if like cost isn't a concern and you really just care about top line performance, then customization is really where the open source models become interesting which is that like you can do more to the open source models than you can do to Claude and you can have a lot more fine grained control of like saying hey this is my eval, this is how I'm measuring performance. We are just going to hill climb this and then it's up to you as a business to define your business logic. Say hey, this is what I actually care about, this is what performance means. And I think we'll see a lot of companies realizing that like that is a useful knob to be able to turn to be able to like. And I think concretely what it'll look like for a lot of cases is there'll be these multi agent products that have their main orchestrator agent that's like one of the frontier models with lots of specialized sub agents for things that are related to the business and specific workflows which are then fine tuned models. I think that's kind of what we see currently as the most dominant paradigm for mix and matching between the proprietary models and the fine tuned open models.

B (133:58)

If you had told someone a year ago that there were going to be probably millions of people running agents locally with custom setups and MD files for various skills, they'd probably be like wow, that's pretty aggressive. Do you think that we'll be in a world in like a year or two where at least people on X will be talking about like my fine tune, I did RL on my pacific problem. My personalized agent is even better now because I did the rl.

E (134:33)

I mean so we see it today already with this a little bit where it's like, I mean there's people who are showing you can get these models to beat any of the closed source models on sufficiently well scoped tasks pretty quickly. It's not rocket science. You can basically vibe code it. You have to be clear that you have a goal in mind. But if you can define the goal and you can spell this out in English and you can do the same sort of prompt that everyone's doing for coding, then yeah, you can just kind of plug it in and get trained to work. But I think it'll become more like a lot of it is still very much like these kind of more proof of concept or narrow research cases. But it does seem like it's quickly especially like code becomes cheap and the cheaper that code gets, the more complex you can make your environments. And I think like a year ago we saw like cloud code's about a year old, came out I think February last year and at the time it was like, wasn't actually that useful yet. But I remember playing with it and feeling like, oh, this isn't actually something I want to use that heavily today because it's kind of slop, it's very chaotic, it just makes a mess. And I went back to cursor for a while because it was much more controlled, but it was like, oh, this form factor feels like it could eventually work. And I think there are other form factors today that don't actually work yet in some ways like the open cloth thing where it's like OpenCloud kind of works, but there's also a lot of trouble it gets into. Same with if you saw the Gastown thing or these crazy multi agent systems where it's like they aren't actually excellent yet for shipping quality production code. But the thing we had a year ago now is at the level where Claude code is used for most production code but by the heavy adopters or codecs. And so it feels like it is a matter of time until the these things stabilize and the goals of having that system end back up in the models for people training for it. But the recipes of how to train these models, they've become robust enough over the past year that it does seem to be a good idea in a lot of these cases to optimize your models for the structure you want them to be in. And if that structure is this crazy multi agent system thing, it's like yeah, why not?

B (136:48)

Yeah.

A (136:50)

Are you expecting real tangible breakthroughs in the first half of this year? I mean our intern keeps saying that he's close to cracking continual learning.

E (137:01)

Oh yeah, learning is going to fall pretty quickly. I Think. Do you think it'll be such a big thing? I mean, I think it's more of an engineering problem.

B (137:09)

Explain.

E (137:10)

No one's actually trying.

B (137:11)

No one's actually trying. Why not?

E (137:12)

No one, like. No one like OpenAI and Anthropic don't want to continuously train their models for each user.

F (137:17)

Like that's.

E (137:17)

It's expensive and annoying and hard to serve at scale. But, like, from a research perspective, like, we're. We do continue learning where the model learns new. They just keep training the model more and it knows more stuff because they put more Internet in it.

B (137:29)

Sure. And yeah, like, yeah, yeah, yeah, yeah. Uneconomical right now, but. But, yeah, but that's very.

A (137:37)

For a product like Frontier, I could imagine that that would be a selling point if you're McKinsey and you're going to a big institution.

B (137:46)

So, yeah, if you hypothetically, like, I don't know, you're a law firm and there's some crazy case update, like. Yeah, the model retrains on that. Like the day that the Supreme Court completely changes the way the law works and then everything else is, like, interpreted from that. Yeah. Makes a ton of sense.

E (138:01)

Yeah. There's enough kind of tricks. I think there's a lot of experimentation around, like, exactly the record recipe that's going to be the most reliable. But we kind of have a grab bag of like six or seven tricks that kind of work or they work in different ways and you can mix and match them. And it's just going to be like whatever's the best combination of these tricks. People are going to experiment with it and find the versions that work the best. And there doesn't seem to be any, like, big wall inside that prevents that from, like, being practical.

B (138:27)

That's cool.

A (138:29)

What are you tracking on the silicon side? We were playing around with Chat Jimmy AI.

E (138:35)

Oh, yeah, that was sick Craz.

A (138:38)

Jimmy's quick, but is he smarter?

B (138:40)

Too fast. You have to, like, scroll up once you get the answer.

F (138:45)

Yeah.

E (138:45)

I was trying to see how many tokens I could get it to print so that I could actually see it go. And I was like, give me every number between one and like 10,000. But like, llama just won't do that. No matter how you prompt it, it'll always stop after like a few thousand tokens.

B (138:57)

Oh, interesting.

E (138:58)

So you can't actually get to feel it, like, blitzing past.

B (139:01)

Whoa. Interesting. Yeah, yeah, yeah, yeah. It was like sort of a throwback experiencing llama 3.8b. Because I remember when that model came out and There was a lot of hype because open source developers just love open source stuff and it was exciting and it was cool. It was like, wow, they really did train a big model and they just put it out there. And I remember some people being like, yeah, like if you actually go talk to it, like it hallucinates a fair amount. Like I don't know that this is like actually at the frontier. Might have done okay on some benchmarks, but it's not quite there. And it was a little bit of a throwback. But you can just imagine baking any of the current frontier back there, giving it access to tools, giving it a reasoning loop. Like, yeah, it's going to be. Even if it's only 10 times as fast, like that's still so much faster than like, okay, got to close the app and come back after 20 minutes because my thing is running now. It's going to be a completely different and I think it'll be a big step change for people that are like, oh yeah, AI hallucinates. And I need to check that out. It'll be like, no, it's like totally. You can just have it right there and it's perfect and it works a ton very fast. It's gonna be a really cool moment. Will you be buying an AI lamp?

E (140:09)

I want the one that goes out of your bed and folds your clothes.

B (140:11)

Oh, okay.

E (140:12)

Have you seen that one? It looks like a Pixar.

B (140:16)

It also looks like it might dismember you if it doesn't like you. It's a little bit horrific, but I do agree if it folds your laundry, that's pretty, pretty amazing.

A (140:24)

I don't care if there's a 1 in 10,000 chance that it goes crazy.

B (140:30)

I don't care if there's like a 1 in 10 chance of me just being dismembered in the middle of my night because it gets mad at me because I was trying to prompt injected or something. No, I am excited for hardware. It feels like even the first gen hardware, like the Humane AI pin, the Rabbit R1, all that stuff with like frontier models starts to get interesting. I really hope we get a solid next iteration there. Even though it's obviously very much outside of your core competency. But maybe some hardware developers will be coming to you looking to fine tune a model RL model.

E (141:05)

Do you want local on device for something that's way to like because yeah, you can, I think especially for like these narrow things like if the R, the Rabbit, whatever. And this is also Apple strategy, it seems like is Apple's like they like keeping stuff on device.

