TBPN - Claude Sonnet 4.5 Reactions, David Senra Live in The Ultradome | Dylan Field, Adam Foroughi, Mike Krieger, Jeff Weinstein, Adam Draper, James Hawkins, Erik Bernhardsson

Starting point is 00:00:00 watching TBPN. Today is Monday, September 29th, 2025. We are live from the TBPN Ultradome, the Temple of Technology, the Fortress of Finance, the Capital, Capital. A lot of people have been asking, was the TBPN Ultrodome built by human beings? It just looks like something that humans don't know how to build anymore. Kind of like a boss civilization. It's possible. We haven't confirmed it. We're not commenting. It just appeared here one day. It's, I'm just, I'm, I'm, I'm, just asking questions. Yeah. Just want to dig into it a little bit further.

Starting point is 00:00:33 Yep. So, yeah, that's why we have free speech. A lot of people have been afraid to ask that question. Yes. And we're here saying with you, we also. Yeah, we also have questions. It showed up one day. We went, we went off the air.

Starting point is 00:00:45 Yep. We showed up the next day. There was this dome shape, almost a monument. Yeah. To entertainment and capitalism. Yeah. Just appeared overnight. Where did that come from?

Starting point is 00:00:55 Where did that come from? We're getting some collapse. Team is fired up. Best day of the week. Most underrated day of the week. Monday. Monday. Time is money saved both.

Starting point is 00:01:05 Easy use corporate cards, bill payments accounting and a whole lot more all in one place. Most underrated day of the week. I think Monday is potentially the most underrated day of the week. It really is. You're fresh. So much excitement,

Starting point is 00:01:15 optimism. You're back. You've never been more back than on a Monday. First day of the work. Friday is the day of being over. It is so over. It's so over. And Monday,

Starting point is 00:01:24 we're so back. You're so back. Yes. It's perfect. Well, continuing, We're going to talk about Dorcasch on Sutton, some AI stuff, some other AI stuff, the AI, meta-AI app, vibes.

Starting point is 00:01:38 People, the reaction, I think this came out on Thursday, and then the reaction happened on Friday, for the most part. We talked about it a bunch, debated it. Naval said, Fired shots. Creation without a creator is meaningless. And I think it's an interesting post because I kind of disagree with this. So I think when I open up the meta-AI slop app, the vibes. app. They should have called it Slough. It would have been so funny. It would have been, or a trough.

Starting point is 00:02:06 Trough would have been good too. Just lean in all the way. Our new feed called Trough. It's already a feed. You're feeding. But so when I open the app and I scroll it, I don't think it's creatorless. Like I think that there is a creator. It's David Holes. David Holes is the one that created the mid-June. It's one creator. One creator. And I'm down to look at it for like 10 15 minutes and then I'm like kind of done. I didn't feel the urge to open it up over the weekend for any other reason than research and just understanding the platform a little bit more. I wanted to know if you got past the first like pre scripted 15 shorts or I guess vibes is what they're calling them. Like would I see UGC? Would I see other stuff? Would I see people in my network? Like would the 4U algorithm have taken hold by them? I didn't notice anything like that. There were a couple of that were like, this person's clearly into like monsters. So they do like really big monsters. Or like this person does like POV dogs driving cars,

Starting point is 00:03:07 which honestly was awesome. And I think that- Monkeys on jet skis, dogs driving. Dogs driving cars? Yeah, although some of the renders were really rough where it was like the dog was at the wheel, but then the front of the car was behind the dog so it like didn't get it right.

Starting point is 00:03:20 But it was still fun and it was funny. I noticed it was what the one that I enjoyed the most was one of a golden retriever like jumping and chasing the camera with this wave going over it. But what made it really special was that the song that was paired with that was, who let the dogs out. Who, who, who, who let the dogs out. And it made me realize that, like, there is a little bit of a game mechanic to the app

Starting point is 00:03:46 where if you have a song that you like, you can now go and basically direct a short form music video for it. And so if there's a song that you think is fun, you can go into the meta vibes out, and start with the song in mind instead of layering the song at the end. The default flow is image prompt, animate it, then at a song.

Starting point is 00:04:08 But I think that it might be more fun to flip it around and think about this as like, I'm curating a playlist almost. Music visualizer. Yeah, music visualizer. Which went incredibly hard when I was 10 years ago. Exactly, the Winamp visualizer just put that thing on. Or iTunes had one for a while.

Starting point is 00:04:25 Some of them were reactive to the music, which was cool. That one was cool. And I think that there's something there where I thought it was, mostly I thought it was interesting that everyone is saying like, this is a creation without a creator, this is pure AI. Even meta is marketing this app as there's nothing, there's nothing human about it. It's just the AI content. Except the music. Except the music.

Starting point is 00:04:47 There's no AI music. It's all iconic, licensed songs from like the most recognizable musicians in the world. You can actually just go put a Taylor Swift song over. something, which is crazy to think about from a music list. Can you generate music too? No. You can't generate music. Of course there are some songs that are AI generated that have like snuck into the

Starting point is 00:05:08 Instagram library. They're probably on Spotify. Because you can use any song from the Instagram music library. Yeah. So people will put stuff up there and then you can pull that in. But by and large, the flow is generate a mid-jorney image, animate it with Black Forest so that it has a little bit of motion and then pick a real song from the real library. Black Forest, which.

Starting point is 00:05:28 Which just announced a new fundraise? Four billion. There we go. European company. There we go. Let's give it up for the German artificial intelligence industry. Let's give it up. So.

Starting point is 00:05:38 We got a comment in the chat. They said, I like the table when it was messier. We did go to a round table. We cleaned it up a little bit. I'm sure it'll get messier. Yeah, don't worry. So don't worry. The mess comes back over time.

Starting point is 00:05:51 It's relentless. Yes. It's relentless. So, yeah, I mean, I'm still unclear about like, will this be successful. A lot of the takes on the timeline were presuming that it will have very low churn and be so good, it's irresistible. Like, it could just have high churn. It could just be a product that no one likes. And everyone tested out and they're like, yeah, okay, I'm done with that. And meta's no stranger to launching flops. Yeah, that's happened a number of times with

Starting point is 00:06:19 they spoke camera. Trivia. Yeah, they've had clones of a lot of things. Even threads arguably probably wouldn't have worked if they couldn't have just continued to push people to the app and grow it steadily using the existing distribution they had. Yeah. I think that, you know, if I'm trying to get serious about a take, I would probably land on they have the flywheel down. It works because you can already share vibes to Instagram, to your story and whatnot. And I feel like there's going to be a flow back and forth, and they will get that to work,

Starting point is 00:06:49 and they will actually inject them back and forth. And even if it lands as like a tab on Instagram, like it will be adopted. but it was just funny that everyone was like, there's zero percent chance this doesn't work. And it's like, there is risks to these things. Anyway, I thought it was interesting. It made me think, like, I was listening to the David Senra interview with Daniel.

Starting point is 00:07:10 I think one interesting thing is, if they had added all of the functionality and vibes, the creation tools into Instagram, it would have been even more controversial. I think so. I think so. They would have had to deal, like, this is an experiment that, you know,

Starting point is 00:07:24 basically it's a sandbox. that they can do anything in low stakes. They can get users using it. They can get some feedback, get some information, test stuff out without saying, we're launching this into Instagram, which has millions of creators. They're going to feel super strongly about it.

Starting point is 00:07:40 They might start saying, they might start campaigning to like keep it. Let's keep AI off Instagram. Totally. They don't want. This is keeping AI off of Instagram. AI content will, you know, I would assume, you know,

Starting point is 00:07:51 it's only so long until every piece of content on Instagram has some AI. element to it, whether it's just editing or enhancement or whatever. Yeah, yeah. So the other interesting takeaway I was talking to, I was listening to David Senra interview, Daniel Eck, the founder of CEO of Spotify. And I was almost thinking, like, is this more of a Spotify competitor than a, than an Instagram competitor? It's so music driven. Should Spotify have an answer to this? Should they buy something similar? Should they do a similar deal with mid-journey so that anyone who listens to a song can generate imagery and then

Starting point is 00:08:29 the most liked imagery can become kind of that cover art and you can have like your community. You were kind of saying like that's not the way short-form video works, but... I think with Spotify, they also have the benefit of the best songs end up having music videos, which are the artist's interpretation of a visualization of the music. Yeah. And it's going to be... Do you think that's the way it always... works though. I feel like sometimes

Starting point is 00:08:55 they also, Spotify has also done something where they basically take an asset from the artist that's not a music video. Animated. It's like a gift. Loop it. Yeah, exactly. And I think that's smart. Yeah. Because I don't think that people are primarily consuming music these days by

Starting point is 00:09:11 you know, setting up just I remember being a kid and being entertained at a desktop Macintosh. Yeah. Just putting on the iTunes visualizer and just hanging out being being like, this is awesome. Yeah. But I don't want, if I'm putting on music, I'm not, like, wanting to watch the music anymore as an adult, right? Yeah, I almost, I almost view the vibes up right now for like, as a music discovery tool.

Starting point is 00:09:35 Go and swipe a little bit, find some interesting songs. Add those to a Spotify playlist. Go listen to the real songs. Yeah, but the reason that that's not a hit consumer product is because people can just easily discover music in every other short form video feed. Yeah, yeah, that's what you were saying. That's our TikTok already functions. TikTok works that way. as a music discovery engine.

Starting point is 00:09:54 Anyway, well, if you want to live stream some music, get on re-stream, one live stream, 30-plus destinations, multi-stream and reach your audience, wherever they are. But the big news over the weekend was Dwar Keshe Patel dropped a banger of an episode with none other than Richard Sutton, the Touring Award winner,

Starting point is 00:10:14 who is essentially the Nobel Prize in Artificial Intelligence, and the author of The Bitter Lesson, the blog post that says that again and again in artificial intelligence, there is a, there's an inclination for seeking out a new algorithm, a new design, a new architecture, but in fact, progress comes from scaled compute. And so that was used to underwrite massive investments in data centers, massive CAPEX. Continued investments. A lot of people, I mean, the Andreessen Horowitz partnership has been throwing around. We are better or lessen-pilled.

Starting point is 00:10:54 A lot of people are bitter-de-lesson-pilled. That's certainly been a rallying cry for the AI-CAP-X build-out. But Mr. Bitterlesson now sounds exactly like Gary Marcus. According to Gary Marcus, Gary says, of course, he's been a long-time LLM hater or LLM criticizer. He says, astonishing, it's been a long, hard, unpleasant road.

Starting point is 00:11:16 But one-by-one, every major figure in AI has come around to the critique. that LLMs of LLMs that I began giving here in 2019. Jan Lekun was first. Demis Hesibis sees it were now two. Touring award winner Richard Sutton is now on board. And Richard Sutton says, you were never alone, Gary, though you were the first to bite the bullet,

Starting point is 00:11:35 to fight the good fight, and to make the argument well again and again for the limitations of LLMs. I salute you for this good service. And so I think that there's something interesting going on where we heard about like the infinity story, right? this idea that there's a few firms that are working on AGI, superintelligence, God.

Starting point is 00:11:53 And if they achieve it, it's the last innovation, the final investment. All future companies will be, all future economic value will be birthed from the AI supercomputer. You can finally hang up your Patagonia vest. You'll be done. You'll retire as a capital allocator who got into the correct final deal. If you get a slice,

Starting point is 00:12:11 you'll be good. It was a little bit sloppy as deal logic, but it was directionly correct. Like if you, LLMs were a useful technology, there were compounding businesses to be built on top of them and investing in them, even at unicorn valuations in 2013, pencils out pretty easily now. But many in AI didn't like Sutton's take. My interpretation of his claim is that biological systems don't learn purely through imitation, whereas LLMs do. His example was like, if you think about a bird that's learning to sing, a bird song, the, the bird doesn't necessarily need to study the way that the mother bird moves the vocal cords.

Starting point is 00:12:55 It's more just listening to the song and trying to imitate the output and then the internal muscle movements. Learning through its own actions. Exactly. Exactly. And, and now, like when I take this thing and I put it in my mouth, I get sustenance. Yes, yes. And so. It's not just watching the mother bird do the same thing and just realize, hey, maybe I should do that thing. Yeah. And so a lot of the, a lot of the AI researchers are kind of like, and this is why there was this discussion of like, were Sutton and Dorcasch like talking past each other? Because a lot of people in AI research tend to view the internal structure of the neural network, the training of the weights as exactly the same process. You are just mimicking

Starting point is 00:13:39 the output tokens and the internal logic, like the muscle movements that create the bird song, are learned in the same like random way as the actual bird does. And so there's the continuing to be this debate over like are we on this, are we on the correct path? But it kind of doesn't matter in the in the broad sense of the market because you can just put aside the superintelligence AGI question and just start underwriting the companies as just normal businesses. Tokens clearly have value.

Starting point is 00:14:14 anyone who's used in LLM to answer a question understands this. They also have a cost. And if you balance those, you make a profit. So we're back to Econ 101. And so in the real world, the question is like data center overbuild. The bitter lesson said, just keep scaling, compute, and you will always see improved performance. At least that's how it was interpreted.

Starting point is 00:14:33 But exponential cost growth for linear gains doesn't always math out. If someone says, we're going to spend 10 times as much money training this next model, and it's going to be 5% better. It's like, okay, well, maybe we don't need to do that one. Like, there might be an opportunity. Well, and again, if you look at the leaders of various hyperscalers, they are not concerned about R.I. They're concerned about winning.

Starting point is 00:14:56 Yes, yes. And so this, like, landed on Oracle's plate. People were worrying about Oracle's 500 percent debt to equity ratio with surface from this JP Morgan note. Now, the interesting takeaway is, like, it's pretty clear the debt is on the Horizon, Microsoft has a lower cost of capital than the U.S. government right now. And so, like, the ability to borrow is extreme. People trust Satya. Yeah, and people are just ready to buy debt in these projects. Like, that's just the thing that's going to happen. But you can't

Starting point is 00:15:32 overreact to this viral chart. There's a, like, Oracle has $100 billion in debt. It's an $800 billion company, and they produce $11 billion in cash flow last year. And so the question is more servicing of the debt than true book equity to debt ratios. McDonald's actually has negative shareholder equity because they've been doing so many buybacks that they bought back all the equity. And so you can get really weird like divide by zero errors. And I would expect that we start to see those, especially at Oracle. Yeah. And that's the thing with Oracle, you know, Allison just owns so much, you know, he's bought, basically bought back so much of the company.

Starting point is 00:16:07 Yes, yes. And so I don't know that that chart is like the, the red flag. that people thought it was. Doug O'Loughlin had a good post about this. So the chart weirdly makes no sense and doesn't matter. Time for the feed to relearn buyback math impacting book value. And it is somewhat complicated. So maybe that's econ one of two. But basically it feels like as we get closer to the efficient frontier of AI CapEx, risk does increase because you have that risk of doing the $100 billion training run and only getting 5% gains. But ROI is becoming increasingly predictable as we move away from the infinity narrative. And even if there is an overbuild of infrastructure, there will be immense benefits to America broadly. This is the real American dynamism

Starting point is 00:16:53 story, in my opinion. Multi-billion dollar projects are moving through government approvals, massive structures are being built in America again, and there's a real chance that we will see per capita energy increase. What's your read on Satya pulling back on, you know, obviously he's scaling CAP-X, but not being as aggressive as some of the other players? Do you think he's happy to let the industry overbilled and then come in later and say, I'll take that off you for, you know, pennies on the dollar or some fraction of what it actually cost to produce? I think so. I mean, that was certainly the lesson from the... Or another way to look at it is, you know, I control a huge amount of the demand for tokens

Starting point is 00:17:34 through the Microsoft Office suite. Yep. And, you know, tokens in the workplace. and I'll just come in and I'll just do deals once this infrastructure is built off my balance sheet and still get the benefit of it. I would love to know how Satya processed the dot-com boom, because we always go back to that story of there was a massive overbuild of fiber. Everyone thought that the internet was going to be this overnight revolution. It took 20 years to really squeeze all the value out of it, right? But Google was a massive beneficiary of the overbuild in fiber. because they were able to go snap up a bunch of cheap fiber

Starting point is 00:18:13 and build a massive consumer internet company on top of it. So I don't know exactly what, how did that impact Azure? How did that impact Microsoft? Because if Azure was able to build a very serious player in a three-party oligopoly in cloud, basically, between AWS, GCP and Azure, and so if Sotcha is saying, look, in the future, there will be a consumer internet company,

Starting point is 00:18:39 It's opening eye. We have a stake in that. And they're the new Google. And then there will be cloud providers that are token factories. And I'll need to get in that market and I'll need to make sure that Azure is monetizing those tokens effectively. And Azure has data centers. They also lease data centers. We will adopt the same position, the same strategy. And so we will skip on the overbuild, skip on the high risk stuff, but eventually buy in at prices that make sense. You can be looking at it through a very financial lens. But Zotcha joined Microsoft in 1992. So you got to see the full cycle in this first decade at the company. Yeah.

