SpaceX Launch: Deep Dive & Reactions | Josh Reeves, Keller Cliffton, Will Brown, Julia Steinberg, Olivia Moore, Flo Crivello

25:44But the Gundos been like a great home for defense tech broadly. Maybe Redmond's the next one. Maybe you got to do some, if you're doing satellite R &D, got to get a Redmond office. If you're doing launch stuff, you want to head out to South Texas because they do seem pretty friendly out there. At the end of the SpaceX stream, the announcer kind of like thanked all the different parties. He was like, thank you to the county where Starbase is because Starbase is like this chartered city within the county. and thank you to the people of Texas and the people of this and that. Well, they got a stock exchange now.

26:17They do. It was yesterday. Got Nicey Texas. Nicey Texas. Something down, something happening in the Lone Star State. Yeah, I haven't been to Texas in years. Last time I actually went was for David Sinner's thing. Really? Yeah. Other than that, I don't find a lot of random time to go there. It's been a lot of New York, a lot of SF, a little Miami. I have been a bunch. When was the last time you went? last year I think I went three or four times for different portfolio companies great place do we have the video of the SpaceX explosions over time they have this montage reel I'd love to play that if you can find that as I tell you about graphite.dev code review for the age of AI graphite helps teams on github ship higher quality software faster I'm throwing it in the timeline This is the best video that I could find.

27:13We will be talking about venture debt and what's going on in that market. But in the meantime, I believe we have a video of SpaceX's launches and crashes. This is from SpaceX themselves, how not to land an orbital rocket. Yes, this is the video I want to watch. Can we go full screen on this? This is great. and just immense confidence to release something like this that on the face of it makes the company look bad but actually demonstrates what makes the company look great, which is not being afraid of failure. The top comment is, this is 100 % the most expensive YouTube video ever made. It just comes down so fast.

27:52It comes down so fast. The first one is just like, you didn't stop at all. Here we go. Uh-oh. Oh, no. Wow, there we go. Beautiful. But this is great. Unfortunately. Look at all this equipment. The smoke's still coming off. Yeah, how is that not just going to blow? That seems dangerous to be around there. Yeah, that's a lot of confidence in your product for Elon. Especially for something that, like, it actually blows up. Yeah, here it is on the drone ship, I believe. Just smashing down constantly. This one trying so hard lands, and then it's definitely gonna blow. There we go. You know it's gonna blow up.

28:36Boom. Every time. The music. Yeah, the music's great. I thought they had a different one with Hall of the Mountain King. Wait, what is coming off of that? Why didn't they use the chopsticks? Yeah, Tyler, do you know why they didn't go for chopsticks on this one? Like originally? Yeah, yesterday. Can you look up why they didn't go for chopsticks grabbing? Because it seemed like they were able to do that a few flights ago. I'd be curious to know why they didn't do it. This is brilliant marketing. Yeah, it's the yes and this is so good The yes and approach. Yes, we blow up Why is it tilting it's not even going is it just a ship?

29:14Oh, it's a sea ship and it's just about to fall This is crazy. Oh, it actually made it. It seems like this one's coming down That was a tough one solid Okay, solid You know it's gonna explode this is insane

29:34so good oh I could watch these all day

29:42not bad anyway on to venture debt yeah let's get into the the really you know what's really burning up the timeline what is the audience is clearly celebrating Gulfstream Aerospace says, join us in officially welcoming the world's longest range aircraft to the skies. The first 8 ,200 nautical mile G800 has officially been delivered to a customer at our Appleton, Wisconsin, completions facility. Look at that plane. So let's give it up for Gulfstream Aerospace. They don't have the luxury of having quite the number of explosions. They said to join us in welcoming the world's longest range aircraft to the skies.

30:32We are happy to join Gulfstream in welcoming the world's longest range aircraft to the skies. Anyway, let's go to venture debt. There are some changes in the market. Yeah, the journal has a story. The asset class has shifted to late stage lending as companies remain private for longer. Venture debt firms are shifting their focus to larger deals with mature private companies as more startups stay private for longer. venture debt value hit a record 53 billion in 2024, up from 27 billion the prior year, and more than a previous high of 42 billion in 2021. Of course, SVB would have been a big player.

31:06So it's actually still, you know, given how significant SVB was in that market, it's pretty incredible to see the growth. As the value of deals rose, however, venture debt deal count fell to 1 ,300 transactions in 2024, the lowest level since 2016. The trend continued during the first half of 2025 with a little more than$19 billion of venture debt deal volume, according to a separate report by Capital Advisors Group. Great name, very generic, but if you can own it, it's powerful. Debt must react to equity, said Stefan Spazik, director of debt placement at Capital Advisors Group. Venture debt providers have to target stable companies with solid revenue if there is a lack of equity or sponsors decide to pull back on equity commitments to their portfolio companies, he added.

31:58I didn't fully understand this because it feels like there is very much not a lack of equity right now. But what is he? Well, I mean, he's saying that's the big risk, right? If a company is running in the red and they raise their next round, that's when you can get. Yeah, yeah, yeah, yeah. If the current backers aren't doubling down at Series C, D, E, F, which is the vast majority of companies. We focus on the companies that successfully raise every single round. Anyway. Startups and growth companies are turning to venture debt as an alternative to raising what is often more expensive equity capital that stands to dilute existing shareholders in an environment where traditional exit opportunities remain heavily constrained.

32:41Venture Debt Provider Say credit also allows venture investors benefits, such as retention of company board seats, which would have to be given away in equity financing. The, quote, liquidity constraints on the part of venture equity are still a big driver in the market, said David Sprung, the founder and chief executive of Runway Growth Capital, adding that he expects the venture industry's liquidity constraints to continue into next year. Runway typically writes checks of roughly$30 million, and it lends to tech, health care, and consumer. Earlier this year, for example, data and AI company Databricks announced a$5.2 billion.

33:17You never know with Databricks. They might be raising$52 billion. Announced a$5.2 billion credit facility shortly after the final close. We want to be 50 % debt to equity on this company, on this high-growth tech company. After Databricks. It's like a$150 billion company now or something. It's over$100. It's a huge... They just used the greater than somewhere. Yeah, that's right. That's right, yeah. Which is like, who's counting it? The last deal was a$10 billion Series J. Well, now they did the Series K. They did the K? They did the K. The company said it plans to use the debt for general corporate purposes.

33:53The debt financing was provided by a consortium of investment banks, including JPMorgan Chase, Morgan Stanley, and Goldman Sachs. Let's give it up. And Blue Owl's in, and Apollo Global's in, and Blackstone's in. Just a couple names. They really got the murders. Technology companies are looking to diversify their capital sources and debt is increasingly attractive as a non-dilutive way to fund growth and lower the overall cost of capital. As companies scale and stay private longer, we see significant opportunities to support high-quality businesses with flexible credit solutions. Earlier this month, on the smaller side, Runway committed$20 million to Swing Education, an online marketplace that connects schools with substitute teachers, which is backed by firms including Apex Partners and Reach Capital.

34:35Runway's commitment consisted of a first-lean term loan and a revolving credit line. And what's interesting is like we hear a lot about the venture debt boom in the context of like the Hadrians of the world or like, you know, hard tech industrials. Like makes a lot of sense to have relationships with debt providers because you're buying a lot of land or you're buying a lot of equipment. And like that has durable value. Yeah. Yeah. Even if you're buying, you know, when you when you read about Crusoe building a massive data center, it's like, well, of course, there's going to be some credit involved.

35:05because a piece of that business looks like a traditional mortgage and it looks like a real estate deal almost. And so you want a certain tranche of capital to be focused on earning real estate-like returns as opposed to paying for everything with venture capital. So since the 2023 collapse of SVB, one of the most active venture debt providers, more lenders have entered the market or stepped up their existing lending, creating a more competitive landscape. Borrowers not only have more appetite for debt, but they now have more choices as well. One firm that has capitalized on the opportunities created by SVB's collapse is longtime venture debt lender Hercules Capital, a great name, which typically invests around$40 million in debt deals.

35:47The firm's deals generally have a 15 % loan-to-value ratio, meaning the borrower's loan volume represents 15 % of its total value. Hercules is focused on tech and life sciences. Earlier this year, Hercules entered into an agreement with publicly traded biotech company Moon Lake Immunotherapeutics to provide up to$500 million in non-dilutive capital. At the close, there was a$75 million drawdown with additional tranches becoming available upon achievement of certain specified milestones. Our biggest competition is the equity market. Again, this should be obvious to everyone listening, but founders are kind of going out and saying, I need$100 million.

36:26Do I want to take the potential risk that comes with equity financing, even if I'm going to be, or sorry, potential risk that comes with leveraging debt, even if I'm benefiting from the lack of dilution? Or do I just want to go and raise a fresh primary round? Anyways, good to see the market recover. The thing that stuck out to me about this was the fact that the deals are getting bigger. So venture debt value hit a record of$53 billion last year. The prior year was garbage. $27 billion in 2023. Obviously, high interest rates, a major pullback in the startup and venture community, kind of healing from the turmoil of the SVB crisis.

37:23It was March 10th, 2020. was when SVB went bankrupt. So it's like, yeah, of course, 2023 was going to be a bad year. 2021, two years prior, was up at$42 billion. So we went from$42 billion down to$27 billion, but are now at a new high watermark of$53 billion. But what's changed is the fact that the deal count fell to 1 ,341 transactions in 2024, which is the lowest since 2016. More lenders doing less deals. Exactly. Bigger deals, fewer deals, more embracement, more concentration, more focus on the power law winners. And that makes sense when you see all the capital that's flowing into the really, really big companies, OpenAI, Anthropic, the Crusoe's of the world that are able to marshal a lot more capital.

38:12There used to be a time in the SVB era where you could go do a$10 million Series A and get$2 million in venture debt. And you would just add that on top to kind of build the relationship, build the credit, stretch the balance sheet a little bit more, extend the runway. Maybe you pay. It was always like kind of it all goes into one cash bucket and you kind of spend it however you want. But it kind of set you up to stretch your financings a little bit further. And a lot of companies just by default got on that train pretty early. And I think that SVB leaving the market, obviously, really led to consolidation.

38:51And so we're seeing venture debt go way, way up the stack to a$5 billion venture credit facility for Databricks. Yeah, and if you remember - Even the$20 million series, like venture debt round, that's pretty big for swing education. Bench, if you remember, got in a rough spot with venture debt and ended up being acquired by that company, Employer.com. I guess they were a decent business, but had tripped some debt covenants and were forced to abruptly shut down. Yeah. And then they were ultimately acquired after that. Yeah. And that's the thing is like, you say like, oh, well, I don't give the venture debt lender a board seat, but they do have debt covenants.

39:34they're at the top of the cap table, even above the preferred. If there's liquidation, they're going to get their money out first. And so even though they don't have a board seat, they do have incredible leverage over the company and they actually have a sort of control that should be definitely considered. It's very, one could call it senior. Senior. It is senior. The other interesting thing is Hercules is targeting 15 % loan to value ratio. that feels higher than what I've seen in the past. So loan to value means if your company's worth $100 million, they would loan you up to$15 million in venture debt.

40:15In the previous era, where I was more familiar with this, it was like you would raise$20 million Series A at$100 million post. The VC takes 20 % of the company in a board seat. Back on low single digits. 2 million in venture debt, 3 million in venture debt, like loan to value of like 3%, 4%. But I think this reflects them being more selective about going into companies that are not just post-product market fit and have some VC hype behind them. They have real revenue, real flywheels. We don't know what the terms are on these 15 % loan to value deals. It could be that you need to keep a multiple of the loan on your balance sheet.

40:55Yep, totally. Otherwise, we're going to require. And also, they probably have done some sort of private equity breakup valuation to say that even if this doesn't, even if the company completely plateaus and all the VCs bail, there's still base case, we can get 15 % of the value out here. as opposed to the previous era, which was maybe more risk on in the sense that they were backing earlier companies, but in another way, more risk off because they were writing so much lower. Jason O 'Connor says, this show is like a Buddhist statue I have in the background to hopefully bring in good luck and fortune.

41:35Thank you, Jason. That's exactly what we're going. Jason, we're praying for you and your capitalist endeavors. Okay. Always. I want to pull up this chart on Polymarket of SpaceX Starship fully reusable in 2025. Of course, yesterday was a major market mover. It does not look like anyone had any inside information because the market spiked right as the launch was successful. It jumped from 11 % chance that Starship is fully reusable in 2025 to 71%. Now the market's hovering at 40%. And the rules here are on February 28th, Elon Musk posted that it was likely Starship would become fully reusable in 2025.

42:22The market will resolve to yes if SpaceX or Elon Musk announces that Starship is fully reusable by December 31st, 2025. Otherwise, the market will resolve no. So you are, you know, riding on you know spacex and elon's comms which of course would be hotly debated and obviously there's not a pure binary in terms of what reusability means like clearly this rocket came down and exploded they need to rebuild it but even when it comes down they still need to swap out parts and fix things like they were saying that during this last spacex flight they deliberately took off a bunch of tiles just to kind of like see if they could do it what would And so you can tell as the rocket comes down, a lot of those tiles got roasted.

43:07And so they might have to swap those out. There are elements that aren't reusable. But Elon Musk has a goal. They're definitely going to swap out the stuff that was flying around. All the insulation or something. That was one of the wildest bits that I've seen. Anyway. Well, we have this chart here. Yeah, we have a chart. We love to trust the experts on this show. We are specifically trusting the experts on the White House Intel deal. If you haven't been paying attention, living under a data center, the U.S. government is now the proud owner of 10 % of Intel shares. You, the U.S. taxpayer, now have a small slice of Intel, which also means you have a smaller slice of figure robotics because apparently Intel Capital invested in figure robotics.

43:53If you were worried that you didn't have exposure, you can stop. Yeah, yeah, yeah. You can send the layered SPV emails that you've been getting to the archive and just be riding with Uncle Sam on that one. But we did a little whirlwind tour of who is saying what about the White House Intel deal. Semi-analysis seems pretty bullish. Gave the latest, it gives the last U.S. chipmaker an actual chance to catch up, spur the domestic sector. The bid here is that the Intel story with the government is not going to stop here and that there will be some sort of pressure or incentive applied to other American tech companies to buy from Intel.

44:37Jordan Snyder, friend of the show over at China Talk, is also bullish. He says it's Intel's only chance is government support. The U.S. needs domestic chip fabrication capabilities, and this is a good path. Ben Thompson has a slightly positive bias. He pulled out the steel man term and kind of steel manned it. He said national security and the economy concerns are too accurate, too acute to not keep Intel Foundry viable. Although he does cite Scott Lincecum, who's extremely bearish. And this says this will lead to widespread inefficient capital allocation in the chip sector. That, of course, is the fear.

45:14The Wall Street Journal is also somewhat bearish. Pitting essential U.S. partners Samsung and TSMC against a government-owned competitor is a bad idea. I was talking to a few other folks that both they predicted this, but they didn't think it's necessarily the good outcome. Yeah, and it has ripple effects if other companies are getting strongly encouraged to use Intel's services. Yeah, could make those companies less competitive potentially. Could make those businesses worse, yeah. The Asianometry was constructive on the deal, said could make the deal work by using Intel as an NVIDIA second source.

45:47And, of course, this could potentially hurt NVIDIA's margins because the CPUs that NVIDIA makes, I believe at TSMC, are probably high margin. If they have to pay Intel, their margin will fall. But this is a small piece of NVIDIA's overall business, and so it shouldn't completely upend the amazing business they've built. Of course, NVIDIA is reporting earnings today, and we are looking forward to tracking that. I have a post here from Buko Capital Bloke, the third technology brother. It says, the fate of the world is in Jensen's hands. Good thing he's a psycho. And he has an excerpt from an article here.