B (141:17)

Yeah.

E (141:18)

The whole privacy thing is part of their whole pitch. And so I think part of the reason why Apple's been slow on the AI stuff is they're shipping a feature once they can do it on device with sufficient reliability.

B (141:29)

Yeah.

E (141:29)

And so that means they're slower in their rolling out of features, but it means that, like the stuff like summarization and the image search, like, they can do this locally now because the hardware is good enough and the models are good enough at that scale.

B (141:39)

Yeah. Yeah. You have to imagine that that the same talus principle of like baking the model down to silicon. Well, it feels like they're doing something maybe like wafer scale, like not iPhone scale. So, like, maybe that's another couple years and then you need another couple years to get it to. Okay. It's now frontier on a chip that's the size of your phone, fits in your phone, doesn't suck your battery down, but you play that out and you get to something like, really, really fun and interesting. I'm excited. Future is bright.

E (142:10)

Yeah, definitely exciting. I think the people always said the Internet was going to run out of data, but I think what we're. We're getting more data, but it's better data because it's just from the last generation of models.

B (142:20)

Oh, interesting.

E (142:20)

And so you can kind of like, you kind of get this flywheel of like there's just more data to learn from and it's all getting better as the models get better. And then you do more on top of that to boost beyond where you were from the old data. And that's where the RL and the filtering comes in and the human data.

B (142:34)

Yeah.

E (142:34)

But like, seems like you just have a pretty clear path of models getting better as you put more data into them and we have the data.

B (142:40)

Well, thank you for coming on the show and producing a bunch more data. That's helpful. It goes on to YouTube.

A (142:46)

It's an honor to produce data with you.

B (142:48)

It's an honor to join the training set with you.

E (142:51)

Yeah, that's the goal.

B (142:52)

That's all you can do. And thank you to everyone in the chat who's also providing data for the Internet. It's God's work.

E (142:58)

Thanks for having me.

B (142:59)

We'll talk to you soon. Will have a good one. Let me tell you about Plaid. Plaid powers the apps you use to spend, save, borrow and invest securely. Connecting bank accounts to move money, fight fraud, and improve lending now with AI. And speaking of data, let me tell you about Labelbox. RL environments, voice robotics, evals and expert human data Label Box is the data factory behind the world's leading AI teams. And I believe we have our next guest already in the Restream waiting room five minutes ahead of time schedule. Michelle Lee from Medra is in what's going on? The TBV at Eldrim. Welcome to the show.

G (143:34)

Hey guys.

B (143:35)

Hey.

G (143:36)

Excited to be here.

B (143:37)

Thank you so much for having me.

A (143:38)

Great to have you and your robots.

B (143:39)

Punctuality. Oh, yes. What is behind you? Wow, there's a robot that's actively working. Explain. Introduce yourself, please.

G (143:47)

Absolutely. So I'm Michelle. I'm the founder and CEO of Medra. And a little bit about me, I study chemical engineering. In undergrad, was there a typical chemistry life science nerd. And then I did an internship at SpaceX and was just really excited about like, what if we can build in the physical world, right? Like I wanted to build in the physical world. I want to build real things with real impact.

B (144:12)

Yeah.

G (144:12)

Ended up doing my PhD at Stanford at the Stanford AI Lab in Robotics, building robotics foundation models. I worked with Jeanette Bogue and also with Tae Fei Lee, Shout Out World Labs. And I ended up, when I finished my PhD, decided I wanted to combine life sciences, robotics, AI and I started Medra and we are building physical AI scientists which we think that is necessary to eradicate disease.

B (144:44)

How narrow do you want to go to start? I mean, it feels like there's pipetting, there's centrifuging, there's different stuff going on behind you. But like medicine, bio, these are massive terms. Can be animal studies, mice models, you could have monkeys in there. There's a million things that you could do. But I feel like you probably want to pick a beachhead. But you tell me what the strategy is.

G (145:07)

Definitely look like one day we will have Medra robots doing animal studies. Like mark my words, right? But you're right, we have to start somewhere. And we are starting with early discovery and development. We have physical AI robots at Medra that can do experiments at scale. We can work with instruments that humans already can use. And most importantly, we truly have intelligent robotics. This is not just lab automation where you program things and they do it exactly like you tell it to do. This is actually physical AI autonomy that is intelligent, that's constantly sensing, making corrections. And more importantly, we also have an AI scientist that can actually reason about the science itself.

B (145:53)

So what is an example in the lab where you actually do want some probabilistic reasoning or some stochastic result as opposed to something deterministic. Because if I'm vibe coding a website, I don't want it to guess what an HTML tag is. I want it to just use a div every time. It does a great job at that. So I imagine there's some things where the pipette always needs to go in the same place. So it's okay to stand on the shoulders of giants and puppeteer that. But where does the variability come in?

G (146:26)

Definitely, I think, like, if you think about the best scientists, right, the best scientists are the ones who are reading all the papers, they have all the scientific knowledge, but they're also the ones going into lab and running the experiments. They can like, sense what's happening, they can smell it, they can like visualize what's going on, and they can make changes, as they say, see things start happening inside the experiments. That's what we're trying to capture, right? The ability to be really flexible, to actually reason about the science as it is happening. And also taking into account all of the knowledge that's come before us, all the scientific papers, all the different results, all the past experiments you run, that's actually what enables good science.

B (147:12)

So tell me about the distribution business model. I could imagine a world where you're basically doing drug discovery, going through the FDA process at the same time. You could sort of sell a lab in a box to a pharmaceutical company. There's a lot of different ways I could see this taking shape. Where do you think this goes?

G (147:32)

Yeah, we are building the infrastructure layer. We want to be the TSMC for drug discovery. So we are partnering closely with pharma companies, biotech such as Genentech, where we are. They can either work with us by using our system, our physical AI scientists in their own lab, or we're actually about to open our own lab, our own fully autonomous lab, one of the largest autonomous labs in the United States in 2026.

B (148:00)

Talk to us about the fundraising. I think we missed you on the day you announced your Series A, but I still want to ring the gong. What happened? How much did you raise?

G (148:09)

Yeah, we raised $52 million for Series A. Thank you.

A (148:17)

Amazing. Who did you raise it from?

G (148:20)

Yeah, Human Capital led. They came in and pre seed and seed and they tripled down on us for Series A. Really preempted the raise. We also have Lux, who is also a repeat investor, also Menlo Ventures, Cataglio. Great investors joining in for a very ambitious mission and very ambitious journey of eradicating disease.

B (148:43)

And 52 million Series A. That feels like a lot of money. Congratulations. But is is I could imagine spending it on a training run for a foundation model or buying a bunch of robots like that stuff behind you. Doesn't look too cheap. Where do you see the money going? What does it unlock?

G (148:59)

Well, actually, the hardware that we use at Medra is all off the shelf. Robots right now especially their hardware is fairly commoditized and we use this off the shelf hardware so we can build AI on top of it, so we can actually reason about the science and actually be able to use what we have trained ourselves, which is the Vision Language Lab action model, to be able to autonomously run experiments. And a lot of what we have raised our Series A for is actually to open our own lab right in San Francisco to be able to scale up data generation. Because ultimately what we want to do is to be a data foundry for life sciences, to be like a mercurt, but for biological and life science and chemical chemistry data so that our partners can train foundation models in biology.