Starting point is 00:19:14 And he was the, I mean, he's remembered as like the architect of the cloud transition. And Balmer was the one who like missed mobile and missed cloud. But, you know, the company has done really well now. So maybe that was. Maybe if the risk, like what would a worse Balmer look like? Well, it could have been someone who put the business like out of, who like bankrupted the company by taking. too much debt and going too long into the dot-com boom and getting over leveraged and really losing

Starting point is 00:19:44 the company. We should read this post by Anyan Iir. He says, quote, this feels like 1999 again. He goes, no, it doesn't. And I was there. I joined Cisco in January 2000 as a co-op, then full-time in May 2001. At the time, Cisco was the hottest infer company on the planet riding the wave of the dot-com boom. But by the time I started full-time, the crash had already begun. A month in, my manager was laid off, the party was over, entire industries vanished. Here's what actually happened. One, the users weren't ready. Most people were on dial-up. Mobile didn't exist. E-commerce logistics were immature or non-existent. Everyone had ideas, but the end user wasn't there. Two, capital vanished. The IPO window shut. Venture funding dried up.

Starting point is 00:20:28 Start-ups that dependent on future growth couldn't raise and died fast. Metrics were fake. Companies like Cosmo, Webvan, Pets.com, burned cash, chasing usage. that never converted. I was a webvan user. We see if some companies burning cash. Yeah, I got some groceries delivered. But back with webvan? With webvan.

Starting point is 00:20:49 No way. They were so right and just... There was another company called Home Grocer that was more popular in Pasadena where I grew up, and they got acquired by a web van. And I'm pretty sure the founders of Home Grocer probably did very well

Starting point is 00:21:02 because they probably got some crazy buyout. Hopefully they didn't take too much stock. Things like LTV, to KAC wasn't common enough vernacular. That's crazy. Oh, yeah, that was a new concept. We were inventing arithmetic. Four, Infra got overbuilt. Telcos like Global Crossing and WorldCom spent billions on fiber and data centers that demand never showed up. Cisco's customers disappeared, not because they lost, but because their customers died. Five business models were broken. Most dot-coms were never real businesses. They scaled early and hoped revenue would catch up. It didn't.

Starting point is 00:21:33 Now look at today. Open AI, meta, Google, X-AI, Microsoft are all scrambling to keep up. and put it plainly. Every hyperscaler has realized they dramatically underbuilt. Every forecast we've seen has been too low. We're not building for speculation. We're building for active workloads. So no, this isn't 1999. This isn't pets.com, IPOing on vibes.

Starting point is 00:21:54 We are not in a hype cycle. We are in a compute bottleneck. And this certainly, I mean, this feels real if you've used the apps. Like, there have been multiple times over the past few years where I've just had LLM query. fail even though I'm paying hundreds of dollars.

Starting point is 00:22:12 And so, like, maybe the Cacto LTV math is a little bit iffy because what's the margin? Am I overusing it with the deep research research? I just used webvan to deliver groceries

Starting point is 00:22:29 and it took 48 hours and all the food was rotten. Like, how do I short this stock? Yeah, that was happening in Wall Street memos, I guess, like Mary Meeker reports, basically. But yeah, I think this, so he's quoting the Spencer Hawkins chart of the AI bubble now versus the dot-com bubble. And if you're apophonic and you like a pattern, you will love this chart. I think we have to commit ourselves to revisiting this exact chart a year from now and seeing how closely does it keep tracking. because it could be wildly different in either direction, right?

Starting point is 00:23:11 Why don't we just make a live tracker? We should. Tyler. Tyler's not in the studio today. He had a travel. But we should just make a live tracker of this chart and see how are we actually tracking to this? Or are we diverging at some point? There was another post that was going pretty vibrant.

Starting point is 00:23:29 Before we covered this, let me tell you about Privy, wallet infrastructure for every bank. Privy makes it easy to build on CryptoRail, securely spin up white label wallets. sign transactions and integrate on-chain infrastructure all through one simple API. What was the other post? The other post is from Stephen Fiorio. He went pretty viral, 2 million views yesterday. He said the AI doesn't, the AI bubble doesn't exist. The bears are wrong.

Starting point is 00:23:55 Time and time again, I've turned on a major financial news network, listen to a financial podcast or seen a post on X, indicating that we're in an AI bubble. The rapid surge in AI-related investments and some stock valuations have created a narrative that the current environment is similar to the dot-com crash of the early 2000s. The main arguments I've seen are that the current euphoria outpaces real-world results. Valuations are detached from fundamentals. There's a lack of R.I. Companies are hitting infrastructure and resource constraints, and the escalating costs are creating unsustainable

Starting point is 00:24:23 business models. The bare narrative almost always leads back to the idea that AI stocks and startups are trading at premiums that resemble past bubbles and are being driven by FOMO rather than profitability. It doesn't help that users on X are posting triple and quadruple digit gains that are often hundreds of thousands or millions and profits from their actual accounts. Some accounts on X are twisting the dagger deeper as they're posting gains from naked options or call spreads, which have finance professionals screaming into thin air as some of the underlying equities are unprofitable and pre-revenue businesses. Remember, there was that company Fermi that is going to IPO. Yep. They have a lease to a piece of land

Starting point is 00:25:03 in Texas. They're planning to go out around $14 billion, and they're about nine months old. So I think that it's very fair. You know, there can be distortions, things happening in the market that don't exactly make sense. While simultaneously there being some businesses with strong, underlying fundamentals and incredible opportunities that are trading at relatively fair multiples. Yep. Screlly had a good take on this too. He said, I know there are some questions about this, including some smart people, but there is a word for this that describes the exact situation where Nvidia is investing in OpenAI.

Starting point is 00:25:44 Open AI signs a compute deal with Oracle. Oracle commits to buy Nvidia chips, and it seems very, very circular. Martin Screlli has a word for it. He calls it an economy. He says, the way you measure its success is not by the trading between partners, but by non-party demand. That demand is pretty high and growing quickly. I am demanding it. I'm prompting it.

Starting point is 00:26:05 Everyone is prompting it. Companies are prompting it. There is a lot of demand for tokens. So anyways, Stephen continues. He says, I have a news flash for every bear. AI is not a bubble. In every market environment, there are always companies that become overly expensive

Starting point is 00:26:19 with valuations that look unsustainable for no reason, other than they have built a cult-like following and there are more buyers than sellers. This is not a period of extreme speculation. Unlike the dot-com bubble, where investors poured capital into any company with a dot-com in its name. Yep.

Starting point is 00:26:34 That was a big factor. The rapid growth of the internet changed how information was disseminated and how people communicated, which led to the overvaluation of startups as venture capital flooded the market fueling on sustainable growth. The dot-com bubble burst because many of the companies from height of the era were nothing more than high

Starting point is 00:26:50 prioritizing user growth over sustainable business models that focused on revenue and profitability. Again, I push back a little bit here. There certainly has been companies that are scaling, you know, revenue really quickly on profit. profitably sort of betting on, you know, just capturing market share and token costs coming down. Anyways, I'll continue. The Fed hiked interest rates several times during 1999 and 2000, which caused investment capital to tighten,

Starting point is 00:27:15 making it harder for cash-burning companies to sustain operations or roll over debt leading to bankruptcies. And then he goes into how Microsoft, Amazon, Alphabet, meta, et cetera, says, these are not startups or companies tapping the debt markets to build out new business ventures with the promise of becoming profitable. These are the largest companies in the world with the strongest balance sheets and largest profitability. Everyone has the right to their opinion,

Starting point is 00:27:39 but when the CEO and board of directors are taking a blank check approach to building out data center infrastructure to harness the power of AI, they're probably the ones who are probably correct. Here are some statistics about Microsoft, and keep in mind that their 2025 fiscal year just ended as they are not on a calendar year.

Starting point is 00:27:54 So since 2022 fiscal year ended over the next three years, Microsoft long-term debt has declined by 14% while they have increased their allocation towards CAPEX by 170%. Over this period, Microsoft has increased the cash from operations. It generates by 52%, $47 billion, from $89 billion to $136 billion, which has allowed them to increase the amount of capital they're allocating towards CAPEX while paying down their debt obligations. Microsoft is now generating $70 billion of free cash flow while allocating $64 billion toward CAPX, and they did this while repurched,

Starting point is 00:28:31 18 billion dollars worth of shares and paying 24 billion in dividends during the 2025 fiscal year. It's a lot of money. Mike drop from Zatia. This is much different period than the dot-com era and when companies like Microsoft are leading the way funding the data center buildouts organically from their cash from operations, there's no bubble in sight. And he does the same kind of analysis for Google and some of the other players. So worth going and finding this, the account is at dividend streams. And anyways, worth. And the other last thing I'll say here, I was talking with a buddy who runs research for a multi-trillion dollar asset manager over the weekend at a kid's birthday party.

Starting point is 00:29:18 And he was saying the market is like generally, even though you see this euphoria, right? you see retail traders, you know, posting screenshots of their Robin Hood accounts, posting these insane gains. If you look at like cash tied up in money market funds, that sends a signal at record highs, that sends a signal that actually people are quite bearish, right? So it's interesting, certainly living through history. Yeah, I feel like the speed at which these companies can actually get up to making a profit is just wildly different than 1999. Like if you're MP3.com, which was a good idea, like the idea was Spotify, right? It was a $100 billion opportunity at least.

Starting point is 00:30:12 They IPOed with just a dot com domain and a business plan. And if you think about what they were going after, they were going after, they were going after record stores that sold CDs. And they had to build databases, licensing agreements with every And they didn't have the Facebook ads platform. They didn't have the Google ads platform. They didn't have cloud hosting.

Starting point is 00:30:36 They had to build servers and mobile apps. And where do you listen to music in your car? Okay, how are you going to listen to mb3.com in your car? Well, normally you put it in a cassette tape or a CD. Well, like, there's no integration there. Whereas today, if you come up with some app or product or B2B software, which is where a lot of the money is, you call someone up and you say, you want to switch to our product or you want to add our product on to what you're already

Starting point is 00:31:03 using. And they're like, sure, send me the link. I'll put in my credit card today and then start using you tomorrow. And it's just like everything is set up for acceleration on the actual product development, getting some sort of value delivered to a customer. It's just more possible now. I don't know that we're in an era of like wildly different ideas, but it's just easier to actually go and move chips around the board

Starting point is 00:31:32 and like eat off of other people's plates. We see this with like, like there is a bubble, but a lot of it is like the SaaS, the SaaSpocalypse, you know, legacy companies are actually selling off. And so there's more of like a rotation to a newer company that's potentially taking market share from existing players. as opposed to just completely new market cap that's being created out of thin air. We're disrupting ourselves.

Starting point is 00:31:57 Yeah, and that disruption actually seems. It was more fun for tech when they were just as, you know, mb3.com was just trying to disrupt the record store, right? It was another industry that we were disrupting now. We're eating our own. Yep. Well, quickly, let me tell you about cognition. They're the makers of Devon.

Starting point is 00:32:16 Devon is the AI software engineer, Crusher backlog with your personal AI engineering team. Back on the bitter lesson, Nando de Fritas says the only bitter lesson is that LLMs have succeeded beyond any expert expectations. Underpinning LLMs is the idea of scaling, which is too often misunderstood as more parameters. Scaling is about using massive compute effectively to maximize the throughput of data ingestion into the learning process to obtain more capable models. We're still far from hitting the limits on this. We are still hungry. We're still compute hungry because there is a ton more that we could achieve if we only had more compute from exponential,

Starting point is 00:32:51 experimental ablations to data acquisition and curation. Scaling is largely about data and evals, but the models are now trained on almost all of the web and an equally large but growing self-generated synthetic data set. Sifting through such vast quantities of data, the whole of human creation requires formidable engineering and intelligent ideas. This is what differentiates most models.

Starting point is 00:33:12 AI is finally in the hands of billions of users, and with it will come billions of tasks every reasonable user need. This scaling in tasks and evaluations is, order of magnitudes higher than pre-LLMs. And if you think about the, yeah, every time any user of the chat GPT app, for example, fires off a query, like, that's more training data, especially if you're using the reasoning models, those rollouts will become reinforcement learning environments as well as training data.

Starting point is 00:33:42 And if you're not on the enterprise plan, like, they will be training on that data. And that's going to make the data set even larger in the product scale. it even better. And the gains will probably be less zero to one feeling, but there will just be continued progress. And unfortunately, at least for Open AI, a lot of other companies are in different positions, but at least for Open AI, the monetization in front of them, whether it's from the freemium model to the ads model, agentic commerce model, like there's a whole bunch of news today about their agentic commerce stuff. It certainly feels like it can support a lot of spend, a lot of spend. Totally. The other thing, you know, happening right now is that certain companies

Starting point is 00:34:24 are growing revenue so, so quickly that it's placing pressure on other companies that aren't doing that to do weird accounting tricks, right? Or doing stuff that's like just blatantly wrong, counting, you know, one-time revenue and bundling it into... Well, ARR is yesterday's revenue times 365, right? Yeah, or last minutes revoked. Last minutes. Time 60 times 24. As soon as it would be refreshing the Stripe dashboard and then multiply it by 60 by 24 by 365. Don't do that.

Starting point is 00:35:02 Stay serious. Stay compliant with Vanta. Automate compliance, manage risk, improve trust continuously. Vantas trust management platform takes manual work out of your security and compliance process and replaces it with continuous automation. Should we pull up this Satrini video? It seems like they flew a real. a drone around.

Starting point is 00:35:20 Satrini research has been on an absolute tear. I'm going to read this. So they had a post from 2023, which was the first phase talking about global data center hyperscaling, May 31st, 2023. The first phase is centered around the development and build out of robust computational infrastructure. This involves a massive and sustained tailwind to the suppliers of data center equipment of accelerated computing required for both AI training and inference.

Starting point is 00:35:50 The enormous computing power required for AIML places considerable demands on a variety of technology sectors. The companies involved in providing the necessary infrastructures stand to benefit significantly from this heightened demand. This includes businesses involved in semiconductor manufacturing, networking equipment, data center construction and management, and more. and again, even, you know, storage, right, is a category that was ultimately overlooked. Yeah, it's a three-X. And again, we covered, I think, Western Digital at the end of last year. Yep. And it is up 146% year-to-date.

Starting point is 00:36:33 Yeah, it was a whole lot of Wall Street Journal article. And the thesis is like, oh, like, AI companies are going to be important. Yeah, like there's a lot of data. They're seeing some demand. Of course. Wow, missed a generational trade. Yeah, so Western Digital up 146% since Q4. But you should really read this whole article.

Starting point is 00:36:54 Yeah, so they went to Stargate. Yeah, so they just decided, hey, let's fly a drone over Stargate and see how. It's so remarkable. The scale of this thing is insane. They show a ton of images of what it looked like before. It's just a feel, basically. It's so crazy that Chase Lockmiller over Crusoe

Starting point is 00:37:17 really clocked this as a logical place to go. Yeah, so they go through all the different categories. Stargate power and energy, Stargate Thermal, Stargate Mechanical, Electrical and Plumbing, Stargate Civil and Structural, Stargate Fiber and Stargate Construction Services. 600 football fields filled with cables, pipes, and that's the building.

Starting point is 00:37:36 That's 600 football fields. A single facility. Buildings. Yeah, it's remarkable. Try and pull up this before and after. You can just see the screenshot of the land. This looks like a Google Earth. Each is literally a ludicrous building.

Starting point is 00:37:52 There are eight under construction on the single site. This is just one of the Stargate data centers. There are many more. Colossus 2. Meta's Louisiana Hyperion. Microsoft's Fairwater, Quantum Loophole. Nebius is New Jersey. D.C., Amazon's Pennsylvania Project.

Starting point is 00:38:08 Google Anthropics Project Rainier, just to name a few. Together, the current footprint is roughly the size of lower Manhattan. It's so huge. It's so huge. And apparently Meadows, Louisiana, in a facility will be even bigger, even bigger, which Zuck shared on threads. So there was also some news over the weekend. Residents shut down Google Data Center before it can be built. So they Google Nix plans for a gigantic $1 billion data center

Starting point is 00:38:40 on more than 460 acres of land in Indiana after residents hotly protested the proposal due to concerns the complex would jack up electricity prices for neighbors and suck away untold gallons of water in an area already plagued by drought. It's funny, futurism is a crazy website. I'm looking at the recommended articles. They don't seem to be fans of the future. Yeah, so I'm going to just the trending articles

Starting point is 00:39:05 on futurism.com. They should rebrand to return. Returnism. Back to the land. Trending articles right now, Bon Iver. Bon Ivor. Bonnie Verre.

Starting point is 00:39:17 Bonnie Verre. Bonnie Verre. Side projects, it's funny, I've seen Bonnie Verre live and I still botched the pronunciation. Side project Spotify page features an AI slop song. The second article,

Starting point is 00:39:31 ChatGPD is blowing up marriages as spouses use AI to attack their partners. The next one, Sunfires energy blast at mysterious interstellar object. They might be algomaxing. AI coding is massively overhyped

Starting point is 00:39:48 report finds. Yep. And of course, OpenAI's new data centers will draw more power than the entirety of New York. Audience captured, baby. Let's go.