46:25That constant kind of relentless improvement, looking down at himself, he never rests on his laurels. One guy told me that after a blowout quarter, Jensen came into the room and said, this morning I looked into a mirror and I said you suck he does these mental tricks to make sure he doesn't get over because he's so scared he's a student of history that the big thing for technology companies to start thinking your hot stuff and then start becoming complacent he is so scared of that so he tricks himself to not do that it's amazing yeah so closing out our trust the experts infographic Bill Bishop is slightly bearish he says Intel fabs haven't proven they can deliver U.S.

47:05government ownership in private companies. That's a slippery slope. So we will be monitoring the slope to see how slippery it is. Hopefully it's the slope of illusionment. What is the Gartner hype cycle again? The slope of enlightenment. The slope of enlightenment. And then the plateau of productivity. The trough. So let us know where you are on the Gartner hype cycle. We are working on some more fun stuff there. We're going to crunch all the data. Maybe we'll throw it in, Julius. Of course we will. What analysis do you want to run, chat with your data, and get expert-level insights? I want to know how people are tracking on the Gartner hype cycle.

47:41Yep. Alex Heath just shared a link. He said TikTok's parent company is making more revenue than Meta and is worth less than 20 % of its market cap. So there's some reporting here from Reuters. ByteDance, the owner of short video app, TikTok is set to launch a new employee share buyback that will value the Chinese technology giant at more than$330 billion, driven by continued revenue growth. The company plans to offer current employees$200 per share in the repurchase program. The people said up 5.5 % from$189. Each it offered them about six months ago. The buyback is expected to be launched in autumn.

48:23And the latest buyback at a higher valuation will come as ByteDance consolidates its position as the world's largest social media company by revenue with its second quarter revenue up 25 % year on year. The jump resulted in the company's second quarter revenue hitting about$48 billion, most of which is from the Chinese market as it continues to face political pressure to divest its U.S. arm. And apparently TikTok's U.S. operation remains unprofitable. The U.S. operation? Yes. How is that possible? Good question. They have so many ads and stuff. And I thought they were taking a big cut of that.

49:01I think TikTok shop is wildly unprofitable. That's right. They were subsidizing a bunch of stuff there. Yeah. Wow. Well, maybe it's got to go to Jensen. Maybe it's got to go to Oracle. We'll keep monitoring. Let's give it to Larry. Yeah. I'm in favor of giving it to Larry. Anyway, if you're going beyond getting your brand mentioned on TikTok, you're trying to get your brand mentioned on ChatGPT, head over to Profound, reach millions of consumers who are using AI to discover new products and brands. AJ over at Semi Analysis says, feels like the fact that Zuck poached only a small amount of researchers from Anthropic is really under discussed.

49:38They have something that really works with respect to culture, mission, and values. And remember, Cursor poached the two people on the Clawed code team. They went over to Cursor for a couple weeks and ended up back at Anthropic. I think it was only one week. One week. That is a crazy short stint. Droid says the market implied mandate of heaven. I mean, yeah, the culture at Anthropic is extremely unique. Like they really do believe that they are on an exponential curve. They believe in straight lines on log linear graphs, as they say. I was thinking about that quote the other day, because if you look at the data center capacity and how pre-training is scaling, like it is a straight line on a log linear graph We're getting orders of magnitude improvements every few years.

50:29And it does seem extremely predictable. Except the last time I was looking at straight lines on a log linear graph, it was COVID. And that, of course, like completely flatlined at a certain point because there are only so many people that can get a disease. And so I'm not, I believe in the straight lines on log linear graphs. But I also believe that straight lines can become flat lines on log linear graphs. And we'll have to keep seeing. What do you think? Then the question is, where are we on the S-curve? Yeah, where are we on the S-curve? I don't know. I think we're pretty early. Pretty early?

51:01Yeah, I do think we're early, but I think that there's a number of small percentage chance things that could result in that S-curve hitting. Everything from bottlenecks in the supply chain that are somewhat intractable. We run out of a certain bottleneck product that causes a delay in the next oom of data center build out. We could also see some regulation. We could see the business model's not quite – the business community could just panic, and it could just become a meme that's like, oh, actually, we overinvested. We need to pull back for a few years. This is what happened during the dot-com boom.

51:46Like a lot of those fiber investments made sense over a 10-year, 20-year period, but they were overheated in the short term. And so that just created like a freeze. And all of a sudden we didn't get a continuous build out. Or we did, but we didn't get continuous capabilities. I don't know. We'll see. What do you think? I mean, I'm bullish, yeah. What do you think about the fact that it feels like model capability is like very discontinuous? like it feels like how do you mean discontinue like it feels like it feels like we get we like like we are on a we are on a smooth path of like bigger models chopping wood training things but like only this is the rune take like like uh as technology rolls out it rolls out in this like smooth curve but then you experience like a few discontinuities with like new capability comes on comes online where like the chat gpt moment like there wasn't really that much like like if you If you were to just look at the straight line on a log linear graph, it would be very hard to say that's the ChatGPT moment, because it was really in between GPT-3 and in between GPT-5, and in between medium-sized models, big models, and then even bigger models.

52:56And you couldn't see, there's no kink in the graph at the ChatGPT moment. There is on the user side, there is on the revenue side, but there's not on the actual scale of data center side. It's not like we went from nothing to something on the actual data center build outside. And that caused the, like, so it's hard to predict where there will be new breakout successes along the curve. Maybe that doesn't matter. I don't know. Yeah, I mean, I don't think it matters that, like, you're probably going to see another kink in user usage when school starts again and everyone's like, wow, GPT-5 is so good because no one's ever used a reasoning model before.

53:34Yeah. Did you see this post about someone said that GPT-5 caused, like, a major drop-off? This is from Marie Martins. The model giveth and the model taketh away. User registrations coming from AI search. ChatGPT was like this huge boom. Can you see this, Drew? And then GPT-5 launched. And this is like way deeper in the slides. Yeah, I saw, I mean, there was another like somewhat similar chart. Yeah. That was like going around a couple weeks ago where it was like, basically you could like tie it to when like schools, it was like in may or something then usage completely dropped off do you remember this chart but then it was actually like oh this is just some like weird yeah this is like for one organization it was only like a hundred thousand tokens like it was obviously not the actual usage yeah i would be very surprised if we're if we're at a phase of chat gpt adoption where like the school usage is still like a major factor yeah just because like the the install base is so big it can't just be students right yeah and then so i mean i'm a little hesitant to like pull a lot of stuff from from that graph sure but it seems it seems very weird yeah i don't know anyway this is super interesting chart from from mary martin she is the founder of tally forms yeah and yeah basically calling out they were getting a ton of registrations from people just in chat gpt and then all of a sudden dropped off a cliff uh so yeah there's some debate in the comments around how GPT-5 stopped adding UTM parameters to requests.

55:13Oh, interesting. There could be an attribution issue there, but they're saying... Yeah, I mean, we definitely want her to zoom out and look at overall registrations and see if there's a fall-off in those because was this purely additive or was it just like an attribution thing? And then I'd love to know, is this actually affecting her business? or is Tally still growing at the same rate? And it's just like, it feels like it's coming from a different source, but in fact, it's all from the same. It's not impacting Popmart CEO and founder Wang Ning. He's now the 79th richest person in the world. According to Morning Brew, his net worth has grown by 20 billion this year.

55:56It makes him richer than names like Peter Thiel, David Tepper, and Steve Cohen. Steve Cohen, wow. LeBoo Billionaire. Laboo billionaire. We helped, we contributed to this by buying a Laboo for Bill Bishop's pot. So it's on the way, Bill. Tashi. Yeah, very excited for that one. Anyway, back to Anthropic. Dario was talking to John Collison on Sneaky Cheeky Pite about how he approaches LLM economics in an interview with John Collison. I thought this was interesting. So Dario says, get kind of burned. There's two ways you could describe what's happening right now in the business model. So let's say in 2023, you train a model that costs$100 million, and then you deploy it in 2024, and it makes$200 million in revenue.

56:43Meanwhile, because of scaling laws in 2024, you also train a model that costs$1 billion. And then in 2025, you get$2 billion of revenue from that$1 billion. And you've spent$10 billion to train the model that year. So if you look at it in a conventional way, the profit and loss of the company, you've lost$100 million the first year. you've lost 800 million the second year and you've lost 8 billion in the third year so it looks like you're getting worse and worse exponentially uh if you consider each model to be a company the model that was trained in 2023 was profitable um uh you paid 100 million dollars and then it made 200 million in revenue uh there's some cost to inference the model but let's just assume in this current tunish example that even if you add those two up you're kind of in a good state so if every model was a company the model in this example is actually profitable and i thought about this in the context of GPT-4, that was obviously, I think that was like around a hundred million dollar training run, something like that.

57:36And they've clearly made a ton of money back from that because GPT-4 is the backbone of something that generates like billions of dollars in revenue. Of course, they've also trained other models that have been used less and maybe were less additive. But this was framed as like, oh no, this is like a nightmare. Like he's exposing the truth that like these will never make money because they're just going to continually invest endlessly. I didn't see it that way at all. This makes perfect sense to me. At some point, you could just stop training. The question is, can you just stop training new models and just reap the profits of old models?

58:09Yeah, well, it's more just like at a certain point, you kind of tap the capital markets because maybe you can't raise$10 trillion to fund the next model on debt or equity or whatever. but there's no reason that you can't like in each of these models if you wait 10 years you can afford the next run right because you train the model for 100 million let's say you were like i'm bootstrapping a foundation model company crazy but you could do it here's how you do it you you you invest you know one dollar in a model that makes you a dollar next year two dollars next year then you take you wait 10 years you save up ten dollars then you do the ten dollar run And you expand until you do the$100 million training run that makes you$200 million the next year and$200 million the next year.

58:59And you know what you have after five years? You have a billion dollars in profit. What do you do with that billion dollars? You train the next model. Then you wait five years. And then you've saved$10 billion. Then you train the$10 billion market. This assumes, of course, that you would have a competitive offering in the market and you wouldn't be getting. Well, yes, yes. But that is a broader market dynamic of whether the market is willing to put up. Currently, the market's 100 % cool putting up$100 million into a foundation model company. There were like 20 companies that raised$100 million to train foundation models.

59:31There's only a few of them left. Then those few, a bunch of them put up a billion to train the billion-dollar model. Now you have who's going to put up$10 billion to do the really big training run? It's going to be a few labs. It's going to be OpenAI, Anthropic, Meta, Google, right? As we get to the$100 billion, yeah, we might have to wait a little bit. We might have to kind of like accrue savings and pay that down over time. And this is another thing that might cause a kink in like the log linear graph if you can't immediately marshal$10 trillion,$100 trillion. Like right now we are in the phase of like snap your fingers and the capital marshals because we're in this era where the gains are clear.

1:00:11Yeah, if you remember Anthropics round was so oversubscribed They ended up, I think, doubling the allocation. Yeah. And so this could continue for a while. I mean, yeah, the concern is that you can't, it seems difficult for even the big labs to do another 10x on these pending training runs. Yeah. I mean, that's what Stargate is, right? It's a$500 billion project. And so that's probably like$100 billion commitment in the medium's term. We're getting into the trillion territory. um maybe it continues forever maybe everyone on earth kind of just says like yes we're willing to like everyone sell all your gold and convert that but i think if you're i think if you're on wall street yes traditional finance yes guy yeah or girl and you're looking at this and and in in biotech you might have something where it's like hey we need to spend five billion dollars developing this drug but then it's going to be good and we're going to be able to sell it for a really long time This is every business ever.

1:01:10So it's like having massively uncertain future capital requirements is going to just be a concern and an asterisk on the business. Yeah. I think that's like FinHub IQ is really the concern here. You have exponentially increasing costs and just wildly uncertain future capital requirements. Yeah. And so the way to deal with that uncertainty, those jitters and the later stage investors who you have to talk to, to Marshall, Once you run out of mosses who are willing to take wild swings, the way to allay their concerns is show them that, yeah, we spent$10 billion and we're making$2 billion in profit.

1:01:47And five years we'll be able to pay that. So you can give us the money now to pull forward what's obviously going to pay off again. And we're confident that the next 10x is going to get us another 10x in profit. And so you can underwrite these at any level. Right now, we're underwriting them every single year. We might stretch out a little bit. We might compress. We might go faster. We'll see. Well, Dan Ratliff in the chat says, Elizabeth Holmes has hit the timeline. She has been posting. She's ripping posts. She quoted Brian, a quote tweeted Brian Johnson earlier. Brian said, defeating death would be humanity's greatest achievement.

1:02:30Elizabeth says, amen. She gives another quote here. there comes a point where we just need to stop pulling people out of the river. We need to go upstream and find out why they're falling in. Archbishop Desmond Tutu. And, yeah, she's ripping some comments. She's getting active on the timeline. Now, this is probably not her, right? She said, mostly my words, posted by others. Yeah, Billy Adams, her husband. Probably over the phone. She just said, she was responding to Brian and said, first they think you're crazy, then they fight you, then you change the world. And somebody quoted and said, she's going to launch cash tag Theranos meme coin, isn't she?

1:03:14And Elizabeth says, never. um anyways having having fun um on the timeline yeah imagine thrill of having posts read to you over the prison phone and then replying okay yes like that one she said last time uh uh uh Signal said, dreams do come true if you put your mind to it. And Elizabeth Holmes followed him. And VC Bragg said, never deleting this app. And Elizabeth Holmes said, last time I deleted the app, it looked different. Where did the little blue bird go? What happened? This is funny lore. This is like an odd. It would be weird to be offline for so long. She's responding to an Elon Musk quote saying, for my part, I will never give up.

1:04:02And I mean never. and Elizabeth says, this is one reason Elizabeth, this is one reason Elon Musk wins. He never gives up. That's what it takes. It doesn't sound like, it doesn't sound like Elizabeth Holmes is giving up either, but she seems to be a big Elon Musk fan. I have been shocked by the out of home campaign. Have you seen this? Which is not by her. Someone is working on a redemption arc for Elizabeth Holmes and Justice for Theranos. I think it's a promotion for the new company of some sort, but it's all over. Yeah, there's a new company. I think they're out-raising. But the billboards are all over LA and they've been up for a long time, like months and months and months.

1:04:49It's been crazy. Anyway. Ben Gilbert says, if you think your app needs more polish, always remember Google Maps, Shift Maps without Europe. That's awesome. That is a crazy screenshot. It's crazy that they do this. Easy mog. Yeah, they didn't even think to put the rough outline of Europe and be like, we don't have details on it. But instead, it's just like you can zoom out and you only see America, the States, and England. And we got Ireland in there, of course. Got to keep Ireland in there. Well, if you're building the next Google Maps, get on Linear, linear.app. Linear is a purpose-built tool for planning and building products.

1:05:23Meet the system for modern software development, streamline issues, projects, and product roadmaps. So Travis Kelsey, right after the announcement, lunches american eagle collection yes and uh michael miraflor says nothing is sacred monetize everything let's see let's check in on the american eagle stock it is up another nine percent today so they're just on an absolute terror yes um everyone in tech has been focused on the the taylor swift wedding um particularly what it means for nvidia earnings i have a little thesis here so So obviously Taylor Swift getting engaged, that's going to draw a ton of attention towards the NFL.

1:06:07We've already seen the NFL had a measurable Swift effect from 2023 to 2024. They saw 9 % gain in female viewership year over year, and game level spikes like a 63 % jump among women, 18 to 49 for a Chiefs game. So imagine this, NFL and streamers, they lean in. broadcasters if they escalate moment capture more shoulder programming alternate feeds multi-language clips real-time highlight reels what do you need you need real-time media it's an ai problem yep you need clipping scene detection player id small object tracking all of this is going to run on nvidia gpus uh globalization you need translation at the edge another classic speech a hundred billion for jensen teach text-to-speech translation that's an nvidia product right there They have a product, Reva, that does that.