B (149:49)

Yeah, because there's probably not a lot of really clean data out there on GitHub or out on the open Internet and so you have to sort of generate it yourself. Is that generally correct?

G (149:59)

That's right. That's right. I mean, if you think about in biology, like the largest biology foundation models are still about three orders of magnitude trained off three orders of magnitude less data than like, you know, even like. Oh, one.

B (150:12)

Yeah, yeah, yeah. I think Google launched one that was. Showed really impressive results and the scaling laws were there, but it was much smaller than what you see elsewhere. So. Yeah, very interesting, very exciting, Jordy. Anything else?

A (150:24)

No, this is super exciting.

B (150:26)

Congratulations in the future and I'm sure we'll have you back on the show soon. Have a great rest of your day.

A (150:31)

Great to meet you.

B (150:31)

We'll talk to you soon. Let me tell you about Shopify. Shopify is the commerce platform that grows with your business and lets you sell in seconds online, in store, on mobile, on social, on marketplaces, and now with AI agents.

A (150:43)

And we got a little bit of time. I think we do.

B (150:46)

That's great for our next one. Well, then I'll tell people about the New York Stock Exchange. Want to change the world, raise capital at. At the New York Stock Exchange.

A (150:56)

And let's pull up this post from Gucci.

B (150:59)

Gucci, what did Gucci?

A (151:00)

Not the kind of account we pull up every day. They say Primavera February 27th, 2:00pm CET. And this picture is created with AI. They hit this, they drop this on main and the photo looks any Hallucinations.

B (151:18)

It looks.

A (151:19)

Looks completely tiled. It looks like it could have been out of any catalog over the last 20 years.

B (151:25)

I'm sure it's really peaceful over there at the Gucci offices now.

A (151:28)

Yeah, I'm sure this wasn't controversial at all.

B (151:32)

No, but they're going. This does feel like a more tasteful dipping your toe in the AI boom than say, the Svedka ad where everyone was kind of like, this just. It's not polished enough. It's still in the Uncanny Valley. And I feel like we're going to go through the same thing as cgi where, like, there are some terrible movies out there that are CGI based. There's this. There's this one that takes place in, like, Greek mythology that's, like, notoriously terrible. There's one with the rock in Scorpion King where he comes out and, like, it's very clear that they just didn't give the 3D artists long enough to make it look good. And so it just, like, looks really awkward. Didn't age well. But then some CGI from, like, the original Star wars in 1979. You see the green screens and you're like, wow, that still looks amazing. It holds up. And so you got to know when. When to actually go in, dip your toe in. There's been another turn of events in the Warner Brothers takeover. Have you seen this on Kalshi? It's been going back and forth, neck and neck. Now Paramount is In the lead, 54% chance that Netflix takes over or that Paramount takes over Warner Brothers. Netflix is at 36%.

A (152:46)

And I believe the final offers need to be. The final offers need to be submitted by tonight or tomorrow night. I forget exactly, but Warner. Paramount's revised offer for Warner Brothers will likely come in at $32 per share. Let's pull up this video.

B (153:05)

Ted Sarandos.

A (153:06)

Ted Sarandos having a chat.

B (153:09)

Absolute goat. While we pull that up, let me tell you about Sentry. Sentry shows developers what's broken and helps them fix it fast. That's why 150,000 organizations use to keep their apps working. And let's go over to deadline.

E (153:21)

I, like you asked that question. We've been working hard on this transaction to acquire Warner Brothers and hbo, and we're deep in that deal every day. Yeah, absolutely. There's. There's no reason for not to right now. Our deal is the best deal. It was determined by the Warner Brothers board. It was reiterated to suggest it to their shareholders. We're going to vote on March 20, and that there's no rational reason to block the deal. It is, you know, we're 9% market share growing to 10. So there's really no concentration risk in our deal. And what's exciting I think is we'll be able to have this hundred year legacy of great storytelling finally in the

D (154:02)

hands

E (154:06)

and invest in it and grow it.

B (154:11)

The Trump language kind of comes. It works its way in because you're hanging out with your friends. You start doing some Trump impressions and then it just comes out. It just comes out. Sometimes it's just one of the greatest impressions ever. So you just gotta do it every once in a while. Wow. The year before, the year before.

E (154:27)

It's going to look like that next year and the year after and the

B (154:29)

year after be the greatest acquisition that the world has ever seen.

E (154:33)

Traditional 45 day windows, theatrical exclusivity.

B (154:38)

You should be like, didn't you read the Citrini piece? Everything's going to zero. Does it matter? Does it matter if two companies that are zero combined? No. Just let it happen.

E (154:48)

So we're excited to be in there. We want to help them win these lapel matches. And Mike who run Warner Brothers, they both opened nine number one films in a row. That's amazing. That's the kind of track record. We're excited.

A (154:59)

All right, we can pause it.

B (155:00)

I love it.

A (155:00)

Over the weekend there is some new reporting from Bloomberg. The Justice Department's investigation of Netflix's proposed takeover. Warner Brothers includes a scrutiny of whether the streaming giants behavior wields anti competitive leverage over creators. He talked about it going from 9 to 10% market share.

B (155:18)

But streaming.

A (155:19)

But the issue is it's taking buyers from like a handful of buyers down to one is more real to be

B (155:25)

and I hadn't considered that.

A (155:26)

Well, and that's, that's, that's what we were talking about with Ashley Vance. Hey, you sell documentaries?

B (155:31)

Yeah.

A (155:32)

You excited to have one last buyer? Literally no one to play offers against each other. Just kind of here's the price. Take it or leave it.

B (155:40)

Yeah. Well Netflix for a while has had co CEOs. Maybe you can. Mommy, Daddy. The them go to one. Oh, the other one said he was going to buy it for 500 million. He's like he didn't say that. I was just texting with him. What do you want to say, Tyler?

C (155:51)

At some point the big labs are going to be buying these documentaries, right? If you have good enough, if you have very high quality training data, you can sell straight to OpenAI.

B (155:58)

Sure, sure. Yeah, that makes sense. Yeah. I mean, honestly, like a Sora deal wouldn't be out of the question for. For Warner Brothers. You know, I wonder if the Disney. The Disney Disney deal is exclusive in that they will not be on another AI generation app. But is it exclusive the other way in the sense that Sora will not add Superman or Batman and they will only have Spider man because it does feel like never the two shall meet. Like, we're not going to see Captain America and Superman fighting in anything other than a Chinese model that's getting a cease and desist. But in theory, Sora could go and do a deal with Warner Brothers in addition to Disney, but that might have been stipulated as like, no, we want to be the exclusive provider of superheroes. Effectively.

A (156:46)

We don't want anyone to outslop us.

B (156:49)

Yes. Let me tell you about Railway. Railway is the all in one intelligent cloud provider. Use your favorite agent to deploy web apps, servers, databases and more, while Railway automatically takes care of scaling, monitoring and security. So I cut you off.

A (157:02)

I think it's going to be a big moment when Disney ip.

B (157:05)

Oh, yes. But also when this concludes, I mean, it's going to be. It is neck and neck as we've seen in the Kalshi chart. And then also there's a lot at stake, like the breakup fees and the billions. I think, like, it's a political story. There's so many different things going on. Anyway, I think it's. We have our next guest. Well, while we bring him in, let me tell you about Cognition. They're the makers of Devin, the AI software engineer. Crush your backlog with your personal AI engineering team. Come on to the TV penultradome. Mike, good to meet you. How you doing? While he's sitting down, let me tell you about vibe co. We're D2C brands, B2B startups and AI companies advertise on streaming TV, pick channels, target audiences and measure sales.