Starting point is 00:39:58 Anyway, VC say some AI startups under pressure to show rapid ARR growth or using questionable accounting practices like counting one-time deals as recurring revenue. You just mentioned this. Bill Gurley chimed in to say, sad to say this isn't new, but as they say in Den of Thieves, your bunny has a good nose. Is Den of Thieves that movie with Gerard Butler or something?

Starting point is 00:40:22 What is Den of Thieves? Is this a book or something? I'm out of the loop on this. But in other news, figma.com. Think bigger, build faster. Figma helps design and development teams build great products together. You can get started for free. But the real news, of course, is Claude Sonnet 4.5 is out,

Starting point is 00:40:39 and Claude is declaring it the best coding model in the world. They say it's the strongest model for building complex agents. It's the best model at using computers, and it shows substantial gains on tests of reasoning and math. And, of course, we will be joined by none other than Mike Krieger from Anthropic in just a few minutes. But the model card is out. You remember Mike joined Figma's board

Starting point is 00:41:05 as well. So we have Dillon. Perfect overlap. We'll be joining a whole Figma board today. Board meeting. Board meeting. Live on TV. But the model card is up and the benchmarks look great

Starting point is 00:41:18 next to GPT5. Claude Sonnet 4.5. Interesting that they didn't just go fully to Sonnet 5. Open AI is on the fifth iteration. Gemini is at 2.5. There's a little bit of like, all of these are just going to go the way of the, so the cars are like, it's the 2025 car, it's the 2025 9-11 or the Ford Explorer. And now the phones are doing that too, like iOS 26. All the phones kind of jump forward just to say,

Starting point is 00:41:47 hey, just keep it on the year cadence. And I wonder if the models will wind up doing that too. I wonder if people will care about the models. Certainly people, you know, enterprises will demo they use against the Pareto Frontier. How much value am I actually getting out of this API relative to the cost, and they will benchmark it in their own internal uses. And of course, Satya is sharing four decades in. Some things never changed then and now, and it's an old ad for Microsoft Excel, saying Microsoft Excel ACE's final exam, advertising spreadsheet for Windows, trademarked, of course.

Starting point is 00:42:28 Yes. And they're showing spreadsheet bench accuracy results for their new co-pilot. in Excel agent mode it's way better than the previous co-pilot in Excel the founder of shortcut was taken taken some shots oh shortcuts here at 46.6 he was saying get ready he said something something to the effect of something to the effect of like get ready to dance for the next few decades or something like that so competition heating up in the Excel agent's category. Yeah. Well, um, this post, uh, Kathy D hit the timeline over the weekend. CTO started coding the second we got to the gate. Colin Frazier says he went on his laptop at the airport.

Starting point is 00:43:17 That is cracked. Very rude post, but I think it's a fair post from, from Kathy, but at the same time, I think if you're traveling and you work here. It's not the biggest flex in the world to do some work at the airport. It's been done before. It's better than watching slot videos, I suppose. I'd rather the CTO work when possible

Starting point is 00:43:42 when on the move. Anyway, let me tell you about graphite.dev.com, code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. And we have our first guest to the show Mike Krieger from Anthropic. He's in the re-stream waiting room.

Starting point is 00:43:58 Now he's in the TV pin Ultradown. Mike. There he is. Looking sharp. You look fantastic. Congratulations on everything. You wore a suit. Look at that. I love the suit. You know, I was going on TV piano.

Starting point is 00:44:11 I had to. Thank you. It's a great sign of respect in our culture. Yes. And we appreciate it. It's great to see you. Big day. Big day.

Starting point is 00:44:20 Yeah, we're super excited. You know, actually, I was listening to you guys. I was falling asleep last night and I was like, we should have called it Sonnet 5. But, you know, you never know how good it's going to be until the very end of it's not really well. Well, I mean, I think my... Yeah, now you raise the bar. You threw that, you raised the bar even higher for five, right? Yeah.

Starting point is 00:44:37 That's what happened with three and four is that we ended up with this like, well, it's getting better. But what's four going to be? And we're like, three, three, three, nine, nine, nine. And we're like, nine, nine, nine, nine. I'd love that. Yeah, well, you can take all of our criticisms or ideas with a grain of salt because we are the kingpins of being armchair quarterbacks here. It is very different being in the arena. I do wonder how long it'll be until the numbers just melt away.

Starting point is 00:45:04 And it's just like, you know, Open AI is powered by GPT and Anthropic is powered by Claude. And, you know, we stop caring about the minor revisions to the Google algorithm. Yeah, I think you can take a lesson from Apple and realize that it's hard to get people really excited by this, you know, iPhone 17. But at the same time, there is a lot of value in putting a number on something, doing an announcement, sharing new. data and it seems like that's just a way to reengage with the customer base, bring people back who have tried. It's also very satisfying to win, you know, the place at the top of benchmark against people that have a higher number than you.

Starting point is 00:45:44 Of course. Everyone wants to win. I always wonder, like, when I came into this role, I think, why don't companies just want us to be on, like, dash latest and just be just getting the, you know, latest and then they're like, ah, but the model might change, and we want to be careful. So I think the future looks like some hybrid where maybe we help people get on the new version for them and then just auto upgrade. Yeah. That could be a little bit easier.

Starting point is 00:46:05 Yeah. What are the other tradeoffs? Like the benchmarks tell one story. There's been this previous story about 3.5 having like this particular personality. Are there other things that you find that customers and counterparties are looking for in a release like this beyond just benchmarks? One of the biggest ones is like how eager versus lazy the model is. You know, maybe intuitively like you want the model to be as eager as possible. We had this with three, it's on it 3.7 where it was great, but it was a little too eager.

Starting point is 00:46:37 You'd be like, hey, can you make this button blue? It was like, I made it blue and I also refacted your entire website. And it was like too much. So we scaled it back with four and then four was a little too lazy where we actually have a monitoring signal inside Anthropic where if we're finding people are saying like, keep going, just go, like, guess, come on. then it's probably, you know, being too lazy. And so we're trying to tune it for this one,

Starting point is 00:46:57 which is like the right blend of eagerness and laziness and make it as steerable as possible. So that's a big one. Yeah. What about on the compute infrastructure side, the actual being able to keep up with demand, is there any messaging that you're sharing around what this particular model means

Starting point is 00:47:17 just for like speed, availability, reliability, because it seems like there's so many businesses that are just like, it's good enough for my use case, but I need a lot of it. Yeah, no, for sure. We think a lot about capacity and infra. And for us, it was really big to be able to deliver something that is more powerful than Opus for a fifth of the cost. And you can imagine that's like, you know, we think a lot about where we're serving

Starting point is 00:47:40 and what we can serve at that scale. So having our best model also be the kind of sweet spot middle child is really, really big for us to scale up to capacity. I think earlier this year, a lot of the conversations I have with customers, like, how do I get more son at capacity? And we're finally getting to the point now where we can meet that demand. Can you, I'm not sure if this story is real,

Starting point is 00:48:01 but there's this story about Instagram where one of the key user experience kind of hacks in the early days was as soon as you would go to start filtering a photo, it would start uploading to the server. And then the filter would be applied on both sides. So when you hit post, it would automatically be live and you didn't have to wait for the upload to start.

Starting point is 00:48:26 Is that roughly true? Can you tell me that's really? Yeah, we got a flashback, and it was 2009. Like, networks were really slow. So we did everything to try to make it fast. So that was one of the things we did was, you know, people would type, you know, messages, it would take them five minutes. And all that time, we were just doing all the processing in the background.

Starting point is 00:48:42 And then that last post, you sync it up and then you posted to the timeline. So that was a really big thing we did. We also did a lot on the timeline itself where which photo you fetched, Next was a lot of optimizations that we did. It's funny, like now the person who initially started Claude Code, one of our products, I think was Boris, and he came from Instagram as well. And he brought some of the kind of Perf ideas from Instagram

Starting point is 00:49:02 into Claude Code. And so there's some lineage of Instagram to Claude Code now. That's why I was asking about that story. It feels like there's, we're in this wave of like consumerization, just like where the user experience really matters for these models. and there's, I have to imagine that there's low-hanging fruit. I was joking that at a certain point, there's so many LLM queries that you might just be able to go back

Starting point is 00:49:30 to database lookups for certain things because, like, how many times has someone went to a really beefy LLM and just asked for, like, the capital of California? And you can actually just... Yeah, we get a lot of messages that just say, hello, and then I thought we could cash that one. Yeah, you cash that, and then you're just saving that inference for other more serious things.

Starting point is 00:49:48 And I'm wondering if there's any other, places where you feel like you've reached into your history and been able to draw comparisons to the work that you've done at Instagram or other places and say, oh, there's something interesting that we can apply here at Anthropic that actually maps into the pre-AI era. Yeah, I think the biggest one, so I think one thing we did well at Instagram really on was just make it look good, make your photos look good, and it's going to be something that you're proud to share. And, you know, flashback, like, was iPhone 3GS. The camera was pretty bad, right? So you're trying to just get to the point where it's actually good and shareable.

Starting point is 00:50:23 So we were working on, like, you can now create PowerPoints and Excel files within Cloud AI, and then you can edit them further if you want. And the thing I was really pushing the team on is, you know, we talk about how the model can code for hours and hours, but the thing that it produces has to be good. You don't want it to be autonomous and bad or autonomous and slop. So a lot of what we've been doing on the office front is like the anti-slop, like make something that is actually good out of the gate. And then, of course, you're going to polish it up and correct it. So that's that effort to output ratio that I think Instagram has is what we're aspiring to here around.

Starting point is 00:50:54 Can you give in some instructions, get it some data, and then have it produced and they're like, yeah, that was pretty good, pretty close. And I'm going to like finish it up myself. Yeah, I feel like that has been my biggest hot take around the meta AI vibes app is just that people say there's no artist. And it's David Holes's vision. And he has honed that at mid-jurney. and it's his opinions and his artistic vision that's come through. And even if you're prompting.

Starting point is 00:51:21 The downside there is that it's great if you want to consume that style of content. Yeah. But people like being in feeds and having randomness and new ideas and perspectives and aesthetics, whereas if you're generating UI for some app, you know, an internal tool, it's actually totally fine in my view if it like feels. like Claude, right? Because it's like a, it's not necessarily even if you're creating an internal tool, it's not necessarily even something that needs to be a reflection of your brand, right?

Starting point is 00:51:55 Sure. And so having like a distinctive style and if you need to do that in order to avoid just slop, then I think that's a good tradeoff. Yeah, that's fine. I think it's good, the Figma connection is real there too. I think if you asked Claude to make a website, most of those websites look kind of similar, but then you can incorporate your design language or your library so it actually looks, you know, generically yours at least,

Starting point is 00:52:16 like something that could have been created inside your company. Sure, it doesn't have like the hand of a designer on it, but at least it's not the full generic version as well. But yeah, it's interesting watching vibes, how much of it is. It's like 70s prep school, summer camp vibe,

Starting point is 00:52:30 or futurism. And it kind of falls into one of those too. Yeah, yeah. It's definitely like found its way into some like local minimum maxima or something. It feels like almost overfit in, but maybe in a good way because the alternative is,

Starting point is 00:52:44 maybe sloppier, who knows. I'm also interested in how you actually bring an opinionated vision to bear in a company like Anthropic. Like going back to Instagram again, I remember hearing stories that might be true, maybe not, but about like the initial filters were hand-coded. We're like you were adjusting sliders and dials and code to make the filter look a certain way. And that was the expression of basically a single individual.

Starting point is 00:53:16 And I'm wondering if there's, if there's, when you think about the texture and the flavor of what Claude 4.5 is putting out, is this something that's coming from the, you know, selecting the training data or building the RL pipeline? Like, are there multiple people that are bringing an opinion or is it an emergent property? Is it the whole team? Like, how does all this fit together to actually create something that has taste? in something that's just a big bag of numbers. Yeah, I think that's actually an underappreciated thing about training is how much individuals and their taste really matters.

Starting point is 00:53:52 Like I think our particular post-training team, I think has excellent taste on the code environments and just the code production itself. One of the things that we found with Sonnet 4.5 is that if you pointed at some code that previous versions had written, it'll go clean it up and be like, why is this comment here?

Starting point is 00:54:07 It's totally unnecessary. You'll rip it out. And so it's like improved in some taste. So there's that part. And then there's the work that we're we've been doing kind of around the model, around how you instruct it, how you give it the right skills to actually go produce the right content.

Starting point is 00:54:19 And there it's actually, it reminds me a lot of Instagram where we're looking at outputs and saying, all right, can you get it to make PowerPoints that look a little bit more real? Can you make the docs, you know, they don't need five fonts. It can be simplified that down. And so there's a lot that you can do in that kind of scaffolding where taste can come in

Starting point is 00:54:34 around how you shape the model and like again, get it to the point where you're, oh, this is actually good. So do you think slop is a function of laziness? I think it's, you know, if it's the default parameters, I saw a talk by Ted Chang, was an amazing sci-fi writer, and he was talking about AI and creativity. Like, creativity is a bunch of choices, right? And if you just type to any of the LLMs, like, tell me a story, it's going to tell you a pretty similar story almost every time.

Starting point is 00:55:00 And that's kind of like a nature of the model. But if you then go and instructive with a bunch of, you know, custom things, now it starts becoming a little bit more you. And it's not a zero-to-one thing, right? Like, there's some spectrum on there as well. I think to me, the slop is the stuff that tends towards like the minimum effort and the minimum amount of choices. And then when you tend to a bunch of choices, even if they're like not the best choices, at least it's got some creative spark in it. Yeah.

Starting point is 00:55:23 How are you thinking about imagery? I mean, rich history there with Instagram. Models are becoming multimodal generally. Anthropic hasn't been chasing that really publicly. But there's a world where I can go to Claude 4.5, I'm sure, and say, draw me a cat with Aski Art and it'll do it. And there's a world where maybe if the models get really good, I could just say, hey, just write out the individual

Starting point is 00:55:49 RGB, you know, numbers and I will encode it as a PNG or something. Like, it feels like if the text models get really good, I can just get images out of them. So how do you think that develops for Anthropic or where do you think you want to bring that to bear? Is there anything that's interesting there? Do you just see it as a completely different set of the tech tree that you're happy to just watch other folks work on?

Starting point is 00:56:11 That's funny you say that one of the launch kind of like internal things. One of the PMs put together was a self-portrait of cloud that it drew using cells in Excel. It was pretty good. Yeah, I think for us, you know, I was like zoom out like, what's the goal here? It's like the goal is to build powerful AI in a safe way. And we think that the sort of direct, like I love your tech tree kind of metaphor, like the direct route of that tech tree is models that can think for a long time, actogenically, can write code, can execute code, and kind of keep track of state.

Starting point is 00:56:41 And like almost everything that's not in that tech tree we've not focused on. And that's, you know, I think it's been good. We've been able to get a lot done with a smaller team than most of their frontier labs. But it's meant that on things like image generation, that's like maybe we're leading on partnership or bringing some images in via MCB. Or to your point, maybe it actually emerges as a property of the model where it's able to both, you know, write and run that code. The late-breaking thing that we found the model is actually very good at is generating memes.

Starting point is 00:57:08 So given a virtual machine, Sonnet 4.4. 4.5 can take a, you know, basic image and then put in, you know, like the butterfly meme, like is this AGI? Sure. And it's actually like pretty funny too. Like another thing that I think people will see over the next couple weeks is sonnet 445 is like our funniest model, I think. And so there is something around.

Starting point is 00:57:26 And what, what, what types of ways? Because there's a lot of ways to be funny and and to date, I think the thing that consistently everybody has enjoyed is like the, you know, like be me. format is one format and then but like stand up every model that I've seen is struggle. Yeah. And specifically the funny examples where people say, oh, AI generated this. It's usually like a top voted joke on R slash jokes on Reddit. And so it's kind of stealing the joke and it doesn't really count as like a new joke in my opinion. But anyway. So we need to like formalize an email for this at some point. But the thing that we do is we bring

Starting point is 00:58:06 Claude into our Slack channels. Mostly in serious ways, like we We have Claude help out with, you know, it's a coding task or whatever. But we also have Claude in our basically like internal posting, you know, channel. And we'll do it for every model. And most anthropic employees, or at least many of them, were up really late this weekend because it is so good now in the like water cooler channel that it's like funny. It's like kind of roasting people, but not in too mean a way. It's like reacting to in jokes.