1:06:56Creators pile in. UGC explodes. Reaction videos, podcasts, shorts, consumer AI effects, noise removal, eye contact correction, studio voice, increasingly GPU-tensive workloads. People are going to be running RTX PCs with NVIDIA broadcast software to get this stuff. Ad dollars also fall as attention. Incremental audience, especially for the new female demos. This is going to pull premium dollars to sports ads budgets. This pushes ad tech recommender systems, ranking, creative optimization, frequency control. These are among the most compute hungry inference workloads and an NVIDIA sweet spot. They have Merlin for recommendations of ads and Triton for serving ads.

1:07:40Stream quality is going to be an arms race. The women, the Swifties, they're going to demand 4K feeds. Platforms are going to race to improve video. They're going to need to accelerate this with CUDA. This means more GPU accelerated pre and post processing in cloud workflows. Sports production is going to modernize. Autonomous cameras, analytics are a blueprint for low tier games. These techniques climb the stack for tent pole events. More AI in the truck, more GPUs at the edge in cloud transcode farms. hyperscale hyperscale capex is the real lever here AWS Google Microsoft meta they fund the compute behind all this imagine 2026 capex guides are going through the roof over here we're gonna be in hundreds of billions of dollars just supporting the swifty demand for NFL content yeah powered by immersive exactly and so all this is gonna translate into earnings of course showcases durable broad-based inference demand in media and ads exactly the vector analysts want to see as the AI cycle tilts from training only to training plus always on inference.

1:08:50So that's what the Taylor Swift engagement means for NVIDIA's earnings today. Groundbreaking. Gabriel says, this is the music that Keith Raboy listened to in 2019 while doing a thousand burpees and right before logging on to reply wrong to a political scientist posting about Pete Buttigieg's South Carolina polling results. Can we play this video? This video is hilarious uh in the meantime while they're pulling it up i'll tell you about numeral sales tax and autopilot spend less than five minutes five minutes per month on sales tax compliance numeral hq.com and you know who's going to need to use numeral travis kelsey's new american eagle line if you buy some american eagle uh they said inside travis kelsey's new american eagle line an ad with athletes including suny lee and kian anthony the day after it was revealed that travis kelsey put An eight-carat engagement ring on Taylor Swift's finger.

1:09:42American Eagle had announced it had a limited-edition clothing line with its sportswear brand, True Colors. AEX True Colors by Travis Kelsey will launch in two drops today and up to September 24th. So get ready and make sure that you're paying your sales tax on TrueColors. TrueColors. TK. Michael Weirfleur says, nothing is sacred. Monetize everything. Let's pull up the DJ. The video is going to be almost impossible to find. Okay. It's deep in the... Oh, they got it. They did it. The team got it. The impossible. Let's pull it up. Let's play this. Get the audio. Get the audio. The audio is key. Let's go full screen on this.

1:10:24This is like so annoyingly... I don't know. I was not a fan of this. Did you listen to this? You think it's good? Very Reddit coded. Yeah. I can't believe there's so many people at this putting up with this Mario song. What event is this?

1:10:48It does inspire me. You know what we need? We need what's coming up next. We need this. You see the... It's like they blow this fog around. Look at this. So great. yeah those like uh co2 can you cannons for sure like that to blow okay we got something good out of this man we need the co2 can you want co2 whatever those yeah whatever those dj equipment are during the lightning round when we're interviewing a cnc can you look into the health impact of being in a studio space with co2 cannons regardless this is very david solomon encoded. Yes. You have to be a Mario fan to understand. Yeah, I get it.

1:11:32I don't know if this would be for me in a big crowd, but who knows? Maybe it's fun. It seems like they had fun, and I'm happy for them. These days not big into paying to be in a crowd. Yeah. Or pay to avoid one. Yeah, yeah, yeah. That's rough. Anyway, let's go to Will Manitis. He says, hard to think of a piece of writing that's had a greater impact on my life than John Ludig's index mindset in 2021. This pro-index tendency, writes John Ludig, pervades the private tech markets, startups, and even our culture through what I call the index mindset, a focus on preservation over creation, optionality over decisiveness, general over specific.

1:12:11Public companies are an obvious thing to index, but the index mindset manifests in many domains. In wealth, instead of risk-seeking expansion, you focus on steady compounding preservation. In venture capital, instead of concentrated bets, you focus on deployment pace and IRR. In public markets, instead of active funds like hedge funds, which we'll talk about later in the show, we focus on passive funds and diversification. In employment, instead of four to five years, ten years and being focused, you focus on one to two years, ten years portfolio building, lots of logos on your resume. In research and development, instead of being radical and innovative, you focus on being incremental and defensive.

1:12:54And in your personal life, instead of spiky and committed, you focus on being well-rounded and committal. Internet software companies are far less risky than they used to be. Even at the early stages, there hasn't been a venture vintage since 2002 with negative median returns. Big tech is now huge tech. So the VCs are not taking nearly enough risk because they're not losing money. They should. It is an indictment of the venture community. if there's not a single fund out there that has a negative or a single vintage that has negative returns because they weren't taking enough risk. Well, it's median.

1:13:25Yeah, median. But yeah, and it's vintage, so it's broadly. Of course, there are funds that don't do bad. But it's like as an asset class. Well, we might be turning it around. Anything's possible. Risk-averse people and money have flooded in. When people have something to lose, they protect their downside. Think of wealth managers encouraging a safe mix of stocks and bonds. The tech industry has too much to lose. And I love Wilmanitis has the most chaotic way to highlight I've ever seen. You can just use boxes in iOS if that's where he's editing this. But he's just drawing crazy. It really draws you in, drawing all over it.

1:14:04Drawing all over it. And then he switches colors at some point. I have no idea why. But Ben Gilbert says he's so good, and I agree. Shout out to John Ludig. Satrini says, time to bust out the no-sequel Negronis again. So I think this was a picture from 2021. I think so. There was a Tiger Global meetup. This is when I think it was roughly 30 % of all dollars in the venture industry were going to founder dinners and little events like this. Essentially. It felt like that at the time. It was closed. They had the NoSQL Negroni. Yep. The Tiger Tonic, the Machine Learning Margarita, the AI Artificial Inebriation Special.

1:14:47Which is a non-alcoholic beverage. Interesting. Interesting. I say bring it back, Tiger. Yeah. I think you could go further. Further. We got to come up with our own little menu of funny things. So we didn't have a menu at our event in New York. This was an interesting screenshot. This one from Food.ai, the number one AI agent for customer service. number one in performance benchmarks, number one in competitive bake-offs, number one ranking on G2, fin.ai. I will never apologize for cutting you off. For an advertiser. For an advertiser. Continue, Jordy Hayes. Event. Fewer people can bench 225 than are worth 10 million.

1:15:24In the US, about 0.4 % of the population, around 1.3 million people can bench 225 pounds. Globally, the percentage is likely below 0.1%. Wow. With some estimates as low as 0.07%. 2.13 U.S. households have a net worth of$10 million or more, constituting approximately 1.6 % of all households. If you can bench two plates, you're already stronger than 99 % of people on the planet, just maybe not richer. And Zach Voll says what everyone is thinking, the only job is to do both. And we firmly agree. Both are a grind, but well worth it. The grind at$2.25 was serious for many on the team. And I think almost all of us have gotten there and are now beyond.

1:16:14It's well worth it, well worth it. Buco Capital says, your stock trades at$59, but we can make it trade at$62 if we manufacture a crisis by changing the logo and immediately change it back to the original logo. the strategy I don't think this was their strategy but feels like something out of Nathan for you episode and I would it's hard to it's hard to properly manufacture a reaction like that of course wow did you so the Cracker Barrel story is in the journal today logo change spawns firestorm it's oh and Trump said something about it Cracker Barrel planned to celebrate a fall menu and logo makeover with a festive country music concert in New York City.

1:17:05I had no idea that's what they were planning. But its new logo stole the show and not in the way the company intended. On Monday, the company apologized for how it communicated the changes but didn't pivot from plans to keep updating the brand. The chain replaced its longtime logo featuring a man in overalls leaning against a barrel with a streamlined version featuring just the chain's name. The move engulfed the restaurant in a culture war firestorm. with some commentators online and some customers accusing Cracker Barrel of eschewing its country charm and heritage for a sanitized image. Critics have lobbied, have lobbed personal attacks on social media against Julie Fels Massino, chief executive of the nearly 56-year-old chain.

1:17:46She scrapped a beloved American aesthetic and replaced it with sterile soulless branding, wrote the woke war room on X in a message shared by Donald Trump Jr. She should resign and be replaced with leadership that will restore Cracker Barrel's tradition. The fallout has shaved tens of millions of dollars from the company's market cap, spawned calls for boycotts and risks the casual dining chain's turnaround plan. Cracker Barrel has defended its changes, saying the all the more campaign was meant to honor its legacy while bringing new energy to the brand. Our values haven't changed and the heart and soul of Cracker Barrel haven't changed, the company said.

1:18:26The new logo, the fifth iteration in its history, is a callback to the original one with its barrel shape and word design. On Monday, Cracker Barrel said the dust-up over the past several days had shown how deeply people care about the chain. Trump said on Truth Social yesterday, Cracker Barrel should go back to the old logo, admit a mistake based on customer response, parentheses, the ultimate pull, and manage the company better than ever before. Make Cracker Barrel a winner again. and they immediately responded, reverting it back to the old logo. And if you had invested in Cracker Barrel and NVIDIA exactly a year ago, you would have outperformed buying Cracker Barrel.

1:19:07Wait, wait, sorry. What was that? Up 52 % in the last year. No way. And NVIDIA is up 41.5%. Well, if you're looking for exposure to Cracker Barrel, either long or short, go over to public.com investing for those that take it seriously. They got multi-asset investing, industry-leading yields, and they're trusted by millions, folks. This weekend, says Jeff Morris Jr., Martin Casado dropped a tweet that took over VC Twitter. I wrote about it and what being consensus versus non-consensus actually means to today's market. Every weekend, there's a new hero, says JMJ, that takes over venture capital group chats.

1:19:43This past weekend was clearly slow on the news front because what I initially read as a fairly standard take ended up dominating my timeline. this is, we talked about this, Martin Casado said, the idea that non-consensus investing is where the alpha is is actually quite dangerous in the early stage. Follow-on capital tends to be more and more consensus aligned. For whatever reason, this concept triggered a bunch of investors. We're living in an era where consensus categories like AI, crypto, robotics, bio, defense, autonomous systems, manufacturing, are big enough to carry entire generations of startups.

1:20:11You don't need to invent a new category to generate massive outcomes. And that is an interesting take because, like, Like, these are supposed to be like tech investors, internet investors, and yet like consensus categories include crypto, bio, defense, manufacturing, AI. It's like everything is now like backable. There was an interesting quote in here that I wanted to highlight. Something about, yes, you can still find non-consensus companies within consensus categories, But at some point, calling ourselves non-consensus investors is just glamorizing the job. Take American Dynamism as an example, a venture category branded just a few years ago by A16Z that now feels like it's industry in itself.

1:20:56Palantir in 2003 was truly contrarian. Nobody else was building forward-deployed CIA software. Palantir was one of one company that probably only Alex Karp, Joe Lonsdale, and Peter Thiel could have created. Two decades later, it's more relevant than ever with a$370 billion market cap. And Earl was equally contrarian, but for different reasons. In 2025, he says, fast forward, American dynamism is a full-blown investment category. Hundreds of startups and dozens of funds are chasing defense, aerospace, and manufacturing. I live 15 minutes from El Segundo, which has famously become a dense hub of these companies.

1:21:27There was another one in here about DSC and saying something about that category being even that category. One note here, I was talking with a buddy over the weekend, and he was like, what's going on with El Segundo? Or is it legit? Are all these companies, are they actually going to build big companies? And something Jeff says here, my reaction was a small number of them will be massive and most will have middling outcomes or, or, uh, will shut down. Uh, but that is, that is what our entire industry is based on and it's okay, but it's still good that we have this excitement and energy and momentum and, and a center of gravity and El Segundo.

1:22:22Yeah. There was something in here. I can't exactly find it, but it was basically saying that like, like consumer is contrarian right now. It's not very hot. Very few people invest in consumer D2C companies. But that's just in like Silicon Valley Venture. There are still plenty of funds, both on the private equity side and on the venture side that are doing deals in consumer all day long. Some brands have kind of expanded and shifted away from it, but there still are lots of funds out there that are doing that. So it's just an interesting noodling on what stands for consensus or contrarian right now.

1:22:59Anyway, let me tell you about Adio. Customer relationship magic. Adio is the AI native CRM that builds, scales, and grows your company to the next level. You can get started for free. With a new$52 million. Do you want to do this John Woo post? Series B. Yeah, this John Woo post is great. We should run through this. So John Woo says, do not trust millennials to run your brand. Their vision of the way the internet works was informed by the web 2.0 Facebook 2011 Instagram vintage filters era and their priors have not been updated since. The internet is no longer deterministic. It's algorithm driven and not even among your own followers.

1:23:32Your posts are constantly being shown to a panel of users. If they like it, the panel is expanded. If they don't, you don't get reach. That means putting out a high volume of posts isn't noise, spam, or quote, low signal. Volume is necessary so the algorithm can pick up your winners and make them viral. You have to experiment with high volume, learn quickly, and double down on your winning formats. The cost of failure is zero. The tree falls in the forest and no one sees it. This is huge for content creation. Just the idea that no one sees the flopped posts. Yep. The cost of experimenting, but the cost of not experimenting is astronomical.

1:24:10Sure signs of incompetence in the attention age. We don't want to put out noise. We need to protect our brand. We need more followers. That's ignorance. Low quality posts are no longer distributed. Follower count doesn't matter. The news cycle is shorter than ever. People will only remember your winners, pump up the volume. Totally agree. I think if you look, you know, on X specifically, if you look at your favorite poster, if you actually scroll their feed, they'll have a bunch of just like flops. Yeah. You'll think of a person as like, oh, every time I see one of their posts, it has thousands of likes.

1:24:39And then you go to their account and you see so many flops. And the real secret is that they're trying a lot of experiments. When the current thing happens, they're putting out every possible variation of the current thing. It's a good strategy. I like this reply by Holin. She says, millennials are still good brand builders, but the execution should be left to the terminally online. And so there is something about like, it's probably not enough just to have like pure slop. It's more like you need a seed of something good, some sort of North Star. But the execution in the modern era does require a lot of aggression.

1:25:14Anyway, we have our first guest of the show joining in just a few minutes in the Restream waiting room. We will get to the Numeri story in just a little bit. Should we talk about the App Mafia quickly? If you haven't been following this, some folks that we're familiar with, one of them who did PMF or die, had put out a course on how to build an app that makes money, don't raise venture capital. They sort of took over the timeline with a viral video showing them driving Lamborghinis, kind of all the classic hallmarks of the guru that you previously found on YouTube with the here in my garage, that guy, and folks on Instagram.

1:26:00It was sort of the first time we really saw it on Twitter or on X. And it caused a lot of, you know, people going back and forth. Harry Gestenter, I don't know how to pronounce his last name, but Harry says, apparently I mentioned in the first few minutes of the App Mafia course, so here are my thoughts. I went to college with Blake Anderson and Nozak and Hunter too. There are a lot of threads going around about profit margin, et cetera. It's not that relevant given these kids are clearly netting a minimum six to seven figures, probably for the first time in history for anyone that age. This represents a major shift in the education system that platforms like WAP have pioneered.

1:26:36People are talking about audience IQ. This course is clearly targeted at kids slash younger founders and certainly something I would have loved 10 years ago. It also represents a shift away from venture-backed consumer apps dying from over-raising. Hundreds of apps like Headspace Calm raised eight to nine figures and were stuck with companies that couldn't quite IPO, but also couldn't cash flow for the founders. There's an entitled socialist mentality of, quote, why do I have to pay? The reality is if it's free, it's not very valuable. I haven't watched the course, but these guys are extremely sharp and can provide value to young founders.