A (157:52)

Just like, what's going on?

B (157:53)

How you doing?

D (157:54)

What's up? Gentlemen?

B (157:55)

What's up? Please introduce yourself. First time on the show?

D (157:58)

Sure, first time on the show. Happy to be here. Long time watcher. I like the hat. Thank you. We were actually talking earlier saying, how long have you been watching the show? And you're like, back to the hotel room. I was like, I can't remember the hotel room, but I remember the printed tweets.

B (158:09)

Oh, yeah, we have a lot of paper today. We always stack up a ton of papers on Monday because we get the weekend edition and the Monday edition and then we print some other stuff. We got to bring back the printed tweet for like the best tweet of the day. But honestly, the printer was like a major rate block, rate limiter for us because we'd be like, we're going live. And we used to just start the show around 11. We'd be like, we're 30 minutes late because we would just do RSS. Now that we're live at 11, it's like the printer's got to work. And we would print like hundreds of pages because we'd be printing whole articles. Anyway, sorry I interrupted your interview.

D (158:47)

No worries. So Mike Annunziata, founder and managing partner at Also Capital and we're early stage hard tech fund investing in inception. Pre Seed. Seed.

B (158:57)

How'd you get into vc?

D (158:58)

How did I get into vc? Had a bit of an interesting path, a little bit non traditional.

B (159:03)

So Stanford.

D (159:04)

Stanford, yeah, Stanford.

B (159:05)

Did you actually go to Stanford?

D (159:06)

No.

B (159:06)

Okay.

D (159:08)

I started at Harker and then. No, no, no. So I've been doing venture for a little bit more than a decade, but actually overnight success. Overnight success. So a little bit more than a decade. Started my went to Cornell for undergrad, which, you know.

B (159:25)

Still Ivy League.

D (159:29)

Exactly. Have you ever heard of it, though?

B (159:31)

Yeah, I have,

D (159:33)

yeah. Cornell undergrad. Did the family office thing for a few years. I actually worked at the Cornell Endowment. So I've been an lp. So I've been on that side of the table. And then business school and then started a. A hard tech company back in 2016.

B (159:46)

Okay, okay. So founder.

D (159:48)

Yeah, founder. Yeah. You know, same year Andreil started, we're worth a fraction less than 6 billion right now.

B (159:54)

But, you know, what were you doing?

D (159:55)

We were doing food technology development. So.

B (159:57)

Okay. Oh, cool.

D (159:58)

Around the same time you're doing Soylent John. So I'm sure we're doing food manufacturing technology kind of from scratch. Me and a co founder in a lab through series B. Company is still going. Built out a big facility. So I've been doing our tech manufacturing for quite a few, quite a bit.

A (160:13)

Give us the history of also then.

D (160:15)

History of also. So, you know, it's funny, Also started as Will Brewery, Mike Anneziata and Colin Smith's backyard angel investing adventure in 2019.

B (160:24)

So on Friday.

D (160:26)

Yeah, he was on the show on Friday. Saw that. So I did Dorman Fund when I was in business school. So I've been a DRF partner for a number of years now. An alum. Started writing angel checks in 2019, kind of scaled up through SPVs, and then raised our first fund in 2020.

A (160:40)

3.

D (160:40)

So that was a $22 million fund. And through that journey, you know we wrote the first check into Radiant Nuclear, you know Varda followed shortly after that. And I've been on the board of Varda since inception. Yeah, we did K2 at seed as well any signal and software radios. We raised our first fund in 23 that I mentioned and then wrote the first check into Northwood. So there's been a bunch of.

B (161:01)

Was that post zerp ending the crash?

D (161:05)

Yeah. So it was post zurp. You know, kind of hard time to raise a fund one as a new manager.

B (161:10)

But you're not saying I'm just going to go into crypto and like the fl. The frothy stuff. You're in the stuff. That's correct.

D (161:18)

Well you know I think you could. Let's just talk a lot more about this. But I think you know our thing from the beginning is, you know, who are your smartest friends and how do you believe in them before others do. And then I think the hardtick thing candidly grew outside of that to group from that.

A (161:30)

When how many checks had you written into Gundo companies or Gundo adjacent companies before John went. Oh yeah, put it on the map with that video.

F (161:42)

Yeah.

D (161:42)

I don't know, we did like six or seven before the Gundo bus and your thing.

A (161:46)

So the pre bus.

B (161:47)

And the bus is really the defining line.

A (161:51)

Exactly.

D (161:51)

You know VC free bus. Yes. I mean we did a bunch of Varda, radiant K2, any signal, that kind of crew of folks. And then since then did Northwood first check there. You guys had mesh on the show. We did that one recently as well.

A (162:08)

Basically every company.

D (162:11)

Yeah. So we've been very fortunate. They let the non engineer guy somehow cosplay as an engineer vc, which is a lot of fun sometimes but I'm not afraid to make myself look silly every once in a while to try to learn something new.

A (162:23)

So what is your non technical but so what is your process for underwriting some of these companies where at pre seed they can often actually seem like a science project and that's like the. I probably made that mistake once or twice across 60 some bets where I invest. You know, I'm not a professional investor but accidentally invest in a science project that was being positioned as like a commercial opportunity. But the ones you listed off started as kind of like far out ideas and now have very real commercial opportunities.

D (163:01)

Yeah, look, I think the unique lens that I kind of bring to this is the fact that I ran a company doing hard tech stuff for seven years across engineering, built the whole team, kind of know what a good engineer sounds like and how they execute, you know, product, go to market operations, all that vertically integrated. We for the most part are investing in people that have done these kinds of things before. And if you look by example, the Varda team, a lot of those guys are doing Dragon at SpaceX, right? They miniaturized it and turned it into Winnebago. If you look at the Mesh optical team, they're doing lasers. At SpaceX, they're doing lasers. Now if you look at any signal, right, doing radios, a lot of the Northwood team doing ground stations, right. They're doing the thing they were doing before, but with a different market opportunity. So, so the common thread is like these are serious people building serious companies. And that's. You can kind of see that once you've lived it for seven years is, you know, I didn't, you know, it's great that more people are coming in and wanting to be excited about investing in this category, putting more capital to work in the category. I think we need that. Yeah, exactly.

A (163:56)

I'm really excited about the broader Silicon Valley community coming in and marking me up five times, ten times.

D (164:03)

Jordy, I got an uncapped note for you if you're really excited about one the of these things.

B (164:08)

It was in the COVID of the Wall Street Journal business and finance section. Today, investors go heavy on AI immune assets. Explain to us what HALO is, what it stands for, what it means.

D (164:21)

Yes. So heavy assets, low obsolescence, which is a term that I just learned a week or so ago. So thank you. J.P. morgan. Are they a sponsor yet?

B (164:31)

Not yet.

D (164:32)

Do we want them to be?

B (164:33)

Depends on what the product is.

A (164:36)

We love J.P. morgan.

D (164:37)

We love JPMorgan. Yeah, I think this heavy assets, low obsolescence is these things that are very difficult to replicate. You build a chemical plant or you build an aerospace production capacity. These kinds of things are much more resilient than your traditional B2B SaaS product that may be able to be replicated by Claude code, for example, or openclaw, something like that.