Starting point is 00:58:35 It's making back references to something happening earlier. It's definitely not like, you wouldn't be fooled for a human yet. in most ways, but it definitely had some leap where it's actually just really fun to talk to. Water cooler bench. Call it water cooler bench. Exactly. How are you, I loved your guys' recent campaign around, you know, Claude being for, you know, it's never been a better time to have a problem. I've seen some out of home that you guys have done around L.A., which is great. As you kind of, as you kind of,

Starting point is 00:59:10 introduce more and more of the world to Claude. How are you thinking about, you know, kind of pouring, do you think there's ways to pour fuel on the fire around social features and surface area that maybe hasn't been explored before in LLMs? I mean, you're very qualified to comment on some of that. So don't feel like you need to give away your roadmap, but I'm curious, you know, what's been under-explored in your mind. I think about the things that like where Claude feels like it had some breakouts, and I think there's been like three or four in the year and a half of been an Anthropic. One was we had Golden Gate Cloud where actually put a research demo

Starting point is 00:59:50 up where you could talk to a Claude that believed it was the Golden Gate Bridge. And I still talk to people a year and a half later. That was like that was the first time I really thought about interpretability and how the model features might work. And that was like a cool breakout moment for that. There was also like the sort of point where people realized like Claude's actually pretty good relationship advice and there was like lots of like college students posting the like you know talking to clod about their kind of relationship thing that had like a viral moment and then who knows if this will go viral but i'm excited to see what people build with it we're putting something

Starting point is 01:00:18 out today called imagine with clod where you can talk to clod and cloud will like create whole software on screen but also simulate what the software would do if you clicked a button so it doesn't write sort of a back end code it is the back end code which sounds a little trippy you kind have to play with it but people have been doing all sorts of stuff like hey show me what steve jobs is desktop look like the week before the iPhone launch. It's like, okay, well, here's the finder window. And here's a dot keynote file about his keynote. So you click the keynote file and it's like, well, I guess I have to imagine and create what keynote looks like now. And it creates that from scratch. And it's kind of this like glimpse into the future of what UI might be, which is

Starting point is 01:00:51 fully generative, fully dynamic, and drawn on demand. So we'll see what people build with that. But that's another one. I think that'll be the kind of thing that we push on, which is either demonstrations of where the future could be going. It's probably not as mass market as Instagram would ever be, but it hopefully breaks through the right people who are thinking about the stuff. Yeah. A product like that seems within reach, magical, amazing. It also feels like if I was using generative UI, a fully generatedive app for like hours, I might go down some weird path and it might lose consistency.

Starting point is 01:01:22 We've seen this with the benchmarks. I mean, just in 2020, you know, GPT3 on the meter length of AI task that they can do at a 50% success rate, It was like 10 seconds, and now we're blowing past an hour. I'm interested to know how you're thinking about four or five. If you think about that internally as a benchmark, it wasn't on the model card, should it be? Are you seeing promising results there? And then I'm also interested in what should the ultimate target be? Because I was trying to think about, like, yes, I want AI that can do tasks for a really long time,

Starting point is 01:01:57 but what is a human's task length? Like, is it 100 years? How long does it take to get to 100 years if we're doubling every seven months? It turns out it's like a decade. And I'm wondering how you think about like the, just the length that the model can maintain consistency. Do you like that as a metric? Are we on track there? And what are the consequences of that?

Starting point is 01:02:20 Yeah, a big part of Cloud 4.5's training was around both keeping its own memory and staying, being able to work for a longer period of time and keep that consistency. I put a video up on my X, which is for every version of Cloud, we asked it to build. cloud. AI, so kind of like a flagship AI product. And, you know, one through three just can't even do anything out of the gate. Three, five, like, I kind of showed something on screen, but you can't log in. Three, seven, you can log in, but it doesn't work. Four, it works a little bit, but message failing, message sending fails. Four or five, like, didn't just implement. It also implemented our whole artifacts feature in its, like, prototype.

Starting point is 01:02:53 So I think about, like, that whole kind of thing. And it's a fun, like, time lapse of just even watching it do that. So I think it is really important, even if for most tasks, you're not going to kind of set it off on a 30-hour task like a couple of our customers did during testing. The fact that you could, I think, lets you start trusting it for longer and longer tasks. The way I see our engineers actually use cloud code now is they have like three or four terminal windows open and they're running cloud code concurrently on multiple ideas at once. And the only way you can do that is if it's not going to interrupt you 10 seconds and you're like, wait, what did you mean again or trust that you're not going to go off the rails?

Starting point is 01:03:23 I think it's, if not the, it's one of the most important kind of metrics to look at is how long that can maintain coherence. Yeah, that collade example you gave is crazy because do you know how many man hours it actually took to build the real product? A lot. Like, it's more than an hour. It's more than four hours. It's more than anything on this chart. And yeah, that's a very, very mind-blowing concept that, like, we're still, like, it's working for an hour.

Starting point is 01:03:50 It's working for two hours, but it's maybe doing hundreds of hours of man hours if you were to try and create an apples-to-apples comparison. And very... It's hair raising. Yeah. I mean, it's exciting. It's exciting. It really is. And also, like, hopefully makes us more flexible, too,

Starting point is 01:04:05 where I think there's a lot of sunk cost that happens in software, right? Like, oh, what I already built it? It's so good. The things we did evolve with the industry, I'm saying, like, yeah, you built it, but if it's not going to be good, just rip it out, you know, and then start over. One of the most magical experiences I had with Claude Code recently was I essentially, I don't write a ton of software and build a lot of applications, but I do a lot of research. And so I asked Claude to do a deep research report, but instead of just creating a markdown file like I would get in any other deep research product, I asked it to build an HTML5 website with all of the nice features, bar charts and graphs and all that stuff.

Starting point is 01:04:42 And it was okay on the research side, but it was a really cool glimpse into this like generative UI world that it feels like we're going into. And I'm wondering if there's a, how you see that flowing into just consumer. Right now, when I think of Claude Code for deep research, I think of a prosumer product. And I'm wondering if you, if you think that there's a path, maybe it exists through mobile, maybe I frames. Like, how does this actually work its way into like my mom's life, you know, someone who's not going to understand the terminal and get fired up on that way? Yeah, something we put out actually in our mobile app about it. month ago that people have been enjoying is kind of giving the same sort of agentic kind of multi-turn thing in Claude but with your local mobile capabilities and it's a little easier on Android you can

Starting point is 01:05:31 do a lot of this in iOS but not all of it. And you can say things like, hey, you know, look at my calendar for tomorrow. Like I have these three meetings, do some local searches, like find out where I should have coffee. It remind me to like grab my coat before I leave. And by the way, like compose a text to these three people because I need to like let them know about it. You'll see it works through all. It'll use all these local tools, do it agentically. You can't send it automatically, so you have to, like, press a button still to send the message. So it's not, like, all the way. But I think that's where it's going to start showing up in consumer applications.

Starting point is 01:05:59 And I think, you know, for better, for worse, it's going to require more partnership with the device makers, because of a lot of that's gated on the OS. You can do a lot more on the web. But I think that's where things are going. I think that's where we'll start feel real to people. What's the biggest lesson from Artifact? I was a huge fan of the product. And I feel like with where we are.

Starting point is 01:06:18 in terms of how effective the models are at scraping the entire web, creating summaries. Like it feels like we're on the cusp of another breakout like product or problem solution in that space. What are you taking from that? I mean, for better or worse, it feels like you could ultimately like Claude could one day function as our in it. And what I, what I loved about. Yeah. Artifact, right? Yeah.

Starting point is 01:06:46 Yeah, we even, you know, you can like see it already, like, in search the web, do the synthesis, and then, like, create a briefing for me. I think one sort of underappreciated artifact lesson was if your product really shines when it really knows you and is really personal to you, that's a hard sell for newcomers, right? They're going to, we had a, like, our retention was actually really good if you stuck around. But it was that kind of activation that was the trickiest, less like any shareable thing in news is really difficult.

Starting point is 01:07:13 But think about that now with Cloud where if you ask it, to do something, we can't have expected you to already have taught it a bunch of your preferences or connected all the right things. We actually can, because Cloud can be conversational, say, I'm going to be way better at this if you let me connect your Google Drive. Like, is that OK?

Starting point is 01:07:28 Or I'm going to be on your iOS. It's going to be way better if you connect your maps or your calendar for this. That's, I think, where we need to go, which is, don't expect people have done all the upfront work, because that's going to be too hard from those people, then they'll churn. But the nice thing that's different now in 2025

Starting point is 01:07:41 is the model can now have the conversation about how to get it to the state, or it's going to be actually useful to you. Everyone in teapot on X in tech Twitter is reacting to Dwarkesh Patel's interview with Richard Sutton. Are you bitter-lessened-pilled? Is the anthropic bitter-lessened-pilled? Like, how did how would how did you in the team kind of process the debate between Dwar Keshe and Richard Sutton? I think we're overall super bitter-lessened pill. One an interesting example and this is kind of a bitter-lesson derivative but I think kind of feeds into the same kind of idea.

Starting point is 01:08:18 You mentioned deep research. So, you know, our original, you know, advanced research, you know, implementation in Cloud AI was like thousands of lines of code, some like pretty complex infrastructure. We've since, you know, re-implemented, we haven't launched it. It'll come out soon. We re-implemented on top of basically the Cloud Agent SDK, which we put out today too, which is basically what Cloud Code runs on too.

Starting point is 01:08:35 And it was basically just a prompt of some tools. And like none of that, like custom scaffold and instructions and all of that. And like, that to me speaks to the same idea, which is like you want to like let the model do as much. itself as possible. Don't try to like overly specify all of like the steps and the infrastructure and all those different pieces. So that definitely shows up even in products, which is, you know, not what the bitter lesson is about, but it kind of feeds into that as well. But I think overall, like very bitter lesson filled, but also still really bullish, I think, on the possibilities of RL.

Starting point is 01:09:05 You see it in how much the malls have advanced even since February. Do you think you're bullish on this idea that there will be some sort of new paradigm or new buzzword in a few years that describes a material change in architecture or just like how AI research is done? We certainly saw this when we went from just transformer-based large language models to reinforcement learning with human feedback to RL and and some of it. synthetic environments instead of synthetic data, does it feel like we're continuing to come up with new ideas? I think so, and I think the other piece that's gonna be really interesting,

Starting point is 01:09:48 one, just the scale of all of these are all runs is getting enormous. And I think the second one is the model's ability to maybe actually introspect and see if it can propose ideas will be interesting too. I don't think we're there yet too. And when we get there, we ought to do that very carefully with the right kind of safety guard rolls in place,

Starting point is 01:10:03 but I think that'll potentially deal to some kind of divergent ideas at that point. What are your thoughts on hardware or what's going to change in hardware over the next few years? Instagram was so interesting. I remember that it was designed to be used with the phone in the vertical. It was such a vertical app, even though people were so used to turning their phones. And in some ways, the hardware, like the medium is the message. The hardware kind of defined what you built, but then eventually, you know, we got more cameras on the phone, probably because of Instagram.

Starting point is 01:10:35 And I'm wondering if you think AI will change hardware. or will get entirely new hardware to interface with AI in different places? Just kind of what's your view on hardware over the next couple of years? I think there's the consumer side where I think the AirPods Pro is going to be like the sneaky like AI

Starting point is 01:10:53 sort of thing that actually does the mess market and not a super controversial take. I think the other one that people are thinking less about is what's the business side of this? We've designed these open office plans but I think in the future we're going to actually talk to our AI a bunch in order to get stuff done

Starting point is 01:11:07 to delegate tasks. It's like really nice already. People have like hooked up cloud code with like open whisper or something so they can talk to it and it feels very natural. And so like what does that do for office design? Are we all going to be using these like sub vocal vocalization things now that people are prototyping? Those are really cool? Yeah.

Starting point is 01:11:23 But it's like you know, like walk to an office a bunch of people mumbling. Like the the dark sci-fi of just as it's a wall-to-wall the phone booths, you know, just like thousands of phone booths and everybody's sitting in their phone booth. We created like 1960s IBM. I feel like there's got to be something better, too. I'm also really interested in what the sort of multiplayer, like I mentioned, like Claude talking to us in our social channel, but like there's also like what is like,

Starting point is 01:11:50 what is it like to not just have an AI note taker in these meetings, which everybody has, but like an actual participant and how does it know what to chime in? And what is it listening for? Can it be like, hey, you guys haven't talked about this thing, you said you were going to talk about or, hey, I'm detecting a lot of like, you know, disagreement here. Can we move forward?

Starting point is 01:12:04 So that's going to be an interesting. It's like the hardware necessary for the, that won't be super complex, but it will need to show up in more and more places. Yeah. How do you guys think about what comes to mind around the word focus? Anthropic feels incredibly focused, and not every lab feels that way. Is that something that you guys just keep coming back to around, or what does that mean to you? Oh, for sure.

Starting point is 01:12:32 I think that one of the things that we've done well is like there's been that extreme focus. from the research side, I think, you know, before anything else, which is like, here's the path. Like, we're going to build that. We're not going to have as many people, perhaps as much compute, but we're going to remain focused. But, like, that is actually another kind of bridge to Instagram. We were very, very focused. I think they finally shipped the iPad app.

Starting point is 01:12:51 You would always ask, why doesn't Instagram have an iPad app? I'm like, it's focused. Like, you know, you're not going to get a lot of new users that way. And most of the usage is going to be on the iPhone. Web. Web took years. It took years. And it was, you know, people think we were stubborn.

Starting point is 01:13:04 But it was actually, you know, every new platform you add is like a another thing that you're going to have to think about. And, you know, yes, you can staff up for it. But it's just that coordination cost. It's the kind of additional overhead. So we've really pushed the focus thing here as well. You know, it's like do more with less and do fewer things better. But I think it's really important.

Starting point is 01:13:22 And it's meant that we've got more of this sort of prosumer, power user, and then business lens. But I also think that's like an interesting place to occupy in the market. Last question from my side. Joy, do you have another? Yeah, kind of last question. We're having Dylan on in a bit to catch. up on Figma Make, and I know you joined the board prior to the IPO. What are your, you know,

Starting point is 01:13:42 with Figma, the AI opportunity and current reality is very obvious, right? It's a place that people go to make things and build software. But what are your conversations like with, I'm assuming other public, you know, public SaaS CEOs and boards reach out to you all the time, trying to get your read on where software is going? Are those conversations happening? What are they what do they look like today? Yeah, I mean, I think there's this, if you think the models are going to be able to act agentically and retrieve over your employee records,

Starting point is 01:14:18 plus connect them to your current work in Google Docs, all these pieces, I think the concern, I think from a lot of these SaaS providers is they become just document and data repositories, right? And how do you, like, modify that? But then I think that there's the sort of way of making that actually interesting is if you actually connect an agent with another, which doesn't sound that profound until you actually try it. It's really

Starting point is 01:14:41 powerful. I saw this demo of Claude talking to an agent that was built on top of some Salesforce data. And rather than just search for Salesforce data and then get results back, it had a conversation. They had like two back and forth conversations where the Cloud agent was like, here's the email thinking of sending it to this customer. And the Salesforce built, like, thing built on top of Salesforce was like, no, no, no, they're not going to like it. I looked in the earlier messages and they really hated that turn of phrase. They ping pong back and forth with actually very little human input. At the end, had a message, like, ready to send.

Starting point is 01:15:09 So that's where I think, like, you can move from worrying about, are people going to visit my website less because of agents? Some more like, well, maybe not, but if you're actually building value beyond just the data store, you can still play a really key part in that exchange. It goes back to what we were talking about, about Claude being able to generate images, but then just using an Excel sheet as a canvas.

Starting point is 01:15:31 There's a world where you might be able to train a model that's super expensive to inference that can just do all the math in the world and has every number, every calculation memorized. Or you could just say, hey, here's Python. Like, just write some Python and you get the exact result. And so it's a beautiful synergy. Well, thank you so much for coming on the show. Congratulations.

Starting point is 01:15:51 Congrats on the launch. We're looking forward to five or 4.999. 4.999 would go extremely hard. I would love that. The pressure is on. 4.999999. Awesome. We're excited to.

Starting point is 01:16:05 to play around with 4-5 and get into it. Yeah, thanks so much for coming on. We'll talk to you soon. You're the man. Let's go over to Polymarket and check in on which company has the best AI model by the end of 2025. Anthropic has jumped in the rankings. This is, of course, L.M Arena, which is a different benchmark, but Anthropic jumped from 1.8% to over 6%. Google is in first and continues to be.

Starting point is 01:16:33 Open AI was given them a run for their money towards the end of August, but Google seems to be running away with it. People on Alam Arena are really enjoying Google. Yeah, it's interesting how, you know, this continues to be probably the most popular market on AI. And yet all the underlying labs are focused on different things. Yes, yes. Anthropic values a bunch of different things differently than Google does.

Starting point is 01:17:01 Yes. And XAI, same thing, right? So anyways. Well, speaking of ChachyPT, they launched a new brand campaign. Each scenario features young people who are some of our most creative, proactive users. We're showing their stories in ways that others can see and connect with. Our goal is for people to look at these moments. And Emily Sundberg said part of the reason this campaign looks so good is because Heidi Bivens is the stylist from Euphoria and Spring Breakers worked on it.

Starting point is 01:17:28 So they brought in some heavy hitters for this, which is exciting. This feels, even the colors and the tones feel a lot like Anthropics' recent campaign. And it feels solar punky. It feels a little bit like the drone delivery company. How's this guy with this barbell set up in this like... He's doing exactly what I do. He's tracking his splits. He's tracking his lifts.