1:27:11So they deserve to be compensated for their time. Now there's a wrinkle where the course is now free. Is that right? The course is now free. Okay. So I think the original video was, was, in their words, designed to elicit a dramatic reaction, designed to bring out hate. They're sitting there in a mansion in Miami. Blake loves poking the bear of the algorithm. He's very good at that and very deliberate about that. But yeah, it is fascinating. And then immediately people piled on and were saying, oh, the apps must not be making a lot of money. Otherwise, why would they do this? which I didn't, it seemed obvious to me why they wanted to do it.

1:27:52They want to build their personal brands. And this is like a business model for building a personal brand. It's a, it's a. What's interesting is I don't think I've ever paid for a course on specifically on like entrepreneurship or something broad like that. But I have paid for courses on like very specific technologies. Like I went to a, like a bootcamp for iOS development where they taught us objective C for two days straight. That had like a real cost and I was able to expense it against the business. I also paid for like a course to learn Houdini, a particular visual effects software. And that kind of had, I thought it wasn't like a get rich quick scam at all.

1:28:30They didn't make any promises. They just said, hey, we're going to teach you the software. We're going to teach you this thing. And so buying books is great. Buying courses on specific topics is great. Lots of what you can learn about entrepreneurship is just available out there for free. You can read Paul Graham's blog. You can listen to Founders Podcast. There are plenty of resources. Maybe this is useful. You'd have to try it out. Now it's free, so you can make your own decision. Anyway, we have our first guest of the stream in the TV. Coming in. Welcome to the stream. Josh, good to meet you.

1:29:02How are you doing? What's happening? Hello. Good to see you both. Hi, John. Hi, Jordy. Hi, Jay. Great to see you. Joining the group. You sound fantastic today. Whatever microphone you have hooked up is working flawlessly. Similar to what we have here. uh congrats on the news break break it down for us yeah so i'm josh head of gusto uh we love serving small business uh we've worked with a company called guideline for a number of years close to 10 years at this point and the news is that we are joining forces guideline is going to be joining gusto amazing amazing i think i've i've actually used gusto and guideline together for my entire career i've used gusto at four different companies i started in college And we have an employee sitting over there that wants a 401k.

1:29:45You have no excuses anymore. Time to set up. Yeah. Talk about the prehistory here. The company's been, was it YC 2013, 2012 or something? 2012. 2012, yeah. I remember implementing it at my YC company in 2012, back when it had a different name. And I remember that for a number of different HR functionalities, I would have to kind of go to a different platform and then the data would be fed back. What informed the strategy back then and then why are you shifting it? It seemed like it was working for a long time. So why kind of, what is it, horizontally integrate, I suppose is the term? I don't know what other business is.

1:30:32We just try to make the life of a small business owner easier. But yeah, first off, I mean, thrilled to serve you and your companies, and we're always trying to get better. So send me feedback on how we can get better, by the way. But yeah, no, I'd say we're going to go way back. We launched Gusto back in 2012, and our first product was payroll. If you don't pay someone, they quit. So it's the least optional part of the stack. And so we've made progress there. You know, we've served over 400 ,000 businesses. We love helping tech companies. That's a lot of businesses. There we go. That's a lot of businesses.

1:31:04But I also love reminding folks, there's more dentist offices in the US than tech startups. So like very focused on that mainstream small business owner. And yeah, to your point, we listen to customers more than anything and they guide us. And so they told us, hey, there's all these other pain points we want your help with. Our second product was health benefits, but we're not going to build everything ourselves. And so when retirement came up and like, generally speaking, what we like to do is take stuff big companies have and bring it to small companies. So from almost a fairness lens, it's like, we need to bring great retirement benefits to these small businesses, they deserve it.

1:31:39And that was basically KB, one of the co-founders of Guideline. It's kind of a funny story. He was previously the co-founder of TaskRabbit. And so back in 2015, we were in the same building. They were on the second floor, we were on the sixth floor, and there was a slow, slow elevator, the slowest elevator you've ever seen. And so we would sometimes just be on these long elevator rides, and that's how we got to know each other. Yeah, I was going to ask when did the conversation first start. Yeah, people say you meet your acquirer like years before, but what they don't say is in a very slow elevator.

1:32:10You're going to hang out for minutes at a time. But yeah, to answer the heart of your question, we started partnering with them when they launched in 2016. And yeah, there's an integration. You can kind of connect the two systems. Like we have tens of thousands of shared customers. You know, KB, I can say this stat very proudly on his behalf, like over 10, around$10 billion of retirement savings in the accounts of our shared customers. Let's hit the side for that. I mean, Jordy, you hit the gong. We like to hit the gong around here for big numbers. Congratulations. I like it. I like hitting the gong for retirement funds.

1:32:46Yeah, yeah. That feels good. That's, again, that's their people's money. That's their retirement peace of mind. But yeah, what we realized is there's a couple of things happening. One, we can do even more together. So we're going to continue to partner with third parties on other products. But especially with 401k and retirement, we're going to really go deep. And that's what Guideline has done. They're a broker-dealer. So there's all these different parts of the stack where if you cut out the market, you can create better cost savings. I'm sure you had investors along the way that said, you're sending so many customers to Guideline.

1:33:18Why don't you just build this? If you need more capital, we'll provide it. Why did that not make sense? And why buy versus build in this situation? So I think all options are always on the table when you're building a company, buy, partner, build. And for what we're doing, it's going to be a lot of connected problems. So how do we bring it together as one simple, easy-to-use product is the key. But I'd say, yeah, the catalyst for shifting from partner to being one company is, like, there's just products that we can do that we can't do unless we're combined. I can give you a couple examples if that's helpful.

1:33:57Yeah, please. Yeah, so, like, the journey of building a company or the employee journey. We think a lot about kind of the person by person, what happens in their life. But let's say someone gets promoted, right, and they get a pay increase. So that's obviously coming through our system because it's going to affect their salary. And in that moment, whether it's in the app or in an email, the, hey, you now have more money coming into your pocket, your choice, but do you want to set aside a bit more in your 401k? Or someone has a kid. In our system, that's a dependent, but that's a human being. Someone just had a kid, right?

1:34:31Maybe you want to go do something there in the context of retirement planning or something related to health benefits. So kind of thinking through the person-by-person journey piece of it, there's just things we can do that we think will drive more folks offering retirement benefits. There's also a compliance reason too, which I can get into. Let's get into it. We love this. Yeah. So like government, right? We're here to like abstract government and sometimes there's carrot stick that government does. So 401k is a good example of a carrot, like it's pre-tax dollars. So literally the government is making it hopefully a no brainer to like put aside money for retirement with meaningful amplification of that money than just putting it post-tax.

1:35:14But they also have compliance requirements. I think over now, close to 25 states have rolled out what's basically a mandate that businesses have to provide some type of retirement benefit for their teams. And so that's good intent, but it's still stressful because there can be penalties or fines. And so we're also excited that basically we can give peace of mind to more small businesses so they're not stressed out, but definitely not getting fined or penalized. And if they want to or need to set up a retirement benefit, we can make it brand-to-and-simple through Gusto Plus guideline. What's the anatomy of a partnership look like in this category pre-acquisition merger?

1:35:52We were talking a few days ago about the relationship between Apple and OpenAI. And we were kind of like, wait, like, should Apple be paying for the right to have good AI on their phones? Or should ChatGPT be paying? because they get all these extra queries. And I think the rumor is they netted out to be like there's no money transferred. And I could imagine that sometimes that comes up. Sometimes it's like, hey, I'm getting you customers. You got to give me a fee. What are the structures? Do you have companies that are in one camp and the other? Can you share anything about the relationship beforehand?

1:36:26I'd be interested to know. Channel sales is something I'm learning about now. So the first thing to start with is Gusto's entire engine of growth. you know, here's another metric. We can tee it up. But we're going to, we're on track to add about 150 ,000 customers this year. Wow, that's a lot. Is that a lot of small businesses? That's small businesses. Wow, that's a lot of businesses. Yeah, yeah, yeah, okay. And so you're talking to some guys that absolutely love business. So this is fantastic. The best news I've heard all day. We can get into our whole, like what is AI going to do on the world tangent if you want to.

1:37:01That's interesting too. But yeah, we'll come back to that. Yeah, so anatomy deals. Yeah, so basically, we are an engine to go bring in more and more small businesses, especially new employers, folks starting out, right? If you don't pay someone, they quit. So it's right off the bat an important key thing for you to do if you're building a team or adding even your first employee. And so there's a lot of folks, as you can imagine, that would love to distribute through Gusto because of the scale that we're at. And so what that looks like first is not us starting with the partner, but starting with the small businesses and finding out, like, what are their pain points?

1:37:36Who are the products? What are the products they like or want to use already? They're voting with their actions. They come in through our integrations. And then we do a lot of vetting. So I think more Apple approach than Android. We want to make sure quality experience, accuracy, reliability, and then, like, just this obsession with user experience comes through of whoever we're working with. And then, yeah, we'll do partnership deals across multiple product areas. Time tracking is another example. Sometimes we'll have first party, like we'll do our own product, and still have best of breed third party.

1:38:05Sure. And then that's a negotiation. Usually if we're going to drive a lot of customers to someone, there's some type of rev split because we're creating value for them. There should obviously be value for us. Yeah, that makes sense. What is your guidance for a ton of startups sign up for Gusto today when there are one or two people? It's just the founders. And what is your kind of framework for, you know, the right companies to use Gusto? Because, you know, again, companies today, they might start with a couple founders today and be 30 people in six months, right? These things happen quickly.

1:38:41We've got to vibe code our own. But how do you look at it? We've got to whip this out and vibe code our own. What did it go wrong? Are you counting the AI agents as people or not is my question? Yeah, that's a good question. But yeah, how are you thinking of serving startups as of today? It's obviously, from my understanding, a lot of where you guys started, but it's always evolving. Yeah, I mean, as a quick backdrop, you're alluding to it. Three founders, we all started here in Silicon Valley. We went through YC. I was born and raised here. My parents aren't from tech, though. They came from, like, my dad was a teacher.

1:39:12My mom was a teacher. And so... Same here. But we are physically here. Oh, yeah? What did you teach? In high school that I went to. High school. My dad's out of high school. Very cool. But yeah, it was funny. I think, I don't know if I asked my dad this. I don't think he would have had a great answer, but I do remember asking somebody back in the day, how do I pay myself? I had a company, it was starting to make money. I was like, how do I pay myself? And they were like, oh, you just set up like Gusto and you just onboard yourself. And I was like, oh, that's cool. Well, we love serving small business.

1:39:45Obviously tech startups are a part of what we do. We love serving them too. I guess from a target customer lens, broadly speaking, one to 100 person companies is our sweet spot. That's 98 % of the employers in America. But I always love, like also, it kind of blows people's mind usually, there's six million employers in America, give or take. Two thirds are less than five employees. Two thirds are less than five employees. So whenever I meet a small business owner and there are three people and they say, I'm small, I'm like, no, no, you're not small. You're like the common business owner in America.

1:40:16You are the average, yeah. Wow. Quick thoughts on AI, because you alluded to it earlier. Yeah, I'm interested specifically in how you think about surfacing AI. Like, do you want to put a chat box in front of someone, and then that's an interface to all the deterministic workflows you've built, or do you want to just turn AI loose within your behind-the-scenes development team and have flows that are working just to improve things, reality check things. Like you're probably using machine learning for fraud detection for years. Is it an evolution of that? There's a product manager inside of you.

1:40:55I would say yes to both and more. So like we do a lot of complex kind of backend processing of either money movement, filings, documents. There's a lot of ways to use AI tooling for productivity there. So that's great. And we've always used technology for how we scale, right? We serve over 400 ,000 customers today. That's only possible through using a lot of software. But from a customer experience lens, actually, again, both, I really think there's this massive understatement here, but paradigm shift around how we use software and how software helps us. When I got into software in high school, what I loved about it is it felt like it gave you superpowers.

1:41:33Then every time I saw books on how to use a tool, or I learned like someone would brag on LinkedIn, I can use this tool. It like really, really annoyed me because software is not something you should have to learn how to use. It should just work. It should just make sense. It should be intuitive. It should give you superpowers. So conversational interface, I think, is going to be a pretty, I mean, again, understatement, but like much more accessible way for a lot of folks to use software. And it'll work in combination with web workflows and stuff like that. And so we have a conversational interface for Gusto called Gus.

1:42:03And to your point, it can take actions, right? You get it. You get it. But like you can like shut up a shift schedule. You can go run payroll. Like you can leverage all of our APIs that we spent 14 years building and Gus can do stuff for you, not just give you information. Funny story. My dog's name is Gus Gustavo and I have a friend who could never, he was so into B2B software. He works in the B2B software industry that he didn't understand that my dog's name was Gustavo and for a long time he would call him Gusto. I was like, I didn't name my dog after the software platform. Speaking of dogs, Bill Bishop is in the Substack stream from Cynicism, and he says he's a happy Gusto client.

1:42:47So you have another fan listening right now. Thrilled to serve you. I have one question, and I'm curious. When I got onto Gusto back in the day, it was close to a decade ago at this point, i was i was confused that i had to hit the button to run payroll oh yeah this is not a payroll platform yes why do i have to and i remember there was a couple times that i would get an email and i'd be like i'd be like you're like you need to run payroll now and i was like why i i think you do you can't automate it now but is there yeah yeah is there is there kind of a generational divide there in some ways where there's like a generation that understands like payroll is something that you hit a button and you run it, versus payroll is just something that is just humming.

1:43:33At the first startup I ever interned at, the CFO hated me, because I would always turn in my time sheets late, and he's like, I need to run payroll. And I'm like, just pay me the same every month, I'm fine. Like, you can just put it on autopilot. And they're like, autopilot doesn't exist, this is 2006, or something like that. So we've rolled out autopilot for payroll back in like 2013, 14. I remember using it, it was fantastic. Yeah, so the key, the difference here between both of your use cases was if you have nothing to input, so if it's like, let's say all salary versus some hourly and or like no one took PTO.

1:44:05But again, there's there's inputs that go into payroll. Now, assuming that's all there, like it should just work. And that's the way it does work. Way back when, this is now 2013 ish, like we have a Penny the pig is kind of our like kind of a part of our company kind of brand culture mascot. I could see Penny and Gus potentially beefing over who's the most important. I mean, they're friendly with each other. There's a friendly competition. But we had to have the GIF. It was a GIF for running payroll when you clicked run, and it went too fast, and we had customers back then go, like, it can't be that fast, something is broken.

1:44:41So we had the GIF just run longer for the sake of it. Just to make people be like, hey, there's actually hard processing happening behind the scenes here. And the world's come a long way since then. So people expect stuff to be fast, simple. We want to automate. I mean, a good example is 401k. So if you want to go change the contribution that's being made, that can now be done easier by us being one company. But yeah, we think eventually payroll becomes something like flowing water. It just works. And you don't have to do much direct action yourself. Fantastic. Well, thank you so much for hopping on.

1:45:10Great chatting. The acquisition makes a ton of sense. And congrats to both teams. We'll talk to you soon. Yeah, and make sure you give that teammate a 401k. We will. We will set it up. It's happening. Have a good one. Awesome. Talk to you soon. Good to see you. Bye. And we have Keller from Zipline in the Restream waiting room. Big news. Zipotle is now live. Chipotle partners with Zipline for aerial delivery. Let's go. Welcome back, Keller. We missed you. Welcome back to the stream. How are you doing? Even more dramatic background this time. Every time you deliver. Thank you so much for hopping on.

1:45:43That's our job, right? Yeah. That was unintentional. Anyway, how did this come together? it seems like the perfect partner. Walk us through the evolution, actually what went into the announcement. Yeah, I mean, we've been talking to Chipotle for a while. We obviously have announced a few other restaurant partners. I mean, you know, for reference, like there was a time in my life where I lived on Chipotle. Like that was my primary. Every entrepreneur has like a Chipotle era. Raise your hand in the studio if you've lived on Chipotle. I think everyone's raising their hand. Yeah, you're getting more.