B (164:58)

This was Delian's bit when we had him on for the slop versus steel debate. Randall debating of, of, you know, the paper, what would be. Yeah, it was a pay per view. What would be most resilient? But unpack that a little bit because heavy assets, high assets, what does that actually mean in the defense tech context? Because there still is a lot of R and D that's happening and then low obsolescence lessons. I want to Know more about like the curve of like what can be obsolesced. Because there are some companies, when I think about like, you know, a lot of people have been saying like oh like buy raw materials like go long gold. And it's like that is fungible. So it's, there's no real moat there. If you just own some gold, you just, it's commodity. So how do you think about non commodity products?

D (165:49)

So, so, so two things. First look, any great venture business at the core is a company that has the potential to generate high return on capital in the long term. Like you have to account for the assets that it takes to generate the revenue. And then how durable is the revenue over the long term. That can come from Dow Chemical which has just such massive scale that it's very difficult to replicate that scale. And they have a very durable business, low obsolescence. You're not going to remake all those chemical plants. Or it could come from something like a radiant nuclear who is building very complex nuclear reactors in a shipping container equivalent. And they're going to scale up not just the design of that but the manufacturing, the supply chain, the regulatory. And that gets to the second piece that we always talk about which is we love companies that are novel in the aggregate. Right. We're not looking to your point on how do you underwrite a science or engineering. Jordy. It's like it's not one specific thing we're trying to understand. It's what are the 32 things that have to come together? Why is this the right model moment right now for this thing to happen? And if it does now all of a sudden we've got a moat. Because it's not impossible for somebody else to get a re entry capsule like a Varda.

B (166:53)

Yeah, right.

D (166:53)

It's not impossible for somebody to build ground stations like a Northwood. But the dynamism of these companies is in the ability to execute and move with speed and integrate these complex systems that become low absolutions. But they are heavy assets.

B (167:05)

Yeah, it's like the factory is the product. I've always heard that thrown around as like. And I always read it as like the real challenge is manufacturing at scale. But I think there's another cut on the factory is the product which is probably something like you actually would be very reticent to invest in a hard tech company like Varda if they were like, yeah, we have a third party manufacturer that produces the capsules. We just send them a CAD file.

D (167:31)

Yeah, I think it really depends. Right. I love the I Love the theory behind the factories, the product. But I also think there's some pretty strong, you know, strategy that that needs to be built on. You know, why should we build a factory? And if you look, SpaceX is a good canonical example of this. Why should they build a factory? Well, it was because they were creating a lot of the demand at the same time as they were trying to deliver it at a given unit price to unlock a bigger economic opportunity. And there was no third party even available.

B (167:58)

Yep.

D (167:58)

That they. So they had to do it themselves so they couldn't meet a spec, so they had to do a it themselves. And I think that's the different dynamic where you should not. And I post about this a lot because it like definitely is a thing I believe pretty strongly you should not vertically integrate just for vertical integration sake.

B (168:12)

Yeah, no carbon.

D (168:13)

It's a poor use of capital to take a bunch of equity and shove it into a commodity machine which you could not do, for example. But if you look at like a Varda for example, cadence is everything for that business. So if cadence is everything, you got to be able to turn fast. So they have a lot of capabilities. They built in house to be able to do that same kind of thing with the Northwood.

E (168:32)

Right.

D (168:33)

In house as much as they possibly can. Because they have to turn fast to move with speed. Speed's really the advantage, but it is built on a foundation of the ability to deliver at rate.

B (168:41)

Yeah.

D (168:42)

And that's kind of the core. When you say factory is the product, I think I agree with that. But where are you pointing that advantage? Where are you pointing that? Because it is an investment that you're making as a company and it's something that even in day zero investment, like all of our stuff is pretty much inception pre seed. We're thinking like if you're going to build a factory in three years, you need to be thinking about it now.

B (168:58)

Yeah.

D (168:59)

So why are you building that factory? Why are you doing that? Like what competitive advantages give you? And I think that's how you know, the people that we backed have really had a good sound like head on their shoulders of how to think about those trade offs. My build decisions.

A (169:11)

How did you react to the general intelligence crisis of 2028? The viral essay that new to the markets. I saw this guy, John Lober had a pretty kind of a response to it breaking it down. He was in some parts saying like institutions have a lot of momentum. They can carry that momentum and adapt to these new market forces. He also suggested that reindustrialization could. It seems obvious that if a lot of our jobs can be just automatic, automated, with a computer, maybe they weren't that real in the first place. But there's a lot of work that needs to be done in the physical world. There's. America needs to figure out how to make stuff again, not just kind of push paper around. And so are you optimistic that this could shift talent from, you know, making the 50th, like vertical CRM to making stuff? Do you think this could be.

D (170:13)

Look, like I. I'll say this. I don't think we need to send our boys back to the coal mines.

A (170:20)

The children yearn for the mines.

D (170:24)

The children. No, just kidding. You know, my kids, if you're watching, which I think you are.

B (170:28)

Yeah.

A (170:28)

Get off mine.

D (170:29)

Get off my.

B (170:30)

It's light blue collar.

D (170:31)

Yeah, it's. Well, I think it's, you know, what are we doing next? Right. So if you go back to horse and carriage, right, it's, hey, Ford has a job for you on the production line. Like, go learn how to do that. And I think the opportunity is in the companies that develop recurring education rework as processes or capabilities as part of their normal operating cadence. Right. So a lot of our best companies, they will get seed funded and they will immediately start an internship program. And that internship program will become a funnel for new talent to come in after. After they graduate, for example. And I think institutionalizing that around these companies that are moving fast in new areas, whether it's lasers or nuclear or space. Right. I think we could bring back that retraining. But inside the corporate entity, as it's the responsibility of the corporation, it's really our company's problem that there's a talent shortage. We should take that off.

B (171:23)

He had someone who was working basically a blue collar job and just wanted to get like a. A forklift certification. And he brought them on. And then pretty quickly he was like writing cam, CAD automation software, computer aided manufacturing, and sort of became a CAM programmer, like almost. And that was like a lot longer path a long time ago. And so the upscaling thing is definitely real at these companies.

D (171:49)

Look, I think that the. We could get into like this whole different thing around, like the student debt crisis is definitely a big bottleneck to people being able to do this upskilling.

B (171:57)

Oh, sure.

D (171:58)

Where they may be constrained financially that they just have to kind of keep working to be able to make their student debt payments, for example. So I think that's a big thing that people.

A (172:06)

Yeah. I was having a conversation with Tyler. We've had a number of interns, all of which have been always paid. But I was thinking if I didn't have the opportunity to just work for free for people in college where at the time I was like working a job at a hotel, like grinding. But then in my free time I didn't have a lot of skills. And so I would just tell entrepreneurs like, hey, let me just pick up little things to do and ultimately learned a number of things that allowed me to start my first business and all that stuff. And going back, it feels like the, the, the sort of like unpaid intern is like completely like no go now. Like it's a net, it's a negative signal if a company is doing it, but it might have to come back and it actually, if people really want it, it's like, yeah, you can work a job that is very low skill and then spend all the rest of your time like reskilling yourself. I don't know. I think it might have to come back. Although I'm sure that there's a million reasons why somebody would hate that idea.