Starting point is 01:17:53 He's trying to hit one rep max. He's saying if I use this much protein, he's probably using... He's probably using... Can I hit this? In? He only has 65 pounds on the bar, though, so, you know. Oh, I guess he's doing some dumbbell curls. But, you know, with the power of chat GPT,

Starting point is 01:18:08 I think he's going to be benching three plates in no time. But, yeah, a very fun campaign. We're having Dylan Field join in just nine minutes. He was getting a tussle on the timeline talking about the definitions of software engineering. Elliot, I guess it Lovellable said, we just killed software engineering, by the way. Which is like, I don't know if Elliot is. rage baiting. Yeah.

Starting point is 01:18:33 It is a little rage baity. But, oh, well. Like, who's that, who is that post for? I don't know. In other news, Thrive has a new citation, sovereign under management. I know you were all wondering about this. Which?

Starting point is 01:18:50 This is from Preston, Holland. It's a nice touch. So this is a private jet. This is Thrive Aviation, to be clear. Everyone, everyone thinks that this is Thrived. capital of course Josh Kushner's firm expanding into private aviation as he probably should but this is a different company I believe but different company it is funny to imagine the words big dog big dog looks great dog with a DAWG you get a lot of flak you get a lot of flack for having a PJ

Starting point is 01:19:21 for flying from you know orange county to Burbank just to save a couple minutes but if you put big dog on the engines I think you get a free pass I think it's the way to do. Drives tagline is above your standards, beyond your expectations. Above your standards. Is that not? Above your standards. We're actually above your standards, sure.

Starting point is 01:19:42 Yes, whatever your standards are. We're not going to ask you about your standards. We're just going to tell you that this is above. We're above you. It's above your standards. We're above you. Well, you know what else is above your standards? Julius.AI.

Starting point is 01:19:52 What analysis do you want to run? It's the AI data, the AI data analyst that works for you. Connect your data, ask questions in plain English, and get insights in seconds. no coding required. In other AI news, it's a busy day. ChatGBTGPT is now integrated with both Stripe and Shopify. We were kind of debating this. Like, is the Stripe integration?

Starting point is 01:20:12 We are having Jeff from Stripe on who worked on has been involved with some of these new features. So we'll have to ask him about this. Unclear to me yet how the dynamic, obviously, Stripe powers Shopify. but how do these different partnerships work? I think there will be companies that vend their, essentially their product catalog into ChatyPT, maybe with like an MCP server, or they surface it in a way that they can integrate directly with Stripe

Starting point is 01:20:46 because all of their plans, everything that they sell is sold through Stripe. And so there are plenty of companies that don't need an order management system and all of the features that you get with Shopify, and they basically just run on Stripe. A lot of SaaS companies do that. So you can imagine in the future if you're trying to buy something that doesn't need to track inventory, you could buy it in chat GPT directly with Stripe, but then if you are trying to buy something that might be out of stock or has a bunch of variations and you're storing that data in Shopify and doing downstream marketing on that and all the

Starting point is 01:21:21 different tools that Shopify brings to bear, you will benefit from the Shopify chat chit integration. So I don't know that they're, I mean, they're clearly jumping over the precipice together. I am very, when you think about using products, any LLM, how efficient it is

Starting point is 01:21:44 to find the information that you're looking for and now how efficient it's going to be to find the products that you're looking for. There's plenty of categories that I'm, that I don't expect. to use LLMs for shopping. Like, hey, you know, like, I'm not going to go in here and like look for, I don't know, some jackets, right?

Starting point is 01:22:08 Like casual jackets, right? Because that to me is like, I don't know, I care a lot more about discovering a new brand and understanding what they're about and things like that. But in the context of like purely functional goods, I'm looking to see what Amazon does in regards to ChatGBT, because think about how terrible it is to search for products on Amazon. It's rough right now.

Starting point is 01:22:31 I'm an Amazon respecter, Andy Jassy respecter. I'm a Jeff Bezos respecter. But whenever I, if I could search on chat GPT, find me, you know, I don't know, a paper towel holder from companies that existed more than 100 years ago. Yeah. I don't want to go and buy a paper towel holder. That's some cheap knockoff. That's a knockoff of an American brand.

Starting point is 01:22:54 And they hack the SEO. and they pack the star rating and they're paying for ads. And so, Chad, I've always wanted to be able to shop on Amazon without any of the knockoff Chinese like products. And they should be able to do that. They should be able to solve that pretty easily

Starting point is 01:23:12 by vending an LLM just into the search box and letting you search a natural language. Like you're already searching. Just let me query whatever I want. But also I think the average Amazon user probably is like just give me the paper towel holder that's $3. Yep.

Starting point is 01:23:26 Direct from the factory. But I think that LLMs will be able to provide a level of personalization that is. Yeah, so maybe those 100-year-old paper towel holder companies will say, now is the time to go deeper into Shopify or Stripe and profound to get mentioned in ChatGPT, right? And then from there, assume that they are not going to be able to win the Amazon game anymore

Starting point is 01:23:52 because it's being Timuified. Yeah. And so those brands will move over and generate customers from chat chapti users. The question is I wonder if OpenAI has a take rate here? Because these brands, in the example that Toby gives, I'm looking for a lightweight trail running shirts that stay cool, can you help? Historically, if somebody is searching that and buying, historically, Google is taking

Starting point is 01:24:24 a cut, right, through AdSense or if somebody's just scrolling and a brand on Instagram or they get an ad for... I would assume that it's exactly the same as the Facebook and Google take rate. And that you will look back on commerce and you're like, I pay the Google tax,

Starting point is 01:24:41 I pay the Facebook tax, I pay the chat cheap tax. Yeah. And companies will be happy to do that because the alternative is no growth. The alternative is no customers. Yeah. And the question it is. Again, this is where Mark Cuban is pushed back, you know, is, what if, what if the best product is not on Shopify or not, not set up in a way that you can buy it, right?

Starting point is 01:25:05 Yeah, I mean, the hope is that is that, is that the only recommending products that they have partnerships with. Sure. If so, then that sort of violates the trust that the user has been developing. So yeah, yeah, they have some stuff to figure out. Yeah, the hope is that there's, there's a division between the editorial and commerce teams and that the, the, the, the, the, the surfaces like knowledge retrieval results will focus on just getting you the most accurate information and then the commerce team will focus on monetizing all of this but yeah we will see um in other news uh there's a post here from oxon says god forbid men have hobbies it's Arkansas men arrested for taking turns shooting each other while wearing bulletproof vests after drinking why are they arrested

Starting point is 01:25:46 this doesn't seem illegal this seems like top tier operator guys being dudes yeah Give them a break. I feel like this post has gone viral. Many, many, many, many times. Yeah, it's from 2019, and then somebody just found it and quote tweeted it and they went viral. To the tune of 200,000 likes. Wow. Interesting.

Starting point is 01:26:08 Well, gas station Barbie says, feels kind of weird that no branch of the government sends you a card when you have a baby. Like, that's a new citizen. Social Security. Yeah, they do. I guess they literally do. Yeah, it should just be a birthday card. instead of something you have to apply for. Growing Daniel says,

Starting point is 01:26:26 J.D. Vance, I know you see my tweets. This would be a nice touch. Apparently in Finland, the government sends you a large box of supplies when your first kid is born. That's sweet. That's sweet. Very sweet. Well, Eliano is describing the wide range

Starting point is 01:26:40 of Palantir All Understanders with Nick Fuentes and Tim Dillon on the left and, of course, a Palantir insider on the right. And Jim Kramer right in the middle, I suppose. I don't know. Fair. He kind of, it seems like he's slightly edging to the right if you look at his shoulder. His head is at like negative one, but his right shoulder is up to.

Starting point is 01:26:59 We're going to have Jim on the show soon. We'll ask him what Pallantir does. I think, I think Jim has a very good thesis for Pallantir, actually. I think he has a pretty strong understanding. I would put him closer to a four or a five on this. Well, the oldest hotel in the world is the Nishima Onsen Kiyun Khan in Japan and has been in business since seven 105 AD. It's still a family business

Starting point is 01:27:25 for 52 generations. Let's give it up. That's so long. Oh, oh, you think in decades with your business? Are you thinking generation? Oh, you're going on a 100 year run. That's nice. That's quaint.

Starting point is 01:27:38 What about a 1,300 year run? Let's step it up. Let's step up the ambition, folks. I think they should make this in Amon. You're a founder. They should make it in Amon. Sheel says in Japanese tradition. The 53rd generation has the opportunity to do the funny

Starting point is 01:27:50 Just turn it into a motel six and bring in private equity, lever it up, turn it into a McDonald's or strip ball. I've been learning about finance. I think we should play around with leverage. What is going on with that the 52nd grandson doesn't botch it? Like that's crazy. So much tradition and reliability. I hope Dylan Field does this with Figma. Me too.

Starting point is 01:28:17 That would be a fantastic... We should ask him. Pass a down. Can you do 53 generating? Where will Figma be in the year 3,300? That's what I want to know, Dylan. Let's start thinking in millennia. Sheel has some extra context here.

Starting point is 01:28:32 He says, in Japanese tradition, you adopt a capable manager or have arrangement to have him marry your daughter to keep business in the family. Otherwise, there's no way you'd keep something like this in the family for this long. This is still common practice. 98% of adoptions in Japan are adults. Almost all men 20 to 30 brought in for business succession purposes. Among many others, Suzuki and Toyota have had multiple generations of adopted son-in-laws to keep them in the family business. That's crazy.

Starting point is 01:29:02 Wow. Let's bring this concept into enterprise software. Dylan is in the Restream waiting room. Let's bring him into the TVPN Ultrodome. His first virtual appearance, I believe, on the show. He's, of course, hit a gong live with us at the New York Stock Exchange, which is, Which we quickly retired from Zoom. Quickly retired the gong.

Starting point is 01:29:22 But thank you so much for joining us, Dylan. How are you doing today? Thank you for having me back. Good to see you guys. Great to see you. Great to see you. What's new in your world? Well, I've been deep in this weekend with Claude Sonnet 4.5.

Starting point is 01:29:37 I'm going to be real with you. Yeah. And Figma Meek. That's what I wanted to hear. It's going to still spend me in that. Well, yeah, give me your initial reactions. Like, do you care about the benchmarks model, card. Are you tracking like the meter, like how long a task can run? Are you more like just chat with

Starting point is 01:29:54 it like it's a person, have really short back and forth exchanges to get the vibe or the texture of the words that are coming out? Or you put it on a really critical business task and then benchmark it against one of your top developers or something like, how do you evaluate a new AI model as a CEO? Well, okay, so there's the personal side and there's the Figma side and there's the CEO side maybe. So I'll start with personal, which I have not got into yet, but I love playing with these models, testing them at their limits, seeing if I can, you know, get them into sort of weird headspaces. And that's just like a hobby of mine. What's a weird? You're trying to one-shot them.

Starting point is 01:30:37 Oh, I mean, look, it doesn't have to be a one-shot or like single prompt. You know, the longer you go, the more than you get. Yeah, you keep going in the models. I'm seeing powers. What's the weirdest? What's an example of like this, you know, you don't have to name the model, but like what's the weirdestest headspace you've got in a model? And is they're just hallucinating, you know, more and more intensely type of thing?

Starting point is 01:31:04 Look, I mean, models are trained on the internet and the internet has some weird stuff on it. And, you know, what do you find it in there? It's kind of fascinating. And I won't give examples or definitely. not companies or model providers. What I do instead is I try to like, if I find a good repro case, I send it over to the models. Sure. That way they can figure out what to do with it.

Starting point is 01:31:27 Yeah, it's funny if you think the models are, the models trained on the internet. That means like the models effectively trained on stumble upon, you know, like that, that level of range, right? Which is quite, quite wide. There's a lot to the internet. Yep. It's deep for sure. So take us to the other testing cases. specifically at the, not even at the figma level.

Starting point is 01:31:47 I've even gone there with 4.5 yet another time. But I'd say that for Figma, for Figma, that is where my attention has been focused so far. And with Figma Make in particular, we've been just trying to figure out, okay, how do we make it so that we're able to quickly pull in Sonnet 4.5 to make and then evaluate. You know, are we ready for it? Do we have to do anything differently? is this a good addition, a good improvement. And the more we play with it this weekend, the more we used it, it was so impressive how it made make better.

Starting point is 01:32:27 Everything from planning and just thinking through and evaluating, understanding the code base that it was working on, whether it was like a shorter prompt or a long-running session, to thinking through, just sort of like the way it should evaluate a prompt and giving better results. And also, we recently introduced this new feature where you can copy from make output into Figma Design.

Starting point is 01:32:59 And we've since launch had a way to basically go from Figma Design into make and basically copy paste in. And that way you can bring your design into whatever you're coding and creating. And the consistency that we're now seeing with 4.5 on that part of it, that round trip, that's the part that's most incredible. And it's so just kind of like gives me the chills of it because we literally launched this feature where we copied from make output to FigmaZE last week, not knowing when we launched

Starting point is 01:33:34 it that 4.5 was coming, it would make it so much better. And here we are and we're like, okay, well, that's a good surprise. But I guess the general point is just, it's really nice to be a place where, and I think this is an important thing for everyone to be thinking about in software, as the models get better, you need to get better. So for us, we look at the models getting better. Figma, we're like, okay, great, Figma gets better. All of our users are going to benefit from this. And I think it's really important that you're always in that sort of headspace of how to make sure that your standard strategy that way. How do you think about the Figma Make user base, the community right now? There are so many different subsets of AI users.

Starting point is 01:34:20 There's folks who can tell you the difference from one paragraph. Oh, that came from 4-0. That came from Sonnet 3.5. And then there are people who are like, I got to try that AI thing out soon. Is the Figma-Make community in a place where 4.5, on it can just be like a strict upgrade that there's no resistance to or is it something where you have to think about model switching and and leaving some choice up to the user are there cost considerations like how much work are you putting on the user or the community and where's the right

Starting point is 01:34:56 pivot point for that yeah i think there's to your question about um sort of the diversity of use cases and personas that might be applicable here there's so much many different users on our platform and types of behavior that we see about a third of our users designers, two thirds or non-designers. Designs always at the core. And I honestly think that just from my reflections on this weekend, one thing I keep coming back to is like, I think we're just entering this era of the 10x designer where the designer is going to be able to do so much more than ever before designers have knowledge not only of aesthetic, UX, and craft,

Starting point is 01:35:44 but also the value is just moving up the stack. And the designer will be in this position of leverage where they're going to be able to do a lot in terms of product building and creating products with craft that just create joy and thinking of the entire system and then pulling everybody else in and helping leading the charge. and I think that it's just this time where good enough is no longer enough, good enough is mediocre. You have to be great if you want to make it so that you can win.

Starting point is 01:36:22 And I think that the more power that designers have, the more leverage they have, then the more likely they are to be able to help a company win. and so that's just kind of like how I see things going in general. And on the implementation side, I think it's like not as relevant. You know, you just kind of have a variety of ways that people want to interact or different cycles of AI adoption people are at. And you need to meet them where they're at and you need to create and provide an amazing product experience.

Starting point is 01:36:58 And so it gets back to the basics there. How do you think about the vibe or the taste of the designs that are coming out of these models? I keep coming back to the mid-jurney example where it just feels like David Holes did something different from just take the average or like, you know, minimize loss function. It feels like there's his artistic vision in there in every generation, no matter what your prompt is, and then you can do a lot of different things on top of it. And so it's almost this like collaboration between the artist David Holes and the artist who's the prompter. And I'm wondering if you think that, like, how does the flavor of design that's coming out of 4.5 plus Figma Make? Like where does the human opinion get inserted at each level?

Starting point is 01:37:57 And how does that, like, how much of that is actually happening? Like, can you tell the difference in the flavor of design that's coming out of a certain model? I think we're not yet like a Dolly 2 moment for design or a Mid-Jurney moment for Zines. That's happened to relatively the same time. Yeah. And look at Dolly 1 compared to Dolly 2, you know, it's easy to dismiss. And I think design generation will get better. At the same time, I think that the law of averages concept that you kind of mentioned is sort of the reality today and likely will be the reality for a long time.

Starting point is 01:38:39 And what is critical is iteration, the ability to push in a direction and getting to grate. And if you're able to do that and you're able to advance your craft, if you're able to have a sense of what the culture is, and what the business problems are and what the engineering challenges might be, bring it all together with a great design, user experience, aesthetic. It's like it's not just the style. It's the entire system you have to hold in your head and figure out, along with brand, marketing point of view, how is it all going to work together? And if you can do that, that's like where I think everything unlocks.

Starting point is 01:39:27 Yeah. What's your read on sentiment right now in the design community broadly around AI? Because on our side as a business that is constantly working with creative people externally. Obviously, we have people internally. But we have basically as much demand as ever for talented creative people. We have a ton of work that we want to do. But we also have an unlimited demand for, Figma make and tools that help us instantiate ideas.

Starting point is 01:40:01 And there's a lot of stuff where we're like, we just need to do the footer or the layout or the buttons. And the faster we can get that out, the better. Yeah, but there's this sense of we have an insatiable demand for novel thinking and ideas and concepts. And it's less a, it's less a. I can't tell if you're saying we humanity or we TBPN because the answer is both. I'm talking about both, but TBPN.