1:46:18it's just a period of your life where you're getting more calories from chipotle than anywhere oh yes absolutely i can distinctly remember uh those moments it's the base of the pyramid i also remember i also remember i feel like you guys would appreciate this you know there was a time when i was living out of my car actually and like eating chipotle and you know ten dollars you could get like two full meals out of it and then you could you could get like the tortillas too that was the hack there's a lot of optimization that you can do and i remember thinking that like Like one day I will know, like, how will I know if I'm rich?

1:46:54I will be able to get the guacamole every time without feeling guilty about it. I have the exact same story. When I went through YC in 2012, my business partner, my co-founder would always get double meat. And I was like, we're trying to save money. We can't afford double meat. Double beans. Double beans are free. It was tearing us apart. And then I finally just defaulted to like, look, if you're going to get it, I'm going to get it too. And we will both fall on the sword of higher burn. And to this day, if I eat out with him and we're at a restaurant, I'll just be like, get whatever he gets. Because I know he's going to get something ridiculous and he's going to get the most expensive thing on the menu.

1:47:32And at least then I get stuck with the half of the bill and I'm happy. And I'm not like, oh, I got stuck paying for his expensive thing. Anyway, congratulations. We love Chipotle. What else? It's exciting. What else can you share about the progress of the business? It felt like classic overnight success, as we love to talk about on the show. Also, how did you get them to actually name it Sipotla? That was their idea. That was their idea, not ours. That's amazing. I think, you know, look, I think that what we're, you know, obviously there's this like big transformation coming in AI and robotics.

1:48:05And there are a lot of, you know, very fancy hyperscalers that get talked about all the time. But the reality is that like all the basic businesses and brands that we use and love also want to be able to take advantage of this kind of technology And so I think a lot of you know, some some of the biggest brands like Chipotle are asking themselves like hey, how can we use? robotics automation to like accelerate or you know dramatically transform our own brand and the kinds of experiences we can provide to customers So yeah them choosing the potlay us starting to deliver for them I mean, I think that we did a delivery in five minutes yesterday from like order made to the thing arriving at a customer's house That's not possible every time just to be clear It does take time to make the food but like, you know, I think that Something showing up super super fast like that.

1:48:50In fact, we've found it has been necessary because we only started really Rapidly scaling these food deliveries in the last couple months It's actually necessary to tell customers to slow down and make sure to blow on the food. We've had people accidentally burn themselves Wow. Good problem to have. It's like a weird thing. Yeah, weird thing. People are just used to it being cold or smush. Yeah, just not that great. So, really cool to see that starting to scale, especially for a brand that's very near and dear to my entrepreneurial journey and heart. I mean, yeah, other things that we're seeing, even since we were last talking, which I think was like, I don't know, 10 weeks ago or something.

1:49:28Yeah, a couple months ago. We have seen a problem that we were having in July was that the service was growing extremely fast, between 25 % and 30 % week over week in terms of flight volumes. In fact, yeah, so Saturday, new record high flight volume. Sunday, we actually beat Saturday by 20%. This is just two days ago. Monday was a record day. Tuesday was a record day. That's amazing. Really cool. If you keep giving us records, I'm going to keep hitting the air.

1:50:03Cooler with the records. No, but even cooler is that right now the service has a net promoter score of 94. Very few services in the United States are close to that today, let alone delivery services. A lot of customers are ordering like three to four times per week. In fact, we have many customers who are ordering a couple times a day via this service. I think that when you realize you can get stuff in five minutes, it actually starts to change the way that you kind of live your lives. It's giving people back like three or four hours a week when they can just be focusing on their kids or on their family rather than like stressing out with traffic and trying to get somewhere.

1:50:42At this rate, in a couple months, our deliveries in the US are going to exceed all of our global flight volume across eight other countries. So, yeah, U.S. is growing super, super fast now. We're launching a new Walmart Supercenter every week at this pace. It's just, yeah, I think the moment for automation and autonomous delivery has arrived. So is Walmart... It's a 10-year overnight success. Should I think about the expansion as driven by go where the Walmarts are and then talk to the city officials to get support and Walmart's kind of like the backbone? Or is it more like find other cities like Dallas, whether they're in Texas or not, and then partner with companies and businesses that are in those kind of friendly cities that understand the value that this can bring and the demand and then kind of roll out that way?

1:51:34Like, is there a particular path that you're taking? Like, how should I imagine the map of Zipline growing or the coverage map growing over the next few years? Yeah, that's a really prescient question. I mean, Walmart has been a key partner. We're building a lot of charging infrastructure across these metros with them. We're also building independent charging infrastructure that's just designed to serve a lot of different customers at once. So, for example, we just opened a big charging site in Rowlett. It's a small town inside Dallas where that one site is going to be able to serve, I think, upwards of 40 or 50 restaurants that are all within range.

1:52:11And then from there, we can deliver out to customer homes as long as they're within 10 miles. So like one key thing to think about here is it's kind of a network optimization problem But we go a lot farther than normal delivery You know where you have a human driving a car wants to go any from any given restaurant You can typically reach ten times as many people via instant delivery as as via traditional delivery so and again You know ten times as fast so we're really focused on going metro by metro right now a lot of our costs You know maintenance kind of happens on a metro basis So we're going to go tall.

1:52:45Like the goal is to go very tall on each metro. We have a lot of metros now that are like super, super excited and basically on the roadmap for us to launch. And we're trying to build the capacity and launch them as quickly as possible. But, you know, building this kind of infrastructure across the U.S., it's like it's a big undertaking. And the other cool thing to mention, just really quick, is that like, you know, one thing I'm very, very proud of, more proud of every day. I mean, Zipline now has 120 million commercial autonomous miles. It's the largest commercial autonomous system on earth.

1:53:18Zero safety incidents. That's the main point. Zero safety incidents. Actually, if you just look at the basic safety of cars globally, the system is already provably about 10 times as safe as cars in terms of fatalities or injuries. And from that safety perspective, I mean, Zipline is actually right now about, we have already achieved a level of safety that is 2x the goal that we had set by the end of this year. So like, again, safety and reliability is kind of everything. It's like the core of what we do. And it's a big part of like why I think cities are starting to understand like this is a incontrovertibly good thing.

1:53:52Is this a problem for Pizza Hut? Pizza Hut, famously, the buildings are huts. You can't land on the roof. How are we going to get pizzas from Pizza Hut in - The important questions. The important questions. But seriously, is there - I thought you were going to ask about the size of the pizza. Oh, no, no, no. Because actually, I want to know about laying on the roof. You are working on a pizza compatible. We just cut it in half. Oh, great. Calzones. The comeback of the Calzones. We already have the box designed for a couple different pizza partners. It works great. Should franchisees be thinking about a flat roof?

1:54:25Is that relevant? What does the actual infrastructure look like? Like as and like should restaurants and businesses be thinking about zipline compatibility as they roll out the next like version of their retail infrastructure or their residential plan or the real estate plan? Yeah, so it's pretty cool. I mean, first of all, one of the things we announced just in the last couple of months is something we call zipping points. And zipping points are really simple infrastructure. We pay for it. We can drop it in two hours. It does not have to be permitted. There's no construction. As long as we drop a zipping point with a restaurant or a retailer or a hospital or primary care facility, they're instantly enabled with zipline delivery.

1:55:03Got it. No power, no permitting, nothing. Do not have to do anything for the building. Interestingly, we are talking to like a lot of hotels, a lot of apartment buildings and a lot of even these restaurants as they because, for example, a lot of these big restaurant brands, they'll only have like two or three or four total designs of buildings that they'll build. And they are in fact now starting to take into account zipline and they're starting to design specifically for autonomous delivery You don't have to do that, but you can You know one big point I would make here is that like the thing that blows my mind I mean starting to see this like exponential takeoff from a demand perspective I would emphasize by the way we freaked out enough about demand four weeks ago that we turned off all Marketing all demand generation marketing in the company.

1:55:44We turned off all the in-app notifications We turned off all the field marketing that we were doing and And it basically made no difference. We continued to grow 25 or 30 percent week over week. I think that the reason this is happening in retrospect is kind of obvious, which is we're now starting to deliver to a lot of offices. We're delivering to a lot of apartment buildings. We're delivering to universities. I think the product is way more viral. Yeah, when you think about the virality of the product, I'm sure this is already happening or will happen, which is a fraternity. How can you not make an Instagram story about this?

1:56:16Yeah, my brain goes to some guys at a fraternity order a 30 rack. It's dropped into the backyard. And I don't know if you're doing that. You're probably not doing that yet. But that is a very viral moment that people will be taking videos of for a long time. You can see, I mean, tons of people are doing TikToks. And a lot of those TikToks are getting seen like seven, eight million times. We were actually initially asking customers to not put stuff on social just because we wanted a chance to do an early access program and kind of like, you know, iron out the kinks. But yeah, even people who aren't customers are sometimes like videoing their neighbor, getting a delivery and then putting it on TikTok and having it get seen millions of times.

1:56:58So it is cool. Amazing stuff. Always great to chat with you. Shout out to Danielle, whoever's running your comms program because your remote call-in setup, I think is the best of any guests. I mean, this is the manufacturing line. So this line is ultimately capable of producing about 55 ,000 aircraft a year. And that's scaling to that as fast as we possibly can. That's so many. We have to learn how to build drones in the U.S. Yeah. Yeah, I'm very excited about it. You're doing it. Thank you. We will talk to you soon. Have a great rest of your day. Always welcome here. Anytime. Talk soon, Kelly.

1:57:32Talk soon, you guys. All right. Cheers. Talk soon. Thanks. Up next, we have AIDSleep.com. Probably get a pot five ultra. I put on a clinic last night. Jordy Pele, my favorite soundboard cue. I got a 90. Oh, good for you. I got an 84. Roasted. 100 % quality. 88 % consistency. Six hours and 31 minutes slept. I'm on a run. I got a five-year warranty. 30-night wrist-free trial. Free returns. Free shipping. Head over to 8sleep.com. I'll be right back. And I will talk to Will. How are you doing? Hey, how's it going? Great to be back. You're good. Jordy just plays a sound cue and steps away. uh what's new in your world uh just uh yeah wait uh what's what's the latest yeah so about an hour ago we did a big launch of something we've been building for quite a while at prime and elect uh so for those who are not met me i'm will brown i work at prime and elect for a uh open source research company as well as a compute platform um and we released something today called an environment hub which is for rl environments as well as evals and uh it's something we've been kind of working towards for quite a while, but I think it's also something that like a lot of people are talking about, but also like wondering, like, what does this even mean?

1:58:46Yep. Um, like I, I saw like over the weekend, like Google trends for RL environments, like shut up. Um, and you guys had made some joke tweets about this recently that I thought were great about like, Ender, that's not an RL environment. And that's you place real Amazon orders. Um, uh, it, it perfectly encapsulates my level of understanding here. So I'd love to go deeper. It feels like a bull case for distributed infrastructure generally because we're maybe like it's maybe it's additive, maybe it's maybe it's replacing, but certainly less of the focus is on like get 100 ,000 H100s in one single cluster.

1:59:28There's more to be done in different places. But try and concretize it for me a little bit more like what is an RL environment? How are they being used? What are the trade-offs in terms of compute and design of the infrastructure that actually delivers some sort of valuable product at the end of the day? Sure, yeah. So I think we can get to the distributed aspect of it. just like first i think like an environment is essentially it's an eval like a lot of the things people make as like popular evals like sweet bench or like arcgi like these are environments yeah they're a thing where you have some input tasks you have some kind of harness and then you have a thing at the end that looks at what your model or agent in the harness does and then says a score at the end and so this is like the exact setup we use for evals but it's also what you use for rl training and so when people are talking about oh we're like going to make a bunch of rl environments What they really mean is they're making evals that are designed for agents to do some tasks interactively.

2:00:27Yeah. Is the Amazon order, like we saw that report that some folks are building like full digital replicas. I mean, I guess it's already digital, but like a simulated environment of Amazon.com or DoorDap. Yeah, totally. Are those kind of the power law opportunities that people are really focused on? What are some other examples? Are people building these environments for old enterprise software systems that people want agents to interact with? How wide is the scale of RL environments right now? Yeah, it's pretty huge actually. So I think there is these flagship websites sorts of things, but I think there's also a long tail of increasingly niche applications that go down towards these very fine-grained narrow things all the way up towards like broad domains where models really struggle because we don't have a good like interface for them like it's excel is one of it's like everyone uses excel but like no one is really like an lm power user in excel because it doesn't integrate very well um no we don't have the cursor for excel yet as a broadly deployed thing yeah i think it's easier to have an environment for a terminal or for a coding agent because the harness is a little simpler and so that's one sort of thing you'd want as like an environment and that's one of the sort of things that like the mechanisms of the world and imagine are like building these sorts of things.

2:01:44I'm sure that like the labs all have their idea of like what tasks their customers want models to be better at. They're currently not being used for as well as just like what sorts of products are people trying to build. So like you can build the greatest agent harness in the world, but like if your model hasn't been trained for it, like some models might just like not be very good because it's not very ergonomic for them. And so you can think of environments as like a way of taking a model, like having an interface that you really want your model to be great in, like a harness, and then just like a feedback loop that lets it get better at being a model or agent in that harness.

2:02:16Yeah. Help me understand the Excel case more specifically. Copilot exists there. We're hoping that Microsoft rebrands it as Clippy eventually, but there are also like seven different startups that are building a cursor for Excel. help me understand the context of verifiable rewards and what that would look like in Excel because if I'm trying to do an analysis, I'm trying to understand a company's financials, build a DCF, like there are templates, but then to actually get the correct valuation, that's something that like even if I have the perfect DCF and it tells me that Nvidia is worth five trillion, like I can't verify that against the market.

2:02:57I could wait a year, but the DCF could still be right and the market could be wrong. So like what does verifiable reward look like? What are people actually building towards in that environment you think? Right. Yeah. So the easiest recipe, which like we've done work on as well as I've seen a lot of papers about this. And it also just kind of like the obvious only real thing you can do is have a correct answer where it's like there is a gold standard, like someone made the CCF and we, the agent doesn't see the finished one sure and it's like new enough that it's out of the training distribution from pre-training like past six months is perfect and then you want to train your model to like be in the harness where it could make the dcf and then look at what it produces and then you have all these different things you can check of like did it get all these fields right yeah um and this those evaluations are a combination of like uh like vibe checks spot checks when humans are looking at it for like this is how kind of people have been doing historically you just like you build a thing You try it out, you look at it, you try to make a list of what's wrong.

2:03:55And that doesn't really scale. So to scale, you need to automate this evaluation trunk, where you want to have a thing that comes out like a DCF, a spreadsheet or a paper or a piece of code, and have a way of programmatically checking this. And this checking usually is going to involve LLMs in the loop, maybe fine tuned LLMs or customized LLMs for grading whether the thing has been done correctly. And then this like, you got to iterate on this meta process a bit to have a good grader, and see like does the result of your grader match your human vibe check, your spot check based on having some kind of experts in the loop.

2:04:28But what you then want to do is take that kind of evaluation process and freeze it as a kind of piece of code. And now you can plug this into your harness. So the harness plus some like input tasks of like do this for every company and your like gold standards for every company of like what a report should look like made by some analyst. You're grading against like the golden answer. And now you scale this up. What was the industry standard before you rolled this out? Would the actual AI researchers or AI engineers who were going to do reinforcement learning on a model in a particular domain actually design and implement the environment with the reward all internally and it didn't exist as an external function whatsoever?