D (173:14)

Yeah, look, I think it's just really important. So. One of the things that's most inspiring to me about this moment in time is it feels like not only are there people spinning off of like the SpaceXers of the world that are now continue to take on really big challenges, they're doing two things. They're doing things that are really meaningful, that when you get at the root of them and there's an authentic drive behind a lot of what they're doing, but they're also willing to tell the stories and inspire other people to be better. If you look at Jared Isaacman coming in as NASA administrator, like he is an inspiring human being, right? That I think we, we need more people like a Jared who's 30, 40 years old, whatever it is, like young

B (173:49)

person, he's been to space, he's been

A (173:51)

to space process payments. Every time I get a shift, shift payment checkout, I just go, yes.

B (174:02)

Talk about debt. What is a reasonable debt load for a series A hard tech company? In software venture debt is seen as cancer. You do not want to go near it with a 10 foot pole. Paul Graham said, it blows up companies left and right. It's very scary. I think a lot of entrepreneurs, over time they get comfortable, they learn how to use it effectively. But then it's always just like, well, I have this money sloshing around anyway. It's all one big pool. It's not like the dollars get allocated one place or Another how are you talking to founders in your portfolio about debt?

D (174:39)

Yeah, like I could get hyper technical about this, but I won't for this audience. No, get technical, get technical, get hyper technical. The right amount of debt at Series A is zero.

B (174:51)

Okay.

D (174:54)

The right way to use debt in companies is in two places.

B (174:58)

Says the guy who sells equity.

D (175:00)

Says the guy who sells equity.

B (175:02)

No one should ever just come to me for another check.

D (175:06)

I'm gonna pop this one at the right time, in the right place. It makes a lot of sense. There's really.

B (175:10)

When does it make sense? When does it start? You could take me out to Series D. I don't care. But where for sure? Because there are a lot of companies that we've talked to where it's hard tech and they wind up buying a ton of machines in a big warehouse. And it just feels like the liability side of the. The business is very different than a couple programmers in a bedroom or in a garage. Right.

D (175:28)

Yeah. So you know, in a past life when I mentioned I worked at the Cornell Endowment, I also got my CFA while I was there and then went and got my mba. So probably. Yeah. Such a non traditional.

B (175:38)

How did you wind up in finance with just a CFA and an mba?

D (175:41)

Yeah. But you know, Revenge of the balance sheets is kind of what I how I think about what we're investing in now. But look, I think you're using debt for a couple situations. Number one, to buy long lived assets.

B (175:53)

Yep.

D (175:54)

But the caveat is they need to be tied to actual contracts that will be pushing out real revenue and cash flow.

B (176:02)

Yeah.

D (176:02)

If you are buying, if you are using debt to speculatively invest in capacity, you are taking a lot of existential risk into the business until you have contracted revenue that has significant bookings and backlog. So if you've got two years of backlog and you want to take on an interest rate of 7% while your equity is 30% and you think you can get to a revenue milestone that blows out your equity multiple, that's actually a really good use. Or if you have. There's a canonical example I always think of when I was working at the endowment, we looked at a venture debt fund and it was super interesting. Like 2015. This was 2015 and they had a very good use case of using venture debt. And it was. They were lending to Pandora and so for Pandora. Cheers to Pandora. So they were streaming service, not the pickup place. Yeah. Yes, exactly. Not the jeweler.

B (176:52)

The jeweler.

D (176:53)

The streaming service they were lending to Pandora. And Pandora basically had like a Cash flow timing issue where they needed to buy servers to be able to store the music and stream it. And there would be revenue that would come off of the stream. So the revenue would lag the cash investments in the servers. But as they were growing streams, their multiple on the revenue was growing.

B (177:16)

Oh, interesting.

D (177:16)

So you actually could use the venture debt to bridge to a higher revenue multiple when you went to raise more equity and reduce the cost of equity. So even if you were paying 18% implied cost of equity IRR on the venture debt, including the warrants, it was still cheaper because the cost of equity was declining as you hit milestones. Those episodes, episodic use where you knew you could raise equity as you hit a specific milestone to pay it down or it was part of the permanent capital structure that was tied specifically to cash flow. The trap is when you use debt to try to extend Runway as a.

B (177:48)

Correct. Yeah. So even in that Pandora example, like maybe it's not an asset backed loan, but it's like almost asset backed because you are just taking the money and buying assets that probably have some resale values.

D (178:00)

And you know the revenue is going to come because you know stream are going to go up and then you'll be able to refinance it because equity investors say, you know, you're at 10 million streams, I need you to be at 15 million streams. And like how do you get there? Well, at the time, like this was one way you could get there and buy the physical asset. So I think understanding where did the slope change of your multiple, like those inflection points or permanent capital.

B (178:21)

That's very cool. Jordan. Anything else?

A (178:24)

No. What founders and what categories do you want to meet?

D (178:28)

What founders and what categories?

A (178:29)

Like what kinds of founders? Not specific.

D (178:33)

What kinds of founders? Good question. Look, our big thing is founders that have fun, have fun, but they play to win. We talk about that a lot. Like have fun, play to win. There's like plenty of research. If you're having fun, that's when you do your best work.

B (178:48)

You see that with the Varda team all the time.

D (178:49)

See that with the Varda team all the time.

B (178:51)

Winnebago is funny. Clearly there a good time. We did the LK99 thing and totally, totally Midnight side project. Very fun.

D (178:58)

Totally.

A (178:59)

Yeah.

D (178:59)

It's like are you building these little cultural things that people have a good time? I think that's really, really important. But also the playing to win side too because I think you can over index on let's have too much fun and realize like hey, you're competing out here. So that's really what we're looking for is you know it. You guys have a ton of fun here, right? Like I think this cultural vibe is definitely what we look for in founders is are they having fun. It's how you kind of get to the hard stuff.

B (179:22)

So let's put the hard tech to work. Tell us about why don't you hit the gong. Yeah, you hit the gong.

D (179:26)

Me to hit the gong. All right.

B (179:27)

And tell us the announcement at 50 million.

D (179:29)

All right. Our second fund. 50 million.

B (179:33)

Go smash that gong. Thank you so much for coming on down to the TBVN Ultra Dome. This is great. Have a good rest of your day and we will talk to you soon. And lastly but not least I will tell you about phantom cash fund your wallet without exchanges or middlemen and spend with the phantom card.

A (180:00)

Mike really backed at least 50% of the breakout hard tech companies of the last five years at the earliest stage. Really wild.

B (180:11)

I like this this deep dive from Teo Berga who says reading a 2005 Paul Graham essay. Gasp. He used the forbidden sentence structure and it says writing doesn't just communicate ideas, it generates them in 2005.

A (180:30)

I think that it seems obvious at this point that PG's blog and Sam's blog both like deep in the training. Yeah core. Core to the to the Anyways we got a bunch of timeline here. Let's. Let's rip. Let's rip through it.

B (180:50)

Take me through it.

A (180:52)

Chairlift Capital says opening eye killing application software to then try to rebuild application software is the funniest timeline. Still opportunities in the enterprise. We got to have somebody on from McKinsey or one of these big consulting firms to hear how they're positioning it to customers make us a deck. See a deck Moving on. Where are we? Jim Cramer underrated poster he this morning at 3:33am was outside I see cold early I am not worried. Anthropic has a solution. I read it in a report what vague post incredible. I mean yeah sent IBM down 11% going to work Dolly Bali says investors simultaneously think AI is over invested and AI will take over industries. Yep it's a funny moment right now

B (181:52)

this last one SaaS is dead. OpenClaw replaced all my subscriptions. I went from $480 a month on tools to $1245 a month on API costs and 15 hours a week fixing YAML files adapter be left behind.