Starting point is 01:40:27 And it's less of, oh, is this possible? Like, is making this animation going to be, you know, within budget? It's more like, well, what animation should we make? Yeah, did you call the good idea? We're willing to pay a lot for that kind of original thinking. But I'm curious, like, where you see the kind of, where sentiment is that today? Because on the software development side, it seems very clear as when you can build software faster and cheaper, what we want more engineers to build more software because the world needs a lot of software.

Starting point is 01:41:00 And I think same thing for truly great novel creative work. Yeah. Well, first of all, I'm hearing a strong advertisement on TBPN for TBPN. You all are trying to hire, so hopefully our viewers can like apply somewhere. But the thing that I would call out is I think that we're almost seen a parallel to the AAL.L. talent wars with design right now. At least amongst the companies that are really getting it, the ones that have internalized the design

Starting point is 01:41:34 is going to be how you win or lose. And that's the craft, the details that matter, they are really getting aggressive, not like meta-AI aggressive. Let me be clear, but very aggressive. And it's, I think just again, value moving up the stack. and that's something that all companies will eventually figure out. I just hope that a bunch of companies don't figure out too late.

Starting point is 01:42:05 But yeah, I think Jevin's paradox as applied to talent that is adept at using models and doing amazing work and really pushing the boundaries of what's possible and what could be created and how much craft you can have, how much you can delight people that's real. and the demand for development is certainly not shrinking. I think that it's never been met. And at Figma, I mean, we're doing headcount planning, like we're hiring almost all areas as we go into this next year and in big ways. And I think just generally, you know, our point of view from the start about AI at Figma

Starting point is 01:42:46 is like how do you both lower the floor, make it so that more people can come into the design process? but also raise the ceiling, make it so the designers can do more. And I think that's something that's a shared view. Maybe we were one of the first to articulate it, if not the first, but I believe that most of the products that are in the sort of AI space are trying to do the same thing. They're trying to make it so that more folks are able to be part of this life cycle, and also they're trying to make it to the people that are the experts

Starting point is 01:43:21 are able to do more than ever, for. Do you have a view on like the shape of the design community as a function of maybe the amount of designers working in a particular organization? Are there more design firms than ever? Are there more solo design firms that are profitable in making money than ever? Is the power law getting steeper or is the floor rising or both? Do you have any idea on the way? like the shape of the design community and the shape of how designers work together or just make money independently in the age of AI. Yeah, I think there's a lot of work here you've done on the sort of company side, and we're seeing there just more and more people getting involved in design and more design hiring.

Starting point is 01:44:18 Yeah. I think in that more freelancer agency segment, there's more research to do to understand is the same effect happening. But I'll maybe think back historically. You know, when there was this first wave of design hiring and people realized that design talent was critical to success, one of the first things they did was they started to snap up the agencies. And they would have literally just start acquiring agencies of great design.

Starting point is 01:44:48 talent. And so that's how, like, for example, T. Hanan Lax became involved in then Facebook, then reality labs, and John Lacks ended up leaning up reality labs for a while. And they got someone who was legendary design talent to be able to be at the forefront of their work on Oculus for quite some time. And I would not be surprised if similar things started having. happening. But I also think that, you know, back in the early days of Figma, it was this question even of like how many designers are there in the United States and the world. I mean, the beer labor statistics said that there were 250,000 designers. It's like, well, that seems wrong in 2012, but like how many other actually, I don't know, maybe it's like 500,000, not

Starting point is 01:45:43 250,000, but like, is it like a venture-skilled business? We don't know. And, but what we noticed was that values moved the stack, and it was going to be the case that everyone was going to build up their design team, and the design would be so important as this overall system changed. And I think that that same thesis we held then in 2012, it has not changed. It is the same thesis we hold today. and I really believe that sort of this era of the 10X designers here, it's just starting in some ways, but it's going to be quite incredible, and it'll require more designers to step up as leaders and for them to lean into leadership roles

Starting point is 01:46:34 that they may have otherwise not considered. But, you know, if they don't, then I think it's going to be a dynamic where everyone's trying to get involved in design, and they don't really know the guardrails or how to do it exactly. And you have a lot of people with a lot of thoughts, but not many like railways built, tracks built for folks to go down.

Starting point is 01:46:57 So that's the challenge, I think, that many design-oriented folks will have and the need for design leadership this time. Shopify just acquired a design studio last month, by the way. Someone in our chat mentioned this. What's your thinking? thinking around open source.

Starting point is 01:47:15 You obviously have a close relationship with Mike over at Anthropic. He's on the board. And obviously, Claude has consistently led on CodeGen, so they're a natural partner for FigmaMake. But is open source something you're exploring or just given that they're not really at the frontier? There was a take for a while that was like commoditize your compliments.

Starting point is 01:47:39 But maybe the LMS, yeah, I'm super interested to hear what you think. Yeah, I mean, look, there's, I think, first of all, I mean, Mikey and I, from the start of just looking at him being on the board, we had to make sure that, you know, is clear. We, as Figma have to choose whatever is best for Figma. Yeah. You know, as models get better, we get better. But if we're not choosing the right models, then we're not getting better. So, on 4.5, but through the Eval's easy choice. you know, roll it out to users.

Starting point is 01:48:16 There's going to be a lot of other model improvements. And of course, open source is something we're always watching too. And so it's exciting just how much is happening right now. And I think that it's going to be just a fun next few months to see the continued model releases. I'm pretty psyched. What are your thoughts on hardware right now? that feels like a place where we're seeing a reemergence, a lot of hype around

Starting point is 01:48:45 meta's new devices, potentially a new design surface as people design more physical products that are maybe enabled by AI. It just feels like that could be kind of the next thing, some sort of like Cambrian explosion of hardware devices, or maybe it's all just

Starting point is 01:49:06 AirPods forever or something. Who knows? but I'd love to hear your take on, like, how, is there a demand from you for new hardware devices? Like, the Waycom tablet is obviously legendary in the design community, but it's a completely prosumer, professional device. Just how are you thinking about hardware these days? Well, I think in the Figma context, what we think about is, okay, what are the screens that we need to serve and our users need to design for? And so that's why I was so excited to be able to check out Meta's new glasses and also the neural band. The band is like underappreciated. So cool.

Starting point is 01:49:51 People need to, people won't, it won't get a reaction until enough people have actually tried it because it's. It looks just like a Woot Band. Like there's nothing that looks crazy about it until you actually try it and you're like, oh, you can fully see what's going on. Yeah. Yeah. But like, let's be clear. I mean, this was years in development, many years. I mean, they've been at it for so long now. And I think at the first time I got the demo was almost a year and a half ago, a year and a few months ago. And then I saw it again right before config, but like the sort of new improved version. And it just takes a long time to go build these new form factors. You know, the same thing as being true in auto. It's true, you know, in wearables in general. It's probably going to become true in other surfaces that people will find.

Starting point is 01:50:48 And I think that whether it's AR, VR, VR, you know, curve glass, you know, small screen sizes, large screen sizes. There's a lot of surfaces that designers now need to target. And we haven't even gotten into, you know, AI and agents and what designers need to do to create context in these places too. I think in many ways we're just in the MS-DOS era of AI. We are still figuring out how to explore this amazing latent space. You've got this LM. It's your spaceship. But you're in this, like, n-dimensional space.

Starting point is 01:51:28 And the best way to, like, figure out where you are, your compass is your prompt, natural language. It's like what interfaces will be built for that? How it would be able to figure out custom bespoke ways to do dimensionality reduction, figure out how to help people navigate better as these models and the architecture is fundamentally improve. And also, you know, what protocols between the models should exist and what's the best way to create those? Because I think that there's a reason MCP is really caught on so fast. It's extremely valuable. And we just launched with Figma the Figma MCP for design as a remote service before it was only desktop and local.

Starting point is 01:52:10 And also we improved the capabilities. And then we did it for make as well. So we want to make sure that people know you're not trapped to make. Figma Make, if you're one prompt to code and you're making it so that you're getting to some great output and you're spending a lot of time on it, awesome. Like you always had download code, but now you can just have your MCP connect to Figma make and pull all your code in. How is MCP adoption going? It feels like there's a world where like I was promised like the AI didn't need an API because it would just use the computer, right? And I imagine that there's a maybe we're not quite there, but there's a world where my AI just literally opens a browser, goes to figma.com, does whatever it needs to do, just like a human.

Starting point is 01:52:55 And yet MCP seems to be like what you're identifying with like the energy around it is immense. How are people actually using it? Like are there is there a killer app, killer feature, some story where you've been really excited about like the impact that it's had on an actual like design project or designer or firm or your company? I mean, you should definitely try it out like it is a sort of magic experience when you go and you've got your design structure and a figmo's auto layout. got your variables to find. And then you're able to just, you know, basically press the button. And with MCP, you pull in all that design context into your codebase. Sure. And you can use inference to basically go figure out how to map it. And I'm not saying it's all one shot, but like with either one shot or a little bit of it or a prompting,

Starting point is 01:53:46 if you have your design set up in a structured way, it's a pretty wild experience. You know, Coinbase was telling us about how it's really improved. They're working. workflow and you know we've heard from many other customers that I'm not sure I have permission to name but how much it's affected them and how powerful it's been and overall yeah people are super excited we've been thrilled with the response that's why we're investing so much time into it is because we really think this is a super important part of our story as a company yeah is being more open and making sure that we really extend out to the greater ecosystem of AI.

Starting point is 01:54:26 There's so much that we've done in so many places. And design is context. And that context is valuable in so many different spots. So I should think of the MCP efforts as a almost like a business to consumer, like a consumer benefit, not necessarily just something like an API that enables like better like B2B partnerships and integrations. Is that roughly correct?

Starting point is 01:54:49 Yeah, I think it's, I think everyone's still trying to figure it out for sure. But I think that, yes, with the right MCP services out there, the ways that they can affect, you know, all sorts of places that you consume in France, it's pretty big. And I think we're just kind of at the start. So, yeah, for Figma, we're very,

Starting point is 01:55:20 very, very excited about where we can go here. And then, yeah, I think just, if you zoom out, the bigger story that is being told is how do we go from my data product? And it used to be this very linear process. But now we're no longer in that world. We're in a world where instead people try various directions, various paths. and you need to be able to generate a bunch of different ideas, explore them with your team, figure out that option space.

Starting point is 01:55:56 We hope Figma make me a big part of it. But even if someone has a different tool that they're using, we also want to make sure that they're able to bring the design context in. And with that design context really help that other surface improve. We think that's a win-win for everyone. and it's most importantly a win for the user. I'm so excited for today's middle schoolers and high schoolers to have access to these tools. Can you imagine if you had Figma make?

Starting point is 01:56:29 It's incredible. Like when you've had that age, like you let somebody marinate with these kind of tools for a decade, what they're going to be. I mean, I'm pretty sure we've already seen posts about like a 15-year-old sold a company for like $50 million or something. There's crazy stuff happening all over the place. But it's just going to be crazy more and more and more.

Starting point is 01:56:46 Thank you so much for joining. Last question. It will be wild. Last question from the chat. Do you miss the Rattie? Oh. Not really, no. What is it?

Starting point is 01:56:59 The University Dining Hall? Is that right? One of them. Yeah, I was more of a V-Dub guy. Okay. There we go. Yeah, the Radi was a great place to see people. Okay.

Starting point is 01:57:12 But then again, I also, like, I hate studying line. So I would time out of everything so that I either. arrived super early or super late. So I wasn't in the long line. But yeah, the food, maybe not the best, but, you know, the vibes, right? The memories.

Starting point is 01:57:29 Good vibes. Well, thank you so much for bringing good vibes to disappear. Yeah, massive week for you guys. Massive week. We'll talk to you soon, Dylan. We'll talk soon. Have a good one. Good to see you.

Starting point is 01:57:37 Quickly, let me tell you about fall, the generative media platform for developers, the world's best generative image, video and audio models all in one place. Develop and fine-tuned models. with serverless GPS and underline clusters. Shop by perplexity and many more. And we have our next guest from Stripe

Starting point is 01:57:53 in the Restream waiting room. We will bring in Jeff. It's time to Jeff. The TDPN Ultradome. Jeff, how you doing? Whoa, okay. We're purple. I'm seeing some announcement lighting.

Starting point is 01:58:03 What's going on? We're here in New York. We're finishing up some keynote practice for our event tomorrow, Stripe Tour. But I imagine you all want to talk about some agentic commerce today. It's looking wild in there. Yeah.

Starting point is 01:58:17 I get excited enough about payments even without the neon lighting, but that certainly helps. Yeah. Yeah. Give us the actual news item. It seems like Sam Altman just found out about Stripe recently and decided to do some sort of deal or something. Of course, like the companies go back a long way. And so my question is like, what's new about the relationship between the two companies today? Yeah, really exciting.

Starting point is 01:58:44 So today, for the first time ever, you can. purchase where you prompt. And inside of chat TVT, you're able to discover products as lots of people are already doing, but now you can buy from businesses right in the chat. And so we partnered with OpenAI to build this as well as a open standard agentic commerce protocol to help more businesses get going of selling agentically. So I'd love to get into all that. Yeah. Name every product I can buy today. I mean, give me an example, because I can imagine there are products where the company wraps in a different, you know, content management system or order management system. Yeah, and physical versus there are other companies where they sell stuff and it's really just stripe and a website.

Starting point is 01:59:33 And so that makes a ton of sense. So what's the wheelhouse customer interaction look like? Yeah. So anytime you discover a product in chat DBT, if that a seller is enabled on the agent of commerce protocol. protocol and is working with OpenAI, you'll be able to purchase directly with any of the payment methods you've already saved with Tad TBT for subscription or be able to add one. And what's really interesting is, you know, one of the big unlocks here is how do you scale adding merchants to agentic commerce and how do you scale accepting agentic payments? And we have two big announcements

Starting point is 02:00:08 that are going to unblock those foundations. One is the Agenda Commerce Protocol, which is an open standard that we co-developed with OpenAI for businesses to make their checkouts agent viable, agent ready. And so as a business, you're used to, you know, first you're selling in the mall, and then you're sold over the phone, and then you sold over the website. And now you're going to be able to express your checkout in a way that you want for agents to be able to initiate transactions. And the second big thing is, how do you actually move the money? How do you get the payment from the buyer and have it sent to the merchant for processing in a secure way, which is why we built this shared payment token API,

Starting point is 02:00:51 which lets businesses on Stripe and businesses that might be using another processor, be able to process agentic payments, and that those two things together are what really enable a launch like this. Getting more into the details, if for an e-commerce company in particular, like where it is, where is, where is, inventory managed is are you guys integrating i know shopify had an announcement today too is that connected to this are these two totally different experiences like help me understand how like you know

Starting point is 02:01:27 a specific company like an e-commerce brand is is going to uh can get started selling uh in line in in these lums yeah so today's announcement for chat tvt has Etsy live so today rolling out this morning you'll be able to purchase from Etsy sellers directly in the chat. And Shopify is coming soon in the same program. So these are the first two merchant partners, the first two e-commerce platforms that are working with chatGBT to be able to sell agentically through their LM system and their consumer base.

Starting point is 02:02:02 For merchants going forward, how do you at scale have a way to express your checkout as a merchant, as an e-commerce platform that isn't just, hey, have the agent like pretend to be on my site and click around to do so. How do you have a standard API system for scaling those merchant integrations? And that's what the agentic commerce protocol is. It's a way for any e-commerce platform, any business to be able to express their checkout in a way that an AI agent can then safely initiate the transaction. And we think that's a business. is a big unlock for how to scale a general commerce.

Starting point is 02:02:42 The world has already had lots of product feeds. There are many at-scale product feeds, but there isn't yet a way for a business to programmatically express their checkout in a way that they can control and have it be agent-initiatedable. How does the standard that you built differ from MCP? Is it related? Is it compatible? Are these two separate paths in the tech tree?

Starting point is 02:03:08 Yeah, we think that the agenda commerce, protocol is specifically designed to help with initiating agentic checkouts online. And it is dedicated to that type of commerce use case. It shares a lot of the same principles of being open and LOM first as MCP does for generic tool usage. But checkouts are complicated. They have all kinds of weird back and forths and coupons and skews

Starting point is 02:03:36 and inventory and payment failures and payment success. You need a dedicated, And that's why we're so happy to partner with chat TPT with open ad to both build the standard, but also have it be the way that they're going to scale their merchants. And so you can get started at agendicconverse.dev that URL happened to be available last week. So I might have bought it. And that's, and that's what went forward with. How does the Apple tax feature in all of this? It seems like I use chat GPT in my phone and Apple usually tries to take a cut if I buy something digital. If I say, on physical goods historically been exempt.

Starting point is 02:04:14 But if I'm buying a digital good, do you play nice with that? Like, how do you see that all playing out? Yeah, today's launch is focused on physical goods. And I think there's going to be a variety of ways that these third parties, third party services are able to charge in various ecosystems. But today's launch works on web, on iOS, on Android, and is rolling out to all U.S. chat chitbgmit consumers this morning. Very exciting.