2:05:12Was that the status quo? uh in terms of like open source infrastructure or like the in terms of like just broadly yeah very few people have been doing this the amount of people who have successfully trained a large-scale agent model with reinforcement learning is very small yeah it is not broadly deployed like the companies that are like on the periphery of maybe doing it are like curse red perplexity yeah so it's not a thing that's like yeah yeah yeah when i hear a lot of like we're an AI agent startup, what that really means is like maybe you're leveraging another agent or some sort of agentic API in that particular case.

2:05:48And then maybe they're using sure. Yeah, you might be, you might train your own. And maybe that's the, is there a significant cost to this? Like, does this affect, if I, if I just take the landscape of like everyone who's building cursor for X right now, that implies that at some point they, like, if we see this bifurcation and the market for Cursor for X really is highly fragmented and there's not just one foundation model company that just eats everything. Are you going to see small RL reinforcement learning environments with verifiable rewards in every single sub-niche of SaaS that these companies are going after?

2:06:26And would they have material costs to that? Is it expensive? I think you have to build the harness anyways. I think one of the hardest things to scale is building good evaluations criteria. If you can do the evaluation bit, depending on like how, so one thing is like the infrastructure problem of like doing this for like a frontier scale model is not broadly accessible. It's not the thing that you can just like put in money and it comes out. That's what we're working on building. But you can do it with like these like tiny models like the non mix. So kind of technical thing is like to serve models at scale, you really want them to be like large mixture of experts models that are efficient for inference.

2:07:05The current ecosystem for tooling for doing that yourself is not great. Most people are doing like Lama and Klan experiments, and these models are like, they're good models, kind of, but they're not really what you want to deploy for a serious application. Okay, so yeah, is that can you give me like a rough order magnitude for the for the level of cost that it would, it would, if I have a problem, a specific domain where I think I need to, uh, train an agent in an RL environment, I have the verifiable reward. My AI researcher has, you know, defined the harness and the reward. And I go to you to actually do, uh, the, the RL on top of maybe an open source model.

2:07:49Am I, am I talking about tens of millions of dollars of GPU costs, try and ground it for me a little bit more I would say this is like napkin math if you want to do a serious run that would meaningfully improve a model of the DeepSeek or Kimmy scale we're talking like hundreds of thousands for a big boy serious one you can do a lot for thousands in terms of raw compute when people were like I fine tuned Lama on this thing and now it now it does this funny thing or i fine-tune this image model on my face like those used to be like kind of like prosumer level projects now we're getting into the okay this is like a enterprise level effort but it's probably not gonna you don't need to call up softbank to get it done right i mean there's like a there's a whole spectrum of these things yeah of course that's why it's hard to like say like yeah you can like you can fine-tune llama on your laptop sure but you also you can't fine-tune like kimmy or deep seek the big ones on your laptop yep um And also it's like how long running is the task?

2:08:57How many samples do you need? How much do you really want to crank it? Like it's one of those things where you can kind of just like get more as you dump in more compute. But currently people are spending a very large fraction of their total compute costs on experimentation and rebuilding the same shit. Like there's not, it's not easy to do this stuff. Like getting your GPUs to work correctly, getting your like libraries to work correctly, deciding what hyperparameters to use. Like people spend, and this is one of the reasons that like the labs have needed so much to compute and why the expenses are so high is that like they have all these researchers and the researchers are all doing experiments and each of these experiments is like thousands of dollars of compute.

2:09:37And you multiply this over a year and a hundred researchers, that's a lot. Okay, I want to, sorry, you can finish, sorry. Oh, sure, yeah. Just that like, I think there's some of these pieces is that like once you can get the hard questions answered once, you don't need to keep redoing them for every new environment. You can kind of have some very cheap spot checks. Like a kind of one-shot eval, like how good is the model at the start, is kind of be a bellwether for like, is the running going to work? Okay, loosely related to somewhat of the concept of generalizability, I want your reaction to this Rune post.

2:10:11Rune said, my bar for AGI is an AI that can learn to run a gas station for a year without a team of scientists collecting the gas station data set. What do you think about that as a bar for AGI? We don't need these aural environments for specific tasks. What are your thoughts? I mean, one thing you might have that look like is the, like, I think that could be answered by an agent who realized it's not good at gas station stuff and figures out how to kind of create its own training environment. It's like, okay, I need to practice. The same way that like if you want to learn how to code, you got to go find stuff to practice on.

2:10:47Maybe that's one way of thinking about what that level of AGI means of like truly self-learning, like build your own tasks and curate your own tasks in a way that allows you to check whether you've been doing them correctly or not. Yeah. Yeah. It does feel like we're in this era where there are a few really obvious things that we want to RL on and get really good at. And you see that with the IMO and the math and the deep research reports. And they're very reliable. But then once you come up with some random task, then you need the gas station data set. And so, yeah, maybe the future is go and find those autonomously and set up the reward function train and then iterate and bake that in.

2:11:26Yeah, but I think one way to do this is just like the environment can be the thing you're going to serve. It's like cursor is an environment. Lovable is an environment. All these things, people are already kind of building things that could be environments if they connect the wiring correctly. But I think we're not quite there yet in terms of having this be a thing that people are scaling. And now every company wants to hire environment builders and RL people. And it's like, they're not all going to hire people for this. We need to have better ways of scaling this out to more people and infrastructure as a service.

2:12:03Yeah, that makes no sense. I wanted to ask about Dario's comments in his interview with John Collison, talking about, basically, I'll read the quote. He says, there's two different ways you could describe what's happening in the model business right now. So let's say in 2023, you train a model that costs 100 million, and then you deploy it in 2024, and it makes 200 million of revenue. Meanwhile, because of the scaling laws in 2024, you also train a model that costs 1 billion. And then in 2025, you get 2 billion of revenue from that 1 billion, and you've spent 10 billion to train the model. So if you look in a conventional way at the profit and loss of the company, you've lost$100 million in the first year,$800 million in the second, and$8 billion in the third year.

2:12:39So it looks like it's getting worse and worse. If you consider each model to be a company, the model that was trained in 2023 was profitable. You paid$100 million, then it made$200 million of revenue. There's some cost inference with the model, but let's just assume in this cartoonish example. And so Fintwit has just been, like, freaking out about this, like, being incredibly bearish. It's obviously a bubble. and just kind of worried you have exponential increases in costs and then unpredictable kind of like capital needs in the future. Why do you think they should stop freaking out given that you were just over at Goldman?

2:13:18More interestingly. But yeah, so like there's an element to the freaking out that's like kind of a good point, which is like you can't keep doubling every year. At some point, there's a plateau. And you can grow every year. You can still have exponential growth, but the percentages are not going to... There's only so many people to adopt AI products. And once they adopt it, if they were going to be spending more and more, it has to be delivering a very large multiplier of new value per year. And there's some laws of physics of how good models are going to get, how fast we can possibly make the chip.

2:13:53The chips are only going to scale so quickly. We can only build so many GPUs year after year. with the current rate of growth. And so I do think there will be kind of like the space we end up in where like stuff broadly works quite well. There are like a few big labs who have very good models that are like you can deploy them. And then there's also like this kind of Pareto curve of like how much can you spend? How fast does it need to be? How hands-on do you want to be for it? Like how narrow is your thing you want it to do? So if your thing is like, oh, I want to build a great browser agent, this is super broad.

2:14:32There's so many things you can do on a browser. If your thing is, I want to build an agent for writing Rust, this is much more narrow. And so if you're making the product for Rust developers, like the cursor, for example, you can make a model that's just as good as Quad4 Opus, that's much smaller and much cheaper, most likely, because the focus is more narrow. And so this is kind of how I think about generalizability is like you can spend more on compute. You can make the model bigger. These are all different axes to spend. And you can also make it like more narrow and go deeper on one thing. And so we're going to have a lot of knobs to turn, I think.

2:15:08Yeah. Are you broadly long RL environments, short, normal pre-training data? Do you have any like extra color on the relative value or like where the low-hanging fruit is across those two? I'm long both. But I think they're going to kind of turn into the same thing. Like one thing, for example, that people have been doing for like pre-training recently is that people are just using like DeepSeq R1 and generating tons of like reasoning samples from it. And then mixing those in at the end of pre-training and calling it mid-training. And it's like, in some sense, this is RL because it's just taking the juice from doing RL on like DeepSeq D3, but it's also pre-training.

2:15:50And so I think one thing that we're excited about with environments and scaling these is like you can get a lot of good data out of these by you have a like a task set you have the ability to generate agent data inside of this task set you have a filter for throwing out the bad stuff and keeping the good stuff and so can you just like do trillions of tokens this and put it into pre-training why not um do you have insight into what's happening in the video or image world equivalent of RL environments, verifiable rewards, et cetera. Like, uh, I, I was just shocked by the level of quality of text in images and chat GPT.

2:16:31Uh, we're seeing the same thing with nano banana from Gemini. Uh, and it felt like, it felt like images and chat GPT particularly was like uniquely good at studio Ghibli and uniquely good at text. And that felt like I was, I was almost like putting on my tinfoil hat and saying like they're using photoshop here and they're doing two different layers or something or like there's something else going on here that's just like a real one really good model like it feels spiky it feels particularly good at text and particularly good at cartoons but it hadn't gotten like way way better at you know super photo real imagery or whatever um so do you have a do you have any view on how you translate all the stuff that's happening in the agents and text-based LLM world into sort of the diffusion and image or video world.

2:17:21Sure, yeah. So like I am not a diffusion network or anything, but like I imagine it's a mix of essentially these kind of environments where you have some greater, as well as like good old-fashioned RLHF, like the stuff people were doing for like Instruct GPT, JAT-TPT era, where it's like you have like upvote, downvote, and then you're training a reward model to like check these. And I think the image domain is easier to kind of spot errors, whether it's a human or an automatic grader. Like if text is wrong, you can like do OCR on the image, get the text back out and see, is it the right image from the prompt?

2:17:54And so you it's, this is like pretty easy to verify versus like, if a chat GPT answer is like slop, how do you verify that it's slop? Like what's the algorithm that checks if it's slop or not? Like that's pretty hard. And so in some sense, like the image models were just like for a while they were bad at things that we could very easily measure um and it's getting it's gotten harder to measure the things lms are bad at much faster um and so i think some of this is like the old-fashioned tricks that made chat bt original version as good as it was being applied to the image domain for kind of patching up the the obvious fixes uh give me your hot take how much uh how much do you think meta is paying mid-journey with that new deal they announced?

2:18:36That's a good question. Do you think it's nine figures a year? Hopefully. It feels like it would be more maybe. I don't know. They're spending crazy amounts on everything. That's what I was saying. It makes sense for MidJourney to want to have a low-key announcement about it, but there's another team that would have been like, we just signed a$500 million a year, 10-year deal or something. obscene that looks more like an acquisition even though clearly from a dollar value standpoint, even though clearly they're going to continue to operate independently. Yeah, I would kind of think of it as like Mid Journey's version of doing an API business.

2:19:20So like it could just be based on usage. I'm sure there's a big picture component of it, but like think of like what you can kind of do the napkin math by saying like, okay, what are other image providers like charging for API usage, like Val or like, replicate or like all these other services, like you can kind of map and map the cost or of an image or back it out from the during subscription prices, and how much they give you and then say like, Okay, how many of these people going to be doing, if they're doing like generative ads for Instagram, that's a lot if they're doing like, every person has the ability to apply it to their Instagram posts or their stories.

2:19:56That's a lot. And so you can, the numbers can get crazy pretty quickly and i think it does kind of depend on like like if it's just ads or if it's just to like certain customers or it's not like deeply integrated then it like the scale is not as crazy as it it's like no you're really getting like mid-journey stuff everywhere for everyone instantly at volume yeah i was uh i was thinking when uh studio ghibli moment happened uh that the response for Meta should be to pre-render or pre-generate a Studio Ghibli of every single person on Instagram's profile photo as a Ghibli. Because that's basically what people were doing in ChatGPT.

2:20:40Like pre-generate that. And then just when you open Instagram, it just says, hey, we did this for you. Do you want to share? And that would probably be like the biggest day of usage on Instagram. But it would also probably bankrupt the company because it'd probably be Like$50 billion worth of inference or something like that. I don't know. It's a lot of inference. It's a lot. And those images are expensive. Got some breaking news. Breaking news. NVIDIA beat on revenue and earnings per share, but is down 5%. Wait, why? They didn't beat hard enough. They didn't beat hard enough. Can we ring the gong for the beat?

2:21:12I mean, let's ring it. Yeah, 4 and 1 % respectively. NVIDIA is both a competitor and a supplier partner, right, to you? Yeah, I mean, we resell the GPU. Yeah, exactly. So we're big NVIDIA fans. So you're rooting for them. That's great. Yeah. And I think NVIDIA has been very friendly to the ecosystem of players. NVIDIA is not trying to be the single... You can even buy GPUs on their website. They don't sell GPUs to people. They sell them to data centers and big companies. Yeah. And I know that there was that news that they were coming for the NeoClouds with DJX Lepton. And it felt like they were dipping their toe in that area.

2:21:56But it didn't feel like this was existential for anyone else in the ecosystem. It felt somewhat additive. I don't know if you have a take. Oh, yeah. So I mean, I think it's also like a different like, they're gonna kind of offer that as like a very premium, like white glove sort of thing to certain enterprises. Sure. I think we are much more on the end of like, get all the data centers we can find like partner with every NeoCloud, and have like, really cheap pricing and then like build features on top. Yeah. Um, where part of this is like doing like core research. So like, we like have friends at NVIDIA where we were like, we talked to them about like, they're, they release a lot of cool, like open source, um, like research stuff.

2:22:34And so like, that is like very in spirit of like the sorts of things we are like aiming to do. And I think we're like friendly. Yeah. Uh, are there any RL tasks that you think are like truly intractable? Something that like is, is fundamental to humans, creativity or comedy or something like what, What's the Mount Everest of developing an RL environment around? I think the hardest is stuff that, and a friend of mine was tweeting about this earlier today, actually. One of our collaborators from Evels is doing some cool open source Evels work for computer use. But things that need a human in the loop to have a fine, green, and accurate simulation.

2:23:13Like stuff that, if models are not good enough at replicating human behavior yet, and that human behavior is key to the environment, then you're not going to have an environment that faithfully captures the task because there's a kind of chicken and egg problem. And so like Twitch streamer was one example of like, like having a model that can like be good in Twitch chat kind of requires like an accurate sim of Twitch chat. And it's like, how do you build that or like a physical streamer, like there's a, this real time interaction problem that you need to kind of solve before you can get started because the, But the scale at which these things move, you can do it if an LLM is simulating the user.

2:23:50You can't do it if you need a real human. Yeah, I feel like the time horizon feels really, really tricky to simulate when you think about the impact on a life of a certain behavior. I mean, we struggle with this with drug development, understanding, like, does something in childhood affect you as a retired person? And it's like, okay, now you have to create a simulation of the entire human body to understand that if you did this thing at 18, it's going to cause you to, you know, have your knee blow out when you're 65 or something. Like that feels much harder to simulate one shot. But maybe we'll get there.

2:24:26Who knows? Yeah, at some point you just got to let the error of time run forward and see what happens. Exactly. Well, thank you for hopping on. Always great to see you. Awesome. Congratulations. Congrats on the launch. Check out the Prime and Elect Environments Hub. We are live now. We're on Twitter. X. See you around. Send it, chat. See you. See you guys. Talk to you later. Bye. And yes, we have the NVIDIA market chart here down 2.75%. Is that what I'm seeing here? Correct? Let's see how NVIDIA is doing as well. But high expectations. But after hours, it's down 1.42%. It's bouncing back right now.

2:25:05The company's trading at$179 a share. Was it$181 dollars a share. Something tells me, Jensen's going to shrug this off. If you look at the six months, the stock has just been smoothly climbing upwards. Anyway, we have our next guest, Julia Steinberg in the studio. Welcome. What's happening? How you doing? Welcome back. I am doing well. I was thinking about wearing like a little hat that said Chinese Communist Party on it to, you know, establish what I was talking about. But unfortunately, I came up with this idea 20 minutes ago. And because I am dealing with California infrastructure and not China infrastructure, I could not get it delivered to my house in time.