A (182:09)

Hope's revenge says woke up to Claudebot running its claw through my wife's Hair. I had explicitly asked it not to do that.

B (182:17)

Stop, stop, stop. There was a post about some meta researcher. Alignment researcher. Got all of her emails deleted by OpenClaw or something.

A (182:29)

Yeah, that was.

B (182:30)

That feels like a crazy thing to.

A (182:32)

I think she needed to stress that. My sense is that she set up a new device with a new email and was not just like running openclaw on her main meta email, but I think people read it that she was just running it on her main account, which seems crazy.

B (182:53)

People were saying like, oh, it's bearish. She's literally in alignment researcher. It's like that actually might make her better at her job, though.

C (183:00)

Well, she was at scale too, before. She wasn't like in, you know, in the less wrong, like, grinding through, like, EA stuff.

B (183:06)

Yeah. Like an EA would never be caught getting their emails deleted. No way. Just all of less wrong gets deleted by open call by accident.

A (183:14)

Palantir is launching Valley Forge grants, $10,000 awards for high schoolers to solve the problem that most inspires them using our software Ditch the coffee run and fake intern projects. Whoa. Shots, shots. Fire at Tyler Claude with ads was incredibly real. We'll pay you up front to. We'll pay you to confront the challenge you care most about. At Valley Forge, Washington soldiers endured a brutal winner, turned the tide, and won the revolution. Their fortitude has carried America to its 250th birthday. We're looking for pioneers to lead America to its 500th. Like the sound of that. Very, very cool opportunity. Based on our audience data, we have very, very few high school students in the audience, but you're listening to this. Go apply. Feel free to DM us if you do apply. And we will try to nudge.

B (184:09)

What happened with this F1 race. Someone smashed. Someone crashed.

A (184:12)

Yeah. Who in the chat was actually at the demo because San Francisco. San Francisco. I've never seen San Francisco go. This is in San Francisco. Let's pull up this video first of the truck actually jumping over one of the hills. John, I want you to see this.

B (184:28)

Okay, let's play this.

F (184:31)

Wow.

A (184:32)

This is San Francisco.

B (184:33)

This is in San Francisco.

A (184:34)

I still got it.

B (184:35)

This is the best part. In San Francisco, they close down roads for crazy stuff all the time. There's beta breakers. Halloween is a whole scene. There's a million other events that happen throughout.

A (184:45)

Honestly, I used to do something like that with my clutch when I'd be

B (184:48)

driving my manual jumping over an F1 car. Wow.

A (184:51)

I used to amazing driving manual as a high schooler in San Francisco. Made me Into a man. Yeah, you're really facing death on every hill. And then let's pull up this. One of the cars did a burnout. People were thinking this was Yuki, so. But apparently it was one of the other drivers.

B (185:09)

This is the Red Bull. Wait, why is it on fire?

A (185:12)

At one point, it caught on fire.

B (185:13)

I thought it just, like, had some minor bumper damage.

A (185:16)

No, this was a wild. This was a wild demo. I think. I think they really pushed the limit. Red Bull really wanted to make a statement.

B (185:22)

They're spinning around here, okay. And then they go forward and smash into the wall. Like, that feels like the easiest crash to prevent. When I was watching this, I was like, while you're spinning around, that's the point where you're going to go crazy. But then it's just like, just smoothly going forward and smashing. Front wing.

A (185:39)

No brakes.

B (185:39)

Decimated. No brakes.

A (185:40)

I guess no brakes. And then at another point, the car caught on fire.

B (185:45)

10 out of 10.

A (185:46)

10 out of 10. Demo.

B (185:47)

Red Bull knows how to entertain. They know how to entertain. Like. Yeah, I see. This was on purpose, actually, in the Apple Vision Pro update, I watched a Red Bull video, an immersive video. It was a skier video, about 15 minutes, and they take you back country skiing, heli skiing with these skiers, and they have the Apple Vision Pro thing. And it is amazing. It's like, one of the greatest experiences because realistically, I'm never actually gonna go heli skiing backcountry alone. But this, like, just you. You feel the scale and everything, it's so good.

A (186:23)

I. I called you at like, 10 or 10:30 on Friday. Yeah, no, I didn't even call you. I just said, you up, question mark. And John sends me back a picture of himself in the Apple Vision Pro. You know, you were probably the only active user at that point.

B (186:39)

Well, you know what? I got a Mac Mini hooked up to my Apple Vision Pro now because I got a capture card that can take in an HDMI input. So I hooked the PS5 up to the Mac Mini and then mirror that to the Apple Vision Pro so I can play PS5 and Apple Vision Pro with way too much latency. And it actually doesn't work well at all. But it was a fun experiment and an obvious feature that they should have launched two years ago, but they didn't figure out how to do it. Just put the HDMI cable on the actual power brick and just let you plug in anything, and you'd have a massive audience of people that just want to play all sorts of stuff that's HDMI compatible. Anyway. Very, very.

A (187:24)

If you are in the market for a western, sort of vintage Gulf Stream, Greg's got you covered. X has got you covered. It's a 1994 Gulfstream GI VSP. The latest ask was only 3.75 million. So.

B (187:43)

This is beautiful. The carpet.

A (187:45)

We don't know how to make jets like this anymore.

B (187:47)

Everything about this is amazing. It's so. It's so good. So opinionated.

A (187:53)

Look at the. Look at the bathroom.

B (187:54)

I love it.

A (187:55)

The bathroom. They went.

B (187:57)

The bathroom is crazy. Steampunk. Yeah.

A (187:59)

Yeah. Star wars mode.

B (188:01)

Cantina. Yeah. Should we watch this latest video from Sea Dance?

A (188:06)

Let's do.

B (188:08)

Starts by saying Hollywood is cooked. Hollywood is cooked. Based on the new Sea Dance AI video model. It's showing Transformers, but let's actually track what's going on here. Starts out as a. As a jet, a combat aircraft. He gets out of the jet, turns in. Around. Now it's Transformer. Okay. Then he gets in another cockpit. Now it's a helicopter, and it has a gun on it. Okay. He's shooting it. That's useful. You needed to be in helicopter mode, but now you got to get back in the transformer again. So he gets back in the transformer. Oh. Turns out the transformer can run like a human and walk, but it turns back into a plane. Back into a plane. Then what do you want to do when you're a plane? You want to land on the freeway, on the highway, on the street. You land, you get back out. Then you get back into your Transformer to get in the front cockpit, and now you're back in human mode, humanoid mode, and then you blast off.

A (189:03)

You would not be criticizing this.

B (189:05)

You turn into a plane and you fly backwards.

A (189:08)

John, if this was an actual scene in Transformers, you would just be watching it being like, that's tight.

B (189:15)

Okay. So I agree. I would be saying that's tight. And it is incredible visual fidelity. Incredible visual fidelity. And just an amazing video and entertaining to watch. And that's why 4 million people enjoyed it. And it only has one community. One community note in the community. Note is just AI slop engagement farm. Yeah. Obviously, this is not cooking Hollywood today. It's a tool. It's cool. But really impressive considering that this truly is the most expensive shot you can do in Hollywood. Like, it is so, so complicated to animate all of those different rigid bodies as they interact with each other or they don't, and they tuck inside. It's so difficult. It's the Mount Everest of motion graphics and CGI Tyler could do it if

A (190:05)

we get 10 minutes, but I see your point.