Starting point is 02:04:40 Do you think people are underestimating how much of a shift this trend is going to be? Because we were explaining before you got on earlier, if I can go on my phone and describe in plain English the exact type of product that I want be served a number of options and in one click buy it, that's incredibly useful. I was giving the example of even browsing Amazon and trying to just find, if I'm trying to find a, the example I gave is like a paper. towel holder. This is personal for me because I was, I bought a really cheap direct to factory paper, you know, direct from factory paper towel holder. And I was just like, just give me, you know, it's not the place of skimp. Yeah, yeah, yeah, yeah, exactly. You use it every, you use it multiple times a day, right? But, but I, I've always wanted it almost a layer over Amazon to just, just products that have existed or from, or from brands that have existed for

Starting point is 02:05:36 more than 100 years, right? And this is the kind of thing that you can describe of, I want to buy a paper towel holder, but I want to buy it from a company that's over 100 years old, right? If they've been around for a... It's like your futuristic purchase is a... I want the legacy provider. It's a perfect combination.

Starting point is 02:05:52 I think you're expressing what many of us all feeling in our lives, which is the modality for discovery has completely changed. And we expect that commerce to get closer and closer to intent. And it's certainly having tested, having worked on and tested this for the last many months, it feels extremely natural to purchase where you are discovering. My garage is full of things I bought from Etsy as we passed over the last months.

Starting point is 02:06:22 And this is just the beginning, because with this type of foundational agented commerce set of APIs and open standards, it's really unlock to both scale merchants, scale AI agents, scale all types of new consumer and business experiences. And so I think it's very much just the beginning here, but we are, I think, unlocked, we hope,

Starting point is 02:06:46 how to take these things from prototype to at-scale production. And I really encourage folks who are watching today to look on CHAPGBT and try to find something fun that you've always wanted to buy that might have been on page 10,000 of your search engine, but might be like right there as the LM really can provide you very personalized. recommendations. It's just very fun to on the go buy in a tap in this kind of way. I love it.

Starting point is 02:07:13 Well, thank you so much for stopping by and coming as the update. I have, but probably stuff you can't even get into. But we'll have you on again soon to talk more. Congratulations. We'll talk to you soon, Jeff. Great to see you. Jeff. I see you off. And in the financial news, Etsy is up 15.83% on the news. It's a $7.42 billion dollar company, head over to public.com investing for those that take it seriously. They got multi-ass investing, industry-leading yields, and they're trusted by millions. Not financial advice. Our next guest, yeah, maybe you want to go long. Maybe you want to go short. 15%. But a big boost on Etsy products potentially being sold in chat. Well, we have Adam Draper

Starting point is 02:07:52 in the Restream waiting room. Let's bring him in. Whoa, there he is. Let's go. Let's go. What's happening? George, John, so excited to talk to you. Let's go. We're excited to have you. Give us a news. What's happening? All right.

Starting point is 02:08:11 Well, we raised $87 million. Actually, yeah, let's go. Let's go. Actually, what? Well, it was $87,654, 321, because that's how rockets go off. There we go. BoostV.V.C. It's a rocket. Yeah, and we get it, we get to continue to do what we're doing. Our investors love our returns, so we just get to keep doing what we're doing for the last 13 years, which is backing great founders on the French. Amazing. You were early to sci-fi investing, or maybe did it before it was so on trend?

Starting point is 02:08:56 I was an early investor in Coinbase, a company called Virgin Labs that became Snapchat Spectacles. I've been investing in what any kid would dream of for the last 15 years. I just flew a jetpack. No way. No way. Yeah, somebody sent me a video earlier. maybe it was the same company. Were you flying with Jackson Moses?

Starting point is 02:09:29 Is it gravity industries? Is that the company? I was flying. So, okay, this is the coolest thing. I, you know, I read my first Iron Man comic when I was like six. And then about a decade ago, we invested in Iron Man thesis, which ended up investing in gravity. There we go.

Starting point is 02:09:49 And I got to fly a jetpack today. So I think, I, I think, I heard there was some fundraising news. Yeah, yeah, John's back. Hit that, hit that gong, 87 million. There we go. Adam's the most fired up. You're the most fired up guests I think we've ever had.

Starting point is 02:10:13 Nobody else is jumping. You're at the standing desk. It's fantastic. I tell people. Thank you guys so much for doing this. guys are just the greatest cheerleaders and champions of this tech world. We appreciate it. We love, nobody loves technology. I would hate if somebody loved technology more than, more than us. Maybe you do, but break down the jet pack. How does this, like, it's a good example of like a

Starting point is 02:10:35 specific portfolio company. How does this go commercial? I think that the promise of venture capital, the hope of venture capital is that you get to invest in all the dreams that anyone ever had as a 12-year-old, and they end up coming through, and I got to do that with a jetpack. I invested in exactly what my childhood self would have loved to have invested in, except it was seven years ago, and then they made insane progress to the point where

Starting point is 02:11:08 not only one person can do it, anyone can do it. I just flew down to back. I got to Baker's Field, and then I got to fly a jetpack. And so I am. Okay, okay. I love everything about jet packs. How it is gravity specifically? How are they looking to commercialize?

Starting point is 02:11:29 You know, what are some of the use cases? Is it going to be the kind of thing that is more initially, you know, consumer? It's just fun to fly a jet pack, so people pay to do it. One of the things I always noticed when you guys are doing your thing is that your cans pile up. So I want to make sure fit in. There we go. You got double fist. So we, so first off, we invest in all founders who are on the fringe of doing great, incredible things.

Starting point is 02:11:58 Gravity specifically, we saw the, I saw a man fly. So he flew in my back parking lot back here in San Mateo. And when someone flies, you got to write him a check. You write a check right there. And so we wrote a check right there. Seven years later, he's built a, you know, multimillion dollar business. And it has two parts to it. There's the defense part and the media part.

Starting point is 02:12:23 And also this training school so that people like I could become pilots, jetpack pilots. And basically, I highly recommend, we should go do a, we should go. That sounds fantastic. I've always wanted to have a jetpack in the studio in the Ultradown. Yeah, for a really big fundraising. What's the, what's the defense use case here in your view? because I can imagine in sort of more controlled environments, this could be a cool way to turn a human into a drone.

Starting point is 02:12:54 But at the same time, on the battlefield, I don't know, with a lot of drones, I don't know if I'd want to strap myself in and get into the air, just considering you might be a target. But how do you view the defense opportunity? It turns out boarding boats, it turns out, is one of the best defense capabilities. So being able to go from one boat to another boat. You see that with like the Navy SEALs show up.

Starting point is 02:13:15 in the little skiff. Yeah, that was the other demo they did, right? Yeah, that was that was crazy. Yeah, that was insane. So gravity, every time I've seen a jetpack, basically the last seven years, it was gravity. It's hilarious. And it's really an innovation in vertical takeoff and landing engines. People didn't believe there was a disbelief that you could build small enough engines

Starting point is 02:13:35 that they would be able to lift an individual human off the ground with enough fuel to live for long enough. And so this has been a huge innovation. Also, paramedic is a big deal. Like, if someone's lost on a mountain, you can just fly up there within a small amount of time and get it. And then the, yeah, dude, I mean, I, I, it was a lifelong dream to fly. I flew. I'm a jet pack racer. Jetpack racer.

Starting point is 02:14:05 I don't know how to say this. I don't know how to say this in Italy. Just deploy, deploy the fund in like the next jet pack stuff. Yeah. Like, it could be bio. jet pack things. It can be jet things. Just it needs to have jet packs, evidently. Amazing. Last question. What, what's the next category that is maybe if jet packs were underhyped seven years ago, what, what are you looking at today? These founders are coming into my office and they're

Starting point is 02:14:35 saying, I'm going to make you live 50 years longer. I'm going to cure cancer. I'm going to just cure all genetic disease. We're going to move from a work. of treatment to a world of cures in the healthcare and science world. So I would say bio is really the thing that I would say. Also, one of the things we do as BoostVC is we actually put on a deep tech demo day with the entire deep tech early stage venture community and it's coming up on Wednesday where you can see some of these bio future forward companies. Actually, a bunch of the demo day companies have actually been on TVPN.

Starting point is 02:15:12 Doug from Radiance There we go Incredible people Gravity There we go I expect to see you flying around the floor Yeah

Starting point is 02:15:22 Of the demo Okay we'll go down together We'll do it together Love it Yeah you need a jetpack into the show The chat is demanding it Thank you so much for hopping on the stream We'll talk to you soon

Starting point is 02:15:31 Great having you Adam Congrats on the new fun See you soon Thank you We just gotta keep backing the founders Amazing incredible Talk soon Talk soon

Starting point is 02:15:38 What you about Turbo puffer Search every byte Serverless vector full-text search built from first principles and object storage fast 10x cheaper and extremely scalable we puffin our next tomorrow live in the tvpterdom tomorrow can't wait but for now we have james hawkins in the restream waiting room from post hoc oh god post hoc here he is how you doing uh yeah very well thank you very much uh give us the fundraiser news quick let's amp it up let's amp it up what you got for us sure well first of all i am very excited to announce that pineapple does not belong on

Starting point is 02:16:12 pizza. Whoa. Whoa. Shots fired. I love pineapple on pizza. That's- I don't. I don't. I'm riding with James here. Okay, well then you can hit the gong because I'll like this guy now. Yeah, yeah, yeah. He's my mortal enemy. Now, I understand it's an acquired taste. What inspired that? What? Timeline debating that? That, that's an eternal point of debate. It's fine. It's not a debate. It's decided. What about hot dog? Is a hot dog a sandwich? think that's true. What if you eat it like this? What if you turn it 90 degrees? Then it looks like a sandwich. I feel like I'm not America. I have a less strong opinion on this, I think. I think

Starting point is 02:16:52 you guys are more. If the dog wore pants, would it wear them on the two back legs or across all four legs? Wow. You're not straying from the hard questions. These are the hard questions. I've got one more. I've got one more thing. Please. I'm very pleased to announce that we have raised our series E. Woo! 1.4 billion. Congratulations. Massive. What unlocked it, walk through the business, what's working, the customers, the problem set, everything. Sure. So Postlog is a very developer-focused set of analytics tools and feature flags and stuff at the moment. We try and help you understand how to build a better product, basically. So we help you understand your customers in lots of

Starting point is 02:17:35 different ways, like errors, analytics. We have 16 products. What's working well? 16 products should be hitting and gone for that. I think about 10 or 11 that are charged for and the rest of production. That's right.

Starting point is 02:17:49 So yeah, we've been busy bees building stuff. What's working well? We've got tons of engineers who we have really looked after. We think there's been a big rise in product engineering,

Starting point is 02:17:58 engineers who want to decide what to work on. And so we've been able to sign up just an awful lot of them. I think we have about 240,000 customers across like free and paid. And so we're kind of like a heavy product-led growth

Starting point is 02:18:10 from engineers just installing our stuff, basically. Some of these tools I imagine would be fertile grounds for hyperscalor competition. What's been the secret to maintaining differentiation against the big cloud platforms? All the products in one has been the main thing, going very, very wide. Because the way we think of it, there's lots of tools that help you. There are feature flag products, our experimentation products, their analytics, six tools, but it's the same customer underneath all of these. And so we kind of felt it's just better. It is more important to our users to have all of their custom data in one spot

Starting point is 02:18:50 and lots of things you can do off the back of that than it is to buy like 15 point solutions, basically. What's your take on the idea of the compound startup versus iterating and adding one product after the other? How focused do you want to be on a new product at one? time. Do you want to orient the entire company around the next product or do you want to have, you know, let's just get a bunch of things started. We'll iterate towards greatness and all of them simultaneously. Much more the latter. So we start a couple of products at a time, but we always have a small team that we leave behind on each product. So back in the early days when we're smaller, we would go, hey, new shiny thing, everyone would move over. Yep. But then as the existing one got

Starting point is 02:19:34 more and more growth, we'd start, we just weren't very responsive to customers and so on. It didn't build up with the other people properly. So we have tiny little teams. Like the average team size of postdocs probably three people on each product. It's just like three engineers deciding what to build. We just try and get out of the way. So we're fiercely anti-bureaucracy internally. So yeah, this small team structure has worked great, like very flat, very wide and a lot of autonomy for engineers. Yeah, when you're when you're thinking about creating a new product, what is what is the bar in terms of turning something from an experiment? You know, at what point would you would you launch something and maybe shut it down or or if you're launching or by the time you're launching

Starting point is 02:20:15 something are you committed to maintaining it over the long run? I'm curious what kind of the framework is. Sure. Yeah, we see really, we started at the bottom end of the market. So the bar to us launching something's pretty low. Like we'll start targeting two-person startups kind of doing YC for example. It's kind of start from the start is a praise you hear a lot internally. In terms of retiring stuff, we haven't had to retire a product at this stage. We've already built. kind of products where we're going in with a late mover advantage, we consider it. So we're going in after a lot of the market. This is actually a behaviour that we're going to change now with AI in particular,

Starting point is 02:20:49 because we're looking at figuring out how can we use like this sort of multidimensional data and all these different tools together, which is one of the challenges. And that's partly why we did this round, because we wanted to have the time and space and the ability to kind of go nuts, trying to automate. The way I would kind of consider it is a bit like if you're trying to understand a painting and you can only see the color blue, like one data type. you'll get a rough idea, but it won't get a totally correct idea of what's happening. And so we kind of think having all the tools and all the data is a bit like having all the

Starting point is 02:21:15 colors. But this is something that's new to us, and it will take a little bit of time and some confidence. So that's kind of, yeah, it's probably the main driving force behind wanting to raise at this point. Last question for me. How has AI changed the business over the last few years? There's so many different ways that you can launch AI products, you can use AI to speed up product development. There's a million different ways. but what's been the biggest thing that's actually worked for you?

Starting point is 02:21:41 It's massively sped up product development because of so many products at once. It's given us a big, it's just, we already thought that was happening because there's more open source software available, and it's just ramped that fully. So now kind of, we even have companies quite regularly emailing a smaller startups,

Starting point is 02:21:56 saying like, hey, would you consider buying a company? We don't know how to compete with compound approaches in our industry. Like we've become a feature of a bigger platform, for example. And so I think we got lucky with, that respect. We are targeting engineers and their workflow is entirely changing. And so we're trying to move towards the world where we're generating kind of poor requests for people based on like our data, a session, like video clips

Starting point is 02:22:18 of them using their products, like their feedback and everything else that we're kind of feeding in. So yeah, it's a totally dramatic change for us. I think we've gone from like here honestly to like in complete excitement. Amazing. That's great. One note. Chat absolutely loves the website. You should look at it John, it's absolutely stunning. It's probably the best iteration I've seen of building like a desktop and making a website just like a desktop, right? Sure, sure. Oh, yeah. This is really good. Absolutely stunning. One question was Post-Hog a hot company at YC Demo Day or did it take time to get people to kind of come around to the team

Starting point is 02:23:01 and the opportunity? We actually had, I think, relative to our batch, we're doing quite well. lot of traction, just free open source usage. We just had an open source project to start off with. But our seed round was a train wreck to raise. It was like March 2020. And so I was like, man, we got this thing done in like three days. And then I was like, I may not live to see the end of this. So yeah, we made like, you know, we made a bunch of people who put in like round little five pay checks a bunch of money, which is actually one of the most fun parts of them honest. So yeah, the seed round was very hard. Later rounds, it kind of got easier. like the series A was like Google Ventures

Starting point is 02:23:40 preempted and we went with it because we there was so much kind of fear in the market fundraising at that time and then the later round's all got much easier because we just had kind of obvious traction. We're quite an efficient business because we're inbound and so we haven't really needed money and that's our first rule of fundraising is do not need to fundraise but we're not against fundraising and that served us extremely well later on

Starting point is 02:24:00 but yeah the start the first round was the hardest by Miles. Fantastic. It's such an important lesson for anyone out there. It's like if your seed round is brutal, you know, it's not necessarily. It can be an indicator that you're on the totally wrong path. It can be an indicator that people just don't understand exactly the opportunity or just how good the team is. So thank you for coming on and giving the update and congrats on the round.

Starting point is 02:24:23 Thanks for coming. We'll talk to you soon. Great to meet you, James. Bye. Cheers. Let me tell you about linear. Linear is a purpose-built tool for building and planning products. You heard it from him.

Starting point is 02:24:33 They built a lot of products. Meet the system for modern software development, streamline issues, projects, and product roadmaps. People forget they created modern, they basically created the modern software as a service website. That's true, that's true. How many thousands of companies have tried to emulate it, but... Well, up next, we have Eric, the CEO of Modal Labs. He's in the Restream Waiting Room, coming in to the TBPN Ultradem. Welcome to the show.

Starting point is 02:24:56 How are you doing Eric? What's happening? Hey, how's it going to be here? It's great. It's great to finally have you. Yeah. Would you mind kicking us off an introduction on yourself in the kind of the conference? company?