2:25:48Yeah. Are you driving from LA to SF or vice versa? Where are you going? Yeah. So this may be a bit TMI for the good listeners of TVPN, but I was rear ended a few weeks ago. So my car has since been fixed, but I must drive it up to San Francisco. and I was actually sort of thinking about the drive north it's a lot of what Dan Wayne talks about in breakneck is sort of like look China's able to build these good things and I was thinking about I'm like why do we not have a functioning highway for highway one yeah why like PCH Pacific Coast Highway it's one of the most iconic American landmarks one of my favorites personally it's ridiculous that it shouldn't work and I tweeted I just sort of out of frustration like I'm really frustrated that I have to, you know, do this ugly, annoying drive where my car is going to smell like cow shit for six hours rather than see the beauty of the Pacific Ocean.

2:26:44That was my favorite thing as a kid. My parents, we'd be on a road trip. My parents would say like, okay, we're going through the cow zone. We're that big cattle ranch. Yeah, you have to turn on the recirculate. Yeah, recirculate. I would just be like, I'm hitting the window down. If you were my child, you would have been booted out of the car window at that point. I was surprised. So PCH between Malibu and L.A. was shut down after the fires for a while. And then it was open to like specific people based on your address. And then it got actually open pretty quickly. and I think that was probably Newsom knowing that like this stretch of PCH is like critical to my popularity in the state because there's just a lot I mean for for no reason other than celebrities drive on that route and if they're and if he doesn't get it open so it was open ahead of schedule yeah he was taking a victory lap of like but one is not really critical to many people's commutes it's more of this especially up by it's big sir that's shut down right yeah big sir so i can drive from san francisco or however far north on the one to big sir but i can't sort of cambria region is cut off that sort of fell into the ocean a lot of people on twitter were sort of i wasn't expecting the tweet to blow up and people on twitter were like how can this like girl expect mother nature to want this freeway here i was like this is absurd like we've sent men to the moon we can't have a bridge like i'm not a structural engineer but dan wang in his book breakneck he was like oh yeah the chinese government has built x amount of like the bridges like the top 10 some of the biggest bridges in the world in this like record short amount of time and it's it's frustrating to me that we can't do that we somehow is admitting that like as a civilization we're just gonna let the mudslides win yeah here's the solution if you want one the highway one to open you got to put a harris ranch out there because that's the highlight of i5 you stop for the a5 wagyu ribeye you get a big steak at harris ranch this is the highlight of my trips up and down harris ranch is a really smelly rant that's the really smelly one but i didn't realize you could pull over and get a meal oh yeah i didn't realize that either oh well you're not doing the five correctly then yeah when you're driving from la to san francisco you uh run out of gas halfway because you're driving a terrible car this is my experience in like 2012 um and you always stop at harris ranch you get a big steak and it's fantastic yeah i think most people are flying by the other alternative that is actually scenic and nice is there even though we don't have a high speed train we do have a train that goes it takes 24 hours like it just takes skill issue get a book i thought you were i thought you were into books You're the GM of books.

2:29:36I do like books. You're the general manager of books. But I like being able to read books while I'm not in a moving vehicle. Oh, it's so nice. This is one of my more controversial book takes is I like being, sitting still. Okay. When I'm reading a book, planes are fine. I can read many books on planes. I've read many books on planes. Okay. Well, speaking of books, take us through your review of Dan Wang's book. Yeah. So I sort of was introduced to Dan Wang about a year ago. I was actually on the five listening to one of his articles put through some voice reader technology how technology grows which is how I first got acquainted with the ideas of process knowledge which is basically that engineering requires a great deal of like verbal transmission from one person to another or sort of one like elder mentor to a mentee in the United States we've like lost a good deal of that process knowledge in how to make things because we've offshored it to China.

2:30:33And Dan Wang really in breakneck does a really fantastic job of describing how these communities of engineering have maintained this process knowledge, this ability to build things, which in the United States, we've just voluntarily shrugged off and lost. He, I would say like the big idea, if you're going into the Atlantic, this is the excerpt they chose was that rather than thinking of the United States and China as a capital society versus a communist one or a socialist one, we should think of the United States as a lawyerly society and China as an engineering society. So in the United States, a lot of our elites go to top law schools, you know, Supreme Court clerkship, one of the most prestigious things you can get.

2:31:15Maybe even Supreme Court is one of the most prestigious positions you can get. Whereas in China, the Politburo is made out of engineers, first and foremost. I think Xi Jinping doesn't have that big of an engineering background, but pretty much most people in the Politburo do. And in the United States, when we focus around rules and regulations, what's getting us from point A and point B is like, well, what is this going to look like? Is this going to violate people's rights? Can we do this? In China, it's get from point A to point B. It's like, okay, how do we do this? And building something from the ground up.

2:31:46Yeah. I was listening to Casey Hanmer on DoorCash and he was talking about how he's trying to install solar panels in the desert and he needs to get some environmental report that says that like if he puts the solar panel over it it'll kill this tuft of grass that like a bird might come and eat meanwhile like if he was building like a chemical plant it would be like fine because like like the chemical lobbies have like lobbied to like get it make it easy and so it's just like nonsense but my my question is um the the the the engineering uh what what what's the phrase he uses engineer engineering society engineering society and lawyerly society right um engineering empire or something uh so i i understand that and it's very it's very alluring like i like engineering more than i like lawyer lawyers i suppose um but my prior my prior is that america is number one and america is the best so uh should we really try and change horses in the middle of a stream here maybe the beauty of america is that we have so many lawyers uh keeping us great.

2:32:49What do you think? Well, my dad, who's probably watching right now is a lawyer. So shout out to the lawyers keeping America great. I guess the bigger question is like, is there a world where we say, okay, let's try and pivot from lawyerly society to more engineering focused. We get halfway there and we actually are worse off because you want to just like lawyer max or engineer max. And if you're halfway in between, it's just a nightmare. I How I view it is that the United States is very good at engineering maxing in the private market, whereas China is very good at engineering maxing just having the state do it.

2:33:26You know, Chinese Communist Party. Well, both. Both. But the state capacity for engineering projects is... The state capacity is a lot bigger. But it's also like, this is my biggest issue with breakneck, is that it was just like way too Keynesian. It was like, oh, it's so great that the government is, you know, basically getting these people in the middle of nowhere in China to dig holes. And by dig holes, I'm sort of exaggerating a bit. But it's taking these bridges that no one is going to use or these airports in the middle of nowhere that have fewer than half a dozen flights a week. Like, I just don't think that's a good use of market resources.

2:33:59It's a sort of interesting question that Wang raises, though I don't think he says it straight up. It's like, is capitalism even a useful sort of metric in the 21st century? And I would say absolutely. But we need to prove that in the United States by doubling down on free markets. free markets could bring us great things like a beautiful bridge uh that would make pch work i i'm a huge believer in market incentives i think the issue that wang sort of highlights though when he talks about si jimping not wanting like a services economy is just that in the united states maybe the culture is a bit broken where our best minds want to do i don't know like B2B SaaS rather than building bridges or building nuclear power plants or other infrastructure.

2:34:45I do think this sort of culture is changing a bit. We're returning a bit to wanting physical manifestations of American excellence. That's not just my app is like 0.02 % more productive than your app in delivering an AI girlfriend. But I think that overall, all what Wang talks about in China is sort of engineering society. It conflates two things. It conflates engineering and it conflates culture. And when the government is in charge of culture and they can say, everyone from this district should want to be a guitar manufacturer, which is true in one of the districts in the book that he talks about, which makes guitars, which is pretty random considering that there's not really a historic guitar industry in this city, in Guizhou.

2:35:30It's just something that they sort of was like, we're just going to make guitars here. And it happened and they got very good at it. But with the United States, culture is a lot more organic. It's harder for the United States to say, we're going to make this value really well adopted. And this is something that's sort of bad about China too. And Dan Wang talked a lot about the one child policy in China. He talked a lot about zero COVID in China, where people - the multiple children policy. Have three kids, please. And Dan rejected that roundly. I got destroyed. I like the three kids policy. I think it's a good policy.

2:36:11I think they have the money. There's incentives. And if you really, really tilt the playing field in the favor of having more kids, people will have more kids. I agree. And it's like talking, of course, I have to bring up Tavis, the engagement that happened yesterday. I was talking to my boyfriend at the end of the day and I was like, Jake, I have to tell you something but you're not going to be interested in this at all. And he was like, oh my God, what did she do this time? Did she get another car accident? And I was like, Travis and Taylor got engaged. He's like, I don't know who those people are and also I'm really not interested in that.

2:36:43He doesn't read the Wall Street Journal? Come on, you guys need to write something about a trend piece from America. Taylor wrote a 2014 op-ed about Spotify. Also, I want to know what this engagement means for America. from arena more people are going to be having kids it's one of the weird things it's like a Taylor Swift baby boom China it's not who's a Chinese pop star who's going to get pregnant from a football player and have kids I don't know that's a good question I do think there should be a Taylor Swift Orchid collab you know she's on the older side I hope she has many kids Orchid could be a sponsor you know there and having D1 athlete that's also world famous superstar kids that's true i mean they're they're certainly they don't shy away from uh the economic opportunities associated with their um with their announcements that's for sure no but i but i do really think that the the tavis sort of engagement announcement the what the royal wedding that will happen in a few years like i unironically do think this will cause a baby boom um because people are very mimetic women are very mimetic and if number one pop star has kids I found the answer.

2:37:55Jackie Chan, born in Hong Kong, has two children. That's the answer. We just need to promote Jackie Chan's children in China. That will cause the boom. Well, maybe the thing with Nepo babies, the sort of weird take on Nepo babies is that we need to have more Nepo babies and just rip the shit out of them. So it's like your kids, if you are successful, your kids will be successful too. Because the line about kids now is like, oh, if you have kids, especially if you have kids young, they'll draw away from your success. But if I'm like, oh, my child is heritably going to be like a kung fu superstar, because I am a kung fu superstar, I would totally have kids for someone to do that.

2:38:34I mean, that certainly happens in Hollywood. There's a ton of celebrities that have kids. I mean, John David Washington is Danzel's son, played football and then was in Ballers, I believe, and in Tenet. I have a sort of funny story about that. I grew up in L.A. and I was on the debate team at my high school. And there is a lot of sort of overlap between girls who wanted to do the debate team and girls who wanted to do the school play. And this one girl whose parents were both very, very famous, like super A-plus list actors, was deciding between doing the debate team and doing the school play.

2:39:06And our debate coach, he said to me, he's like, well, it's usually pretty easy to convince them that they're not going to have a famous career as an actor and they're a lot more likely to become a successful lawyer. But if both of your parents have stars in the Hollywood Walk of Fame, then that's a lot less likely. makes sense sorry i'm prepping a chart george do you have anything else chart for what our next guest oh who's in the restream waiting room uh we've been julia it's always great to see you i've got to come on more often yes we should have scheduled more time but we have someone else in the waiting room so next time i expect a custom chart to be pulled up next time you're in la come do the show in person i that i said i said i'm here okay fantastic let's do it it's the car drive over to the team in Ultrasound.

2:39:48Glad you're okay after the accident. Thank you. Appreciate it. Always great to catch up. And we will lobby in California to fix the one so we can drive scenically in the future. Thank you so much for hopping on. Our next guest is in the Restream waiting room. Olivia Moore. From Andrewson Horowitz. How are you doing? What's happening? It's great to see you guys. Thanks for having me. Great to see you. I was just sharing with the team the news today, but you break it down for us first. What are you announcing today? Yeah. So we're announcing what we call the consumer AI top 100, which is where we literally, thank you.

2:40:21I appreciate when other people are this excited about data as I am. We live for market maps. Yes. Yes. This one has, I would say it's market map adjacent in that we every single website and every single mobile app around the world in descending order of traffic. and then we pick out the top 50 in each that are AI-native companies to just get a look at what consumers are actually using. I consider myself a bit of a list connoisseur myself. Yes. I enjoy making lists from time to time. I'm sure you're going to get yourself in hot water with this list. People are going to say, take me off this list.

2:40:55I want to be under the radar. Put me on this list. I want – I'm doing better than that other company on here. Talk about the methodology. How confident are you? Is there an element of vibe and curation? Is this like the New York Times bestseller list? No, there's no elements of Vibes. It's all data. For the web list, we have a provider called SimilarWeb, and we literally just look at monthly number of visits. We pick the first 50 that are AI. And then on the mobile list, we do SensorTower, and we pick the top 50 by monthly active users. But we do get lots of fun texts, emails, inquiries of people either wondering about their rankings or wanting to be on the list.

2:41:35So I'm happy to inform everyone that it is completely unbiased. There are some things on here that stand out to me as completely unsurprising. ChatGPT, number one. Gemini, number two. DeepSeat, Grok, up there. That seems obvious to me. Stuff that's standing out are specifically like - Janitor AI. Like hyper-specific tools, like remove background, remove BG. I've actually used that tool. Janitor AI is number eight. Yeah, what else stuck out to you was surprising? What did you learn? Yeah. Every time there's apps that are kind of like literally single purpose, remove.bg is a great example. It was a company that was acquired by Canva, but grew to like 32 monthly active users, literally by being the best products to upload a photo and get the background removed from it.

2:42:23And it's a catchy URL that people remember. So it keeps getting usage. The janitor one is a good call out too, because one of the surprises for me list after list is how many companion products make the list. Like it's always a dozen or so each time. Janitor is one of them and one of the highest ranking ones behind character. So it doesn't seem to be. It looks very, uh, grok XAI. I would have thought Janitor would be like back office B2B. Like I thought that would be something to clean up my data warehouse. Uh, but you're telling me that it is in fact a companion of some sort. Interesting. I also see Spicychat.ai, that sounds like something we shouldn't pull up on this stream.

2:43:04Cushion.ai has a heart there that seems a little suspicious. Juicy chat. There are a lot of these use cases there. Is this a new thing? Is this the first time you've done the top 100? What are the biggest movers that you're noticing or that you expect? Yeah, this is the fifth list. We do it every six months. We've done it for about two years now. We started it six months after ChatGPT came out. Interestingly, there's 14 companies that have made every single list. Most of them are in kind of creative tools. So think like MidJourney and Eleven Labs. Quite a few of them are general LLM players like ChatGPT, Perplexity, Poe.

2:43:45One of the big, I say surprise, but probably not a surprise to anyone who's been on Twitter is like Vibe Coding completely exploded on the list this time. So Lovable. wasn't even on the last list. And now they're like number 23. Replit's also on there. It's just below the cutoff. Cursor is on there as well. And it's interesting because like you would think that the developer market is much smaller than the consumer market. So for developer facing products to make an app like this is pretty impressive. And it shows that like a significant majority of developers are using them. Well, I think the way that I look at that market is there's so many people in the world that have had an idea for an application without the engineering co-founder like a cto and so that's just like decade decade plus now of like pent-up demand for people that had an idea for an app and then all these companies whether it's lovable or replet or things like that have people on tiktok saying like you can make an app now or they're running ads that say like, you can make an app, right?

2:44:51And so that's just like a flood of demand for people. Yeah, non-developers becoming developers for the first time. How are you thinking about Meta's strategy? I noticed Meta AI ranked 46th. That feels pretty low compared to the amount of effort Mark Zuckerberg is putting into that project, although it's early days. Also, the Meta AI app, I went to go test it. I realized I had it downloaded because I had bought a pair of Meta Ray-Bans, and they kind of iterated on that product to turn into meta AI. But LLMs are vended into Instagram and yet Instagram is not on here. Do you think that you'll see more companies that are in the AI bolt-on narrative era of their business kind of jump on here?