B (190:08)

And the final step in any of these, you can actually go and animate the rigid bodies in cinema 4D or Houdini or something, rig all this up. But the final texturing, the color grading, blending everything in, that's another major step. And it just nails all of this. The lens flares, the. The reflections on the glass, all of that's like another step because you create the jet and then it just looks like a 3D render of a jet. So you have to blend it in. You have to make sure the colors match, make sure it matches the background. This stuff is so time consuming. And having a tool that at least allows you to do some previs, animate, interpolate, do a bunch of different things, obviously would require a lot more art direction to get that to a place where it's. It's amazing. And it makes sense because a lot of the CGI that happens in Transformers. I know some of the Transformers clips are silly, but a lot of it's very motivated. It shows you how things move. And there's a decision driving one transformation to another. It's not just randomly switching from a plane to a car to a plane and back and forth. And it's usually meant for dramatic weight. And there's some weight and timing, and there's art direction that sits on top of the actual cgi.

A (191:21)

Anyway, apparently. Openclaw Field Ordering Frenzy creates Apple Mac shortage. Delivery for high unified memory units now ranges from six days to six weeks. I called it. I did call it.

B (191:36)

You did?

A (191:36)

We did the math a couple weeks ago, and it seemed obvious that if the frenzy kept up, there would eventually be some shortages.

B (191:45)

I mean, demand for AI is continuing unabated. People are using this stuff. Martin Shkreli shared the Nvidia demand check on Lambda. Lots of things are out of capacity right now. People are using stuff. If you need to hop on Lambda, hit us up and we'll introduce you.

A (192:02)

But people were pretty triggered by Gary Tan and the YC crew jumping on a podcast. Lobsters. And I was triggered by people's reaction to it because we have done similar gesture maxing many, many times. I think they were just having. They were just having a little fun.

B (192:23)

Let them have fun.

A (192:24)

But. But apparently fun. Fun is illegal.

B (192:29)

Fun is illegal.

A (192:31)

Hunter Weiss, you have to pack it up.

B (192:32)

Wait, we got it. We gotta talk about our secret plan. Oh, yeah, we're leaving. We're leaving California. Not because of any taxes. Those don't apply to us. But because there is an entire main village with a church and multiple homes that's on the market for $6 million. And we're gonna move everyone there. I'm hearing some claps. I think people are in, they're down.

A (192:54)

We're gonna build that up.

B (192:55)

The whole crew. We're gonna build that 40 acre village. And it was first listed for 5.5 million. You get the whole town.

A (193:02)

What actually qualifies as a village though?

B (193:05)

Well, it has 21 structures. 21 structures. 21 structures. And this is where it gets funny. So we were saying let's move the whole team there. There's 21 structures. We're a small team. We got like 10 people. That's enough. There's not 21 houses, Jordy. There's 21 structures. So it's entirely likely that Tyler over there will have to live in a barn or shed or perhaps the church

A (193:26)

fitting in the stable.

B (193:28)

But other properties found on the unique compound include a Greek revival styre, dwarf style dwelling, antique barns and multi bay garages. Would you go if you had just had to stay in a multi bay garage, you can sleep in a GT3RS. You can also sleep in a multi bay garage.

C (193:43)

If we get some cloned horses, cloned

B (193:46)

ponies, let's do it. Everything a homeowner needs to build their own thriving community or set up one of a kind rental venue. Very fun option. If you have a need for 21 structures, head over to Main and pick this up.

A (194:00)

Hunter Weiss shares one of the best product ads ever.

B (194:02)

I agree.

A (194:03)

The I run the ipod shuffle. We don't know how to make ads like this anymore.

B (194:07)

Really good.

A (194:09)

Really, really good.

B (194:09)

Tells you exactly what it's going to do. Ride around the park.

A (194:12)

Really, really, really good.

B (194:13)

Beautiful.

A (194:14)

Bring it back. Will Depew says whoever builds Gmail app search should be burned at the stake. Every time I use this app I want a kms 75 likes Proof of insurance and it's just pulling up a bunch of Delta Airlines receipts.

B (194:33)

This is actually extremely annoying. I don't know how the search got so fuzzy, but you can search for exactly the term and it'll just be like there's this one cookie that's buried in white text that sort of matches it and it just shows it to you. They got to do something here. I think it's a big app so there's probably some overhead to like fully rewriting this. But search is tough. Search is hard. Even in the AI apps, I find that they generate so much text now that if I search for One keyword and I have it in my mind. It's like, well, that word came up the last 50 chats. Like it comes up all the time. And so it's been very hard, but I do think it'll get better. But certainly an opportunity.

A (195:17)

Dylan Field was having a little fun. He says nothing to see here. Sometimes the chairwoman of the Task Force on the Declassification of Federal Secrets just likes posting pretty pictures. Please continue talking about the Olympics and Anna Luna sharing an image of what looks like a wormhole. And just vague posting now. We got Congress. Congresswoman Vague posting now. I love it. I love it. Let's keep it up.

B (195:47)

You got a vague post every once in a while. Bone Vague post to go through.

A (195:51)

Last one says my culture is not your costume. Brian Johnson. Because Brian said he decided to live life on Friday. He was spotted just playing some video games, having some Taco Bell, some pizza, some Dr. Pepper, having a lot of fun, looking not his usual self, but locked in on the big game or something like that.

B (196:15)

Plant the bomb. I have some words of inspiration for the listeners.

A (196:20)

Wait, no. We have to say this last one.

B (196:22)

That's my words of inspiration. After you plant the bomb.

E (196:24)

No.

A (196:24)

And then I actually have a few more last ones.

B (196:26)

Okay, read this one off.

A (196:29)

Let's see here.

B (196:30)

Excuse me. Neil Renick says everyone you meet is. Is fighting a battle you know nothing about. Send them a teams meeting link and finish them off. I love it.

A (196:42)

Last thing we'll cover today. Anthropic Posted earlier. We've identified industrial scale distillation attacks on our models by Deepseek, Moonshot and Minimax.

B (196:51)

Wow.

A (196:52)

These Labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to to train and improve their own models. People were having a lot of fun with this. They said no crying in the copyright casino in all caps. Or Daniel Luke brought up a vintage growing Daniel Post. Aw. Did someone take your hard work and use it to train a model to mimic your expertise without compensation?

B (197:20)

Oh, pot cuddling, little black. Potentially.

A (197:24)

Another person says, I can't believe someone would just steal from Anthropic like this. The millions of man hours anthropic spent, handwriting, code, text, art books, et cetera to generate enough data for training must be taken into consideration here. Where is the respect for IP?

B (197:39)

Wild fun times.

A (197:42)

And here's where we'll end. Neat says reading one LinkedIn post is equivalent to unreading five books. And that is a good time to remind you to leave us five stars.

B (197:55)

Yeah. Follow our LinkedIn. We're on LinkedIn. We're posting regularly now. We got a bunch of fun stuff. Little clips from the show rewritten, little takeaways, little things that are on our mind. We really would appreciate a follow over on LinkedIn computer. Leave us five stars.

A (198:09)

Make sure everyone in the audience has the best evening of their life.

B (198:13)

And sign up for the TVPN newsletter@tvpn.com goodbye. Nice work, brothers. I'll see you on the next one.