2:45:37Do you want to keep them separate? What is your thinking about all of that? Yeah, meta is really interesting. it's funny the the larger tech companies i think have really uh stepped up in the past year or so on the ai front like google is probably the best example of this um term and i was number two on the list huge they also had the ai studio on the list notebook lm on the list google labs which is where you get vo oh wow they have four four companies yes four four separate entities on the list which were all ranked separately with their own traffic yeah grok had a big debut on the list from X, formerly Twitter.

2:46:14Meta, I think, has struggled a little bit, both on web and then on mobile. I don't know if you guys remember the big scandal where they realized that a bunch of the users were accidentally posting their very private posts to the public feed. And so usage fell off a little bit. They didn't even make the mobile list. But to me, it's like if you guys have tried the Meta AI products, they're a little bit underwhelming. The quality of the images generated, They did the weird celebrity companion thing. So now that they're working with MidJourney, I would expect to see them make some improvements here, but it might take a little while.

2:46:48Yeah, even the MidJourney thing, it just feels so natural in the other apps. And if you have a billion people opening one app every day, like where Meta has been successful has been put stories into Instagram, put videos into Instagram. They used to launch sidecar apps. Like Facebook had an app called Facebook Camera that was supposed to compete with Instagram. and then they were like we gotta shell out for you think do you think midjourney's ranking gets hurt because people still use it in discord or is it a shift because it feels low at 28 sure sure yeah yeah yeah just recently kind of rolled out the web um i think that's true for both midjourney and suno and a couple of other the creative tool apps that also have discords it's funny because in the first iteration of this list like no one had a website everyone had a discord and now a A lot of those companies are really trying to shift their traffic over.

2:47:39That's interesting. Yeah, I bet you could do a different list of just like the top discords because all of that's very quantitative and available and even relies on even further data. Jordi, do you have anything else you want to take through this? I'm wondering if I can continue. Go for it. I'm just interested in understanding your predictions for what the next version might look like. What are you tracking on the earlier stage side that isn't quite ready for breakout adoption as measured by unique monthly visits, but we might see an uptick of in the next year or two on the consumer AI top 100?

2:48:17Yeah, I'm definitely expected to see more in prosumer and productivity. I feel like both the models are just now reliable enough to actually do work for you or getting there. And they're also agentic enough to like Perplexity Comic can draft an email and put it in your Gmail drafts folder, which is amazing. yeah um so we saw a couple companies like manis make the list this time around that thesis but i'd expect to see things like fixer or serif and others in that category gen spark is another big one that i would not be surprised to see on the next list the other what's what's been happening to all the companies that actually were gpt rappers where they were for a while you could build a business to a few million of ARR, probably more just basically.

2:49:06I remember one just advertising against chat GPT, uh, talk to PDF. And well, yeah, there's those, but I'm saying there was other ones that would be called like open chat and it would be like, sure. Yeah. Yeah. When you would search chat GPT and they would have in-app subscriptions and we're just literal rappers. Yeah. That was a loophole that was open on mobile. So we we've only done the mobile list, I think four times now and it was complete chaos the first three times because the app store was not policing what we call fleeceware which was like all the developers mostly abroad that were just kind of ripping off the free version of chat gbt putting it in an app and acting like it was the same thing you got if you paid for chat gbt premium and they were called things like chat and ask ai or chat gbt or chat gpt or something but they finally cracked down on that over the last six months so now we have a mobile list that's like yeah maybe more representative of true products but it's all like beauty cam filters now and scan your homework and get answers oh really what's your read on uh how apple's done from a curation standpoint there's like a lawsuit happening right now between xai yeah and opening eye and apple i mean this is probably gonna get printed out and blown up on a big chart in court soon because grok looks great here um but yeah I mean, I'd love to know your thoughts on how Apple's curating the App Store.

2:50:32Apple is a little bit struggling overall with AI. Like we've even seen this with Siri and the fact that they're now potentially going to turn to Gemini and Google to fix it for them, which would be kind of unheard of. I think they do an OK, but not a fantastic job with curation. It's definitely not like a pay to play situation from what I've heard. But I also don't think that they're necessarily kind of at the cutting edge of what's truly the best products that are coming out next that people should be spending time on. Yeah. They also - They should hire you. They should. I mean, they also obviously prioritize, they editorialize in what they recommend.

2:51:08And so I would be surprised if they're pushing companionship anytime soon, since the lineage going back to Steve Jobs was, this is a very clean productivity, creativity. Well, and Apple - I would imagine gets promoted. Mid-Journey, I would imagine gets a promotion. Apple comms is going to be every day there's an article in the New York Times. about how somebody had something tragic happen to them while getting advice from a model. And so Apple Coms doesn't want to be like, we recommended, you recommended Companions, and the Companions went haywire. It's just very... Anyway, we got to hop on with our next guest.

2:51:39Thank you so much for joining the show. Thank you for joining. Thanks for having me. This was so fun. Thank you for list maxing. Thank you for list maxing. It doesn't get said enough. Thank you for reading it. Let's hit the gong for the big list. We hear the gong is too loud. Too loud. Sorry for your ears, everyone, but great to catch up. Thanks for hopping on. We'll talk to you soon. We'll talk soon. Bye. Up next, we have Flo from lindy.ai. Big announcement today. We will bring him in from the Restream waiting room while we are waiting for him to join. Let me tell you about adquick.com, out-of-home advertising made easy and measurable.

2:52:14Say goodbye to the headaches of out-of-home advertising. Only adquick combines technology, out-of-home expertise, and data to enable efficient, seamless ad buying across the globe. Flo, how are you doing? Sorry for keeping you waiting. what's new in your world? You're good. Don't worry. I'm doing great. You know, it's a good day to build B2B SaaS. Yes. Let's go. Let's hear it. You're B2B SaaS. Finishing out the show with some just amazing news. Tell me more. Yeah. We just announced a Vibe Coder today. I just saw Olivia's appearance here. She mentioned Lovable and Reklit and a few of those. They've been blowing up.

2:52:48What we announced today is a Vibe Coder that can test its own work. Okay. And it's kind of insane to say it because you would assume like, of course you can, but actually none of those products check their work. Like imagine a software engineer that's just like, here, and it's like, it doesn't work. Did you even test it? And it's like, oh, no, I didn't do that. So basically these VibeCodes rely on you to test the work yourself and then like work with them to debug it. And it's like it ends up being a lot of back and forth and painful work. are you talking about generating a test suite in code or spinning up a web browser interacting with the Vibe coded product it's the latter thanks for asking is it okay if I show my screen you can share your screen but we are live so anything you share will be shared with the internet burned into the training data of GBT12 yeah Vibe code us something right now.

2:53:42Let's see. Live demos. They never go wrong. I'm not going to do a live, live, live demo because it does take three minutes or something, so it'd be a bit boring. Wow. Singularity delayed. The example. Built a Scrabble checker. Okay. And this is when we realized it's actually an emergent behavior that comes out of this agent that we gave a computer to. It's like, all right, I built this Scrabble wheel checker for you. I'm going to open it. Okay. I'm going to type a wheel that exists like thunder. And yeah, it it tells me it's valid. Now I'm gonna type a wheel that doesn't exist, like ZZQXW and oops, it tells me it's valid as well.

2:54:17I know this might be an issue. It's showing ZZQXW as valid even though it should be valid. Let me fix the API to improve the validation. Fix the API, try again, doesn't work. So that's, it does have its own computer, its own browser and it sees the work and can click around and actually test its work. Yeah, very cool. What do you think the landing, the landing, not landing page, but like, what do they call it? Like the landing market, the first, the first like product market fit moment. Do you have any glimmers of hope? Like, is this going to be used by restaurants who want to spin up a landing page?

2:54:53That was kind of like the previous boom of, it wasn't vibe coding, but when you saw folks stand up landing pages with, you know, Squarespace or any of those Weebly, like there was a whole boom in that. It feels like there's another boom in in vibe coding right now uh do you have expectations about where the sweet spot is because i i would imagine that people aren't going to build like you know they're they're not going to go out and raise a hundred million dollars for a company that's built on the back of a vibe coded solution it's more of like a prototyping tool right now but then for small businesses this could be something that they really rely on for a long time like talk to me about the shape of the early customer i think that's actually the perception of trying to fight.

2:55:33I think the reason why Vive coding so far has been limited to these landing pages is because the tools have been so limited. And landing pages are just like the fastest dumbest thing you can build. Same for this perception. They're like, oh, it's like a cute prototyping tool, but you're not actually going to build your company on this. The thing that is unlocked by this new paradigm of something as simple as testing your work. If you can't test your work, you can only build very, very simple stuff. If you test your work, we're seeing instances of like this running for hours and building like one of the craziest examples we've built internally.

2:56:06We put it like Lindy B &B. So it's like an Airbnb clone. And it's functional. Like you can select things, you can book them, you can put your place up for rent. It's functional. You can sign in, you can sign up. It's got Google Us, it's got all of that stuff. So, you know, we're actually seeing a lot of, so yes, we're seeing the landing pages, but now we're also seeing a lot of like internal tools. So you're almost getting into like retool territory. Sure. Because people are like, I'm just going to hook it up to my API and I'm going to build like ROI calculators and like various internal tools for the team in like a couple of minutes.

2:56:39Yeah. Do you think of this as like a separate product or does this like kind of sit on top of everything that you've already built? And can you, is there a way to like hook in the traditional like Lindy's into what you're vibe coding? No, it's super integrated with the rest of the product that we built, which is agents, obviously. And so that's the whole value prop. It's like, hey, building your website is step one. Once you've done that, you actually have to build the rest of your business. I joke, building your website is sort of the equivalent of getting your business cards. Like, okay, you did that.

2:57:11Now you need to build a business. So now that you built your website, Lindy can market it for you, find customers, do the entire customer support for you fully autonomously. I have wanted to actually launch Lindy B &B and have it fully, because I've got enough of one business, I don't want to run like Lindy B &B, but have it fully managed by agents, you know, and see how far we can go. Yeah, that's great. Yeah, keep sharing demos. I'm excited to check them out. Anything else, Jordy? Yeah, just, I'm interested, you know, clearly, I think it's almost contrarian at this point to launch another prompt to app builder, right?

2:57:47and you're insanely intentional about everything that you do. Do you think that the category actually needs a bunch more people to try to be experimenting here, right? You see some of these revenue ramps. I think a lot of entrepreneurs would assume that the market is run away. I should focus on something else, but you've clearly made a different decision. From the traction we're saying there is plenty of room in this market, I think it's like the very, very, very beginnings of this market. as crazy as that sounds. Also, and we've not told that story yet to anyone, the way this came about was like, we actually built the V1 of this by accident.

2:58:27So we had this release that I went on the show a couple of weeks ago where we were like, hey, we gave a computer to Lindy. And then we realized after we gave it to computers, like, oh, in fact, you can build websites and you can QA them. Because that's what happens once you have a computer, you can do everything. So then we basically packaged it up and released it as a Lindy build and improved it, obviously. But I think it's basically an emergent capability of the platform. And the reason why we've decided to release it is like, well, one, we were like, fuck, it's already working. And two, we then started to look into the other products and we were like, I can't believe none of them checks their own work.

2:59:00Like this is actually done. That's like a massive deal. And when we A, B tested, like, OK, same prompt on like Lovable, Replit, Lindy. This Scrabble example, for example, every Vibe coder out there is a simple example. it's like a very mini app. Every single Vibe Coder got it wrong. Every single one. Including Lindy. Like it got it wrong first shot as you saw. And then the difference is like Lindy then tried it on work, saw that it got it wrong and fixed it. So I think it's day one for this product. That's very cool. Well, congratulations on the launch. Keep us posted. Last very quick question.

2:59:34Where are we on the Gartner hype cycle? Peak of inflated expectations, trough of disillusionment, slope of enlightenment or plateau? of productivity. I have this conversation every day with a different friend. I forgot the different labels here, but look. It's basically, are we coming down from the peak towards the trough, or are we headed up the slope of enlightenment? I think we're heading up the slope of enlightenment. Wow. Bullish. Bullish. Bull market continues. Very bullish. Very bullish. Well, thank you so much for hopping on. We'll talk to you soon. Have a great rest of your day. Thanks, everyone.

3:00:10Bye. Cheers. Well, if you - Shout out to Ilhan Villani. Yes, shout out. In the chat. Yes. He is an adorned - At Boston University. Let's hear it. For B.U. Watching with five others. Across the coast. Five others. Having a window in Silicon Valley. Bullish. Well, let me tell you about Wander. Find your happy place. Book a Wander with inspiring views, hotel-graded menus, dreamy beds, top-tier cleaning, and 24-7 concierge service. It's a vacation home, but better, folks. Not going to find a Wander on Lindy's vibe-coded Airbnb. He's got to solve getting good inventory. That's what Wander's done. Jeff Dean is, I don't want to say he was inspired by us, but he posted a beautiful one-of-one AI-generated trading card of him playing soccer.

3:00:57He's the chief scientist at Google. Of course, he says, our latest Gemini image generation and editing model is quite good. See the examples in the thread below. It lets you indulge in a bit of creative fun, helps you make new business cards, et cetera. Try it out at gemini.google.com. Jeff Dean, of course, one of the greatest to ever do it. Very, very cool. So we got to, I mean, this. So Logan one-shotted a card generator. Oh, yeah, yeah. Can we play with it? Maybe we should play with it after the show, and we'll bring it up tomorrow. It's looking good. He sent me a preview. Love it. Also, Paki said, I've mostly had X deleted from my phone for the past month or so, but every now and then I re-download it briefly.

3:01:37It's been funny seeing the slow Nikitification of the onboarding flow, like this new pulsing blue orb on the allow notifications button. I didn't even know Nikita could pull this off. There's levels to the game. There are levels to this. Very, very smart. That's obviously going to increase the amount of people that turn on notifications because it draws your eye right to it. Also, just nicely designed, too, that little subtle gradient there to highlight allow. Brian Halligan. This is interesting. Saying, re, Sonos, I have two ideas offers. One, it has a big-ish brand, tons of customers, no real competitor.

3:02:13and absolutely horrible products. I'd love to help a PE firm who would want to take it private, fix it, and take it public again for a big gain. Two, if you're a very sharp, qualified entrepreneur that wants to disrupt them, I'd like to hear your pitch. This is the curse of being Sonos. So many capital allocators, private equity guys, probably have thousands of dollars of Sonos products that they are unhappy with. And every time they go and turn on their TV or their sound system, and they're like, I should take this company over. I should go do it. Keith Raboy in the comments. Keith Raboy. Let me know if you want to work together on this.

3:02:49Yeah, Keith Raboy is getting in the game. Naval is also there. Naval's in. He's talking about another company that has an amazing product that needs to be distilled to a lower mass market price point. Kenneth Castle has a fantastic post here. Yes. He says, I guess his wife says, T.S. and T.K. are engaged. He says, question mark. Taylor Swift and Travis Kelsey. He says, oh, I thought you were talking about T.K. from Uber slash Cloud Kitchens. What? Locked in. Absolute banger. Did you notice what Taylor was wearing in her engagement photo? Absolutely not, but I did see it. Cartier Panthier. Oh, the watch.

3:03:29I did see that. Eight-carat diamond ring. But more importantly, she was wearing a Cartier. Where could you get a Cartier? Bezel. Get bezel.com. Your bezel concierge is available now to source you any watch on the planet. Seriously, any watch. if you're going to be proposing to a pop star. NASCAR has joined Substack. I don't think anybody would have called that four years ago. We've got to do a collab with them. This is the whole thing on Substack. We're going to do collabs with other people. So we've got to get a live video going with them. We've already talked to folks at NASCAR. This will be the second time we're collabing with NASCAR.

3:04:03An absolute tear. Well, folks, thank you for tuning in today. Thank you for tuning in. We've had a fantastic time. We hope you have, too. We hope you have an amazing afternoon. Follow our substack, tbpn.substack.com. We're going very hard over there. And we will see you tomorrow. See you tomorrow. Have a good day. Cheers.