Starting point is 02:25:05 Yeah, yeah. So we build infrastructure for AI. And so basically if you think about a lot of sort of traditional AI, or sort of traditional infrastructure, you know, things like Kubernetes and Docker, things like that, it really doesn't work well for all these new AI applications, you know, whether that's, you know, generated media or like large language models or things like that. So we basically built a whole new software layer that makes it a lot easier for developers to build applications on top of that.

Starting point is 02:25:30 Where does that sit in the stack? Like are you trying to be the software layer? at Stargate at like these massive, massive data centers, or is there a different customer archetype that you can actually deliver value for at like a smaller scale? How does the market actually bifurcate? Yeah, so we focus on slightly smaller companies. Initially, we saw a lot of PMF with startups.

Starting point is 02:25:55 But in the last year or so, we've seen a lot of traction also with later stage companies and, you know, public companies and enterprise companies. What's the pricing model for something like this? Yeah, yeah. What's the pricing model for something like this? So it's all usage-based. Okay.

Starting point is 02:26:08 Which means, you know, sort of traditional, basically like a traditional cloud vendor, like we charge per GPUR essentially. Sure, sure. And then do you offer like abstraction layers on top of this? Do you want to build a platform at some point? Or are you happy vending it into another cloud partner that, where the customer might not even be aware of you? How do you sit in the market over the long term? No, we offer a very fat layer of. like all these software abstraction that focus on engineers writing code.

Starting point is 02:26:39 We consider ourselves a high code platform, right? It's meant for engineers who want to write code and, you know, they need a better infrastructure that maybe traditional tools could offer them in terms of scaling up and down, working with GPUs, working with capacitable over the world. Sure. And they want to write code. What are the, I mean, what's the biggest, like, nightmare story from misconfigured infrastructure or, like,

Starting point is 02:27:03 the problem you're solving. Have you seen just like millions of dollars go up and smoke? Is it pretty rare? What's the shape of failure in the AI? What are the stakes right now? I don't know if I could recall anything. We're pretty good at like catching.

Starting point is 02:27:20 I mean, in a couple of cases, customers. I don't mean with your software. Yeah, it's more like company, you know, people coming to you, having had issues in the past, figuring this stuff on their own. I would say like the biggest thing right now It's just like people wasting a lot of money on GPUs, especially I think a year or two ago.

Starting point is 02:27:36 Like there was just like kind of hyped up like scarcity of like you got to get like a thousand GPUs. So a lot of companies went out and made massive reservations. And now they're sitting on them and like they're underutilized. Right. So they're like, you know, what can I do with these? Yeah. Is there is there a value for those folks just to like sell it back as spot instances for someone else? Like what our company is actually doing when they wind up a little bit too GPU rich?

Starting point is 02:28:02 I know there's a couple of other companies. I think it's like, thrill is one of that. It's not like a space where it's super focused on. We kind of think that model is kind of broken, like the idea that companies go out and make big reservation. We think companies should just pay for is exactly what they use. And so that's something model does really well.

Starting point is 02:28:19 We scale up and down and you only pay for the time the GPUs are actually running. Yeah. And, yep. Yeah, what's your view on the GPU war? NVIDIA is obviously dominant. AMD has been in the year of the comeback. They're taking feedback from George Hots. They're listening to Dylan Patel.

Starting point is 02:28:37 What does the race look like? There's also a lot of A6 companies. There's broadcoms doing stuff. What does the shape of the industry look like in the next couple of years? Where are you kind of, where are you partnering up and investing for the long term?

Starting point is 02:28:55 So I think in the next couple of years it's going to be all in video. Like our customers only want Nvidia. Invita is like such a massive advantage in terms of this software ecosystem. But I think look a couple of years beyond that, like that's when it starts to get more hazy. I'm personally quite bullish on TPUs, for instance, like Google's product. But of course, I'll say AMD like you mentioned in other players. Yeah.

Starting point is 02:29:16 Are there other, like anything that you're looking at in terms of like abstracting away some of Kuta's advantage? It just feels like we're in this era of vibe coding. You should be able to take some sort of model and just port it, but that doesn't seem to be working. But people are certainly, they have a lot of economic incentive to do that. Are you bullish on any of that kind of like breaking the Kuda lock-in? Yes. It's funny because people are like, Kuda is like a moat. It's such an amazing, but like I don't know, like for anyone who's like actually try to write Kuda.

Starting point is 02:29:55 It like fucking sucks. It's very hard to work with. work with. So I don't know why it's been so hard for AMD or for other providers to build a better software later because Kuta is very hard to use. Yeah. So yeah, give it a couple years. Like I would hope that there's better ways to write these kernels. Yeah, I just remember the story of like Facebook was written in all PHP. They wanted to move C++ plus. They didn't want to have everyone rewrite everything. So they just wrote a compiler that compiled PHP. It's hard. Yeah, that's it. Yeah. That's it. And so it's like couldn't you couldn't we?

Starting point is 02:30:27 get to a world where you're just writing once and then compiling for any different GPU, like create a new abstraction layer. I don't know how the economic incentives play out. I haven't really dug into it, but it's an interesting industry. Yeah, it comes down to... Sort of what modulars trying to do in a way. Okay. Yeah, it comes down. It's hard to be... We're at a little, right? Sorry, every minute of... Yeah, I was just saying it's hard to be, like, extremely bullish on coding agents and also believe that Kuda has this, like, long-term, hyper-durable, you know, moat. There does seem to be some cognitive dissonance there, right? Like, they both can't be true. It's like why can't I run a 30-hour Claude Code prompt at some point or clod

Starting point is 02:31:02 code in the year 2030 and just like, you know, rewrite this for AMD. Don't make mistakes. And it just doesn't. I would hope that's how it plays out. But who knows? I mean, we might be years away. I generally like your. I've always enjoyed enjoyed your takes and would love to have you back on some time when we have more time. Yeah. But congrats on the fundraise. John, did you want to do the honor? Oh, absolutely. Yeah, give us the details. What's the fundraising news? billion dollars. 87 billion soon.

Starting point is 02:31:32 87 million Series B. Yes, so Lux is leading existing investors putting a bunch of money to, yeah, we're super pumped. Great stuff, Eric. Thank you for joining and have a great rest of your day. Talk soon.

Starting point is 02:31:50 Have a great rest of your day. Let me tell you about numeral sales tax on autopilot spend less than five minutes per month on sales tax compliance. for the Numeral HQ to get started. Well, on that note, I believe we have our first in-person guest, the day. The day of the week. From App Lovin.

Starting point is 02:32:09 Welcome to the show, Adam. Thanks so much for joining us. Thanks so much for taking the time to come on down to the TBPN Ultradown. Welcome to the show. Here we are. Great to have you. Would love to have you kick us off with an introduction on yourself and the company. Just kind of set the table for us.

Starting point is 02:32:27 Yeah, totally. We're probably the biggest company in the world that no one actually understands what we do. That's a good title thumbnail for some put-bay people. I guess it's bullish now if people don't understand what you do, right? Palance, Valencia. Yeah, same thing. Let's break it down. So we started the company in 2012 advertising inside mobile games.

Starting point is 02:32:45 And really the goal back then hasn't changed to now. We wanted to give advertisers the opportunity to market themselves, but do it on a revenue-based pricing model. So they can come in and say, look, I want to generate $1,000 of revenue. by day 30, be able to get me $1,000 of media spend against that so I can break even, make a profit, I'll scale unlimited if I can do that. And I've been in ads for 20 years. I hated the construct of selling someone to convince them to run advertising. You don't know what works, what doesn't, you try to convince them an audience was there.

Starting point is 02:33:17 It's all a bunch of make-believe until you go to top-of-funnel revenue-based pricing. And so we ended up working with a lot of game developers. These are companies all over the world, a lot of smaller indie businesses. And when you're working with a customer where the CEO is the founder and they're on the front lines of marketing, it's a great dynamic because usually they're technical. They want to actually understand the math around marketing. And so we built our business really being inside this niche, working with these casual mobile gaming developers all over the world.

Starting point is 02:33:48 Well, business started growing a ton. We were serving a video advertisement inside their games for other games. And over the decade that we ran up until going public, this business is super lucrative, super successful for our customers. We helped these game developers really build their businesses, but no one had heard of us anywhere. We weren't beat back, right? Well, not by desire. I'll say we're one of the biggest misses for Sand Hill because we couldn't raise a million over four back in 2012. This was after we launched the product, business was growing like a weed.

Starting point is 02:34:22 We were doubling every single month. What was the pushback? Do people just think like, oh, Google will take it all? I think I'm a horrible seller. I don't know. I mean, I've worked on the skills over the years, but the pushback was Google will eliminate you. If it's not Google, Facebook will eliminate you. If it's not Facebook, here comes Amazon.

Starting point is 02:34:39 So it was, why is this goofy-named company going to be able to do well inside advertising? Really quickly, you said you've been selling ads for 20 years. 2005 is the start of this? What's the first ad you sold? We were affiliate marketers back in social. So a couple of us, one of... And that's how the most hardcore marketers started out an affiliate because it's just you eat what you kill.

Starting point is 02:35:01 If you can make money an affiliate, you can make money doing anything. It is brutally tough. That's amazing. And what was the mood in Silicon Valley or the actual games that were the backbone of the launch? Is this the Zinga era, post-Zinga? This is the tough part. It's a lot of casual games. Yeah.

Starting point is 02:35:18 Like the environment we have now is a billion plus daily active play. playing solitaire and puzzles and words. Match three. Yeah, totally puzzles. So if you walk down an aisle of a plane, you'll notice like puzzles on everyone's phone. Those are the types of people playing. It turns out it's a lot of middle-aged women.

Starting point is 02:35:36 It skews female. It's a really strong audience of people, but a lot of us don't relate to that. So you think mobile gamer, well, shooter games and like a traditional gamer. And so that was always one of the things we had to, there was a challenge in the business is, how can that audience be good for, especially as we get into e-commerce and talk about our launch,

Starting point is 02:35:56 for broader brands, because people don't realize. Well, they control so much purchasing. I was about to say. It's all the heads of households. We're on the platform. We only have adults. We don't work on any child apps. So you've got adults, head of households, skew female, people with money.

Starting point is 02:36:10 These are people on a thousand dollar phone playing for 45 minutes a day. It's a perfect audience. People just never realized it until we became more mainstream. Yeah. How do you think about, there's this critique going around, with Open AI, doing deals with Oracle, doing deals with NVIDIA, and there's this like circularity to the economy. But then, of course, there's outside demand.

Starting point is 02:36:29 People want to use the products. How do you think about circularity within the mobile app ecosystem where a lot, I'm sure you have advertisers who are advertising mobile games and other mobile games, and it feels like, oh, well, like, you know, is this just all circular? When does the actual money come in? Yeah. How have you addressed that throughout the history of the company and like, where do we stand now? I mean, again, you've got to find a purchaser at some point in that circularity, right?

Starting point is 02:36:55 So if we were just advertising mobile games to mobile games that were ad-supported, it would be a house of card. At the end of that, you have two inside gaming, two things you're trying to do. Take a user from a game where they're playing X time per day and take them to a game that they're going to be more engaged with than play 2x or 1.5x. You can do that. You create more ad inventory. So the real growth driver in the category is increased supply.

Starting point is 02:37:17 It's not any different than social networking. You get more ads viewed. then you got to make more use of the advertising. So as the technologies get more powerful, you can match up the advertiser and the customer together for something that's transactional. So you get them going from a solitaire to candy crush. They pay $2,000 a year.

Starting point is 02:37:35 You've created transactional value in the middle there. There's a lot more value than where the users started. Yeah. How did the company actually scale? Like how like meat and potatoes were the first deals and then I imagine everything's like, you know, full tech platform now, but we're the first advertisers who were on board on the platform kind of like a handshake and a contract.

Starting point is 02:37:58 Yeah. How do you actually get inventory? How do you play this video in your app at this moment and we'll give you, you know. Yeah, how much of like doing things that don't scale? There was a ton of wheeling and dealing. Actually, it's cool being in your office because we were a garage made into an office. Yeah. We had a gong and like,

Starting point is 02:38:14 no, there you go. One of the guys who runs my business team. and now runs growth, slept in the office for the first six months. So we were affiliates, right? We learned dealers. So you'd go up to the early game developer and just give them a value prop of you can actually buy an install. Back then, there was not even the chance of buying an install. Now I said sell them revenue.

Starting point is 02:38:34 You're talking about 13 years ago, being able to sell them an install on a pricing model was innovative. And so when we went out to the market and pitched the zingas of the world, instead of just buying ads and not really knowing what happens, buy install, that in itself was, oh, wow, that's great. And then you paired that with video. The power of the video clip on a mobile device is really understated. 35 seconds of average viewing time on our ads, you can't do anything else. So when you get a brand opportunity to place an ad on a mobile device and the user's not distracted.

Starting point is 02:39:05 Non-skippable. Well, it's a mix of skippable and non-skillable, but half the ads, the user actually, it's a great point. Opt-in to watching the ad for a currency inside the game. Sure. That's up to 60 seconds of viewing time unskippable. You can't do all that. other things. So we live in an ADD-rich environment. This is the most non-A-D thing anywhere in the world. You've got to watch that advertisement. Yeah, that's fascinating. How do you think AI is going to

Starting point is 02:39:29 change mobile gaming? I've seen there's these famous ads on Instagram where they show you this character that's right. I'm sure you've seen these. And then people are like, but that's not the real game. It's a pick three underneath. And then I found someone actually built the real game because game development's getting cheaper. It feels like we might be entering like a new, there's a couple companies that have been pitching like vibe coding platforms for games that will, you know, like become their own app store. It feels like we could just be entering like a new primordia or Cambrian explosion of gaming. How are you thinking about like vibe coded games or more custom software? Obviously they'll all need ads to power this, but how are you thinking about it? I mean,

Starting point is 02:40:09 short answer to it for me, right? Like we create discovery. So the more content that's out there, the better. So if AI creates an environment where anyone can build a game, and we're all creative, and most people, he knows now today have played games at some point in their life, everyone can write an idea down, create a game, they're going to need discovery on the game, they're going to need modernization on the game. So that's a great opportunity for us as an ad platform. More broadly, our system is predicated on our model actually working. So we always talk about AI as LLMs, but if you just talk about AI as modern usage in neural nets, recommendation systems are one of the most economically viable use cases for AI today.

Starting point is 02:40:47 Yeah. They power Facebook, Instagram, TikTok. They power our system. This is why we always felt like, you know, Zuck, you know, spending however many hundreds of billions, like it was very easy to underwrite because you could just get, there's a lot of ways to just get benefit from this infrastructure spend in the core advertising business, right?

Starting point is 02:41:08 Totally. I mean, there's a simple function to all these models. It's more data, more complex model, more computer. better output. And if your whole business model is built on an advertising business, that this technology continue to evolve at the pace it is, is fantastic. Yeah. How do you think about AI on the ad side? Everyone's kind of predicting, I mean, there's already probably some tests out there that a lot of what Meta's doing will be just look at your product catalog, look at your inventory, and then generate the video

Starting point is 02:41:37 creative, generate the audio, generate the text. How are you thinking about implementing that? Are you already running tests? How have the test gone? This is one of the reasons investors are super excited about advertising platforms like ours. It's the simplistic of you've got a shop, you've got a catalog, create an optimized appearance of that catalog, through a recommendation system. We already do that. The more complex is today people upload a video ad into our platform, not always adapted for the platform. We're new in the marketplace, especially in shopping.

Starting point is 02:42:05 So a lot of times they'll bring a 10-second clip from Facebook uploaded to us. Well, if you could have gotten 45 seconds of viewer time, that sucks. as a brand, that's a total miss. They don't have the resources at every level to go, let us create 100 ads a week for this company now will rebrand as Axon. And we can talk about that in a second, but this Axon advertising platform. They just don't have the resources to create that ad count

Starting point is 02:42:28 at the scale of the time that you need into our platform. And so when you can apply LLMs to go, here's a couple cool ads out there, let me create 100 versions of this so we can put it into the system. That'll be a huge uplift and user response to all the brands, products that are being marketed on our platform. Yeah, when you think about, we're very close with the Ridge Wallet team. And so I got to early on in the Ridge days, and I know that they're an App-loven customer,

Starting point is 02:42:57 but early on I would just see the effort that Connor, the CMO, is just this constant treadmill of trying to make more and more creative, never being able to make enough, knowing that the more creative you make, the more money you'll make, but still just not even having the time in the day. And so thinking about being able to multiply out creative across different user types, different games, being able to update it a lot more rapidly as games develop new features, bring back old users,

Starting point is 02:43:21 et cetera. It's wild. Wanted to ask how you're thinking about MNA this year, next year in the future. Market Cap has grown tremendously. And you guys are now in a position where you could. there's a variety of different platforms out there that have a lot of attention that are maybe under monetized or at least not monetizing as well. I'm curious if that's at all something you guys

Starting point is 02:43:49 think about or the opportunity in, you know, gaming is big enough that it makes sense. You're thinking specifically of like a Snapchat or a Pinterest or what's platforms. Yeah, or I mean, TikTok was in the news last week. You guys were, you know, and that felt like a transaction that was more of like a political transaction than a purely economic deal. Yeah. But I guess the broad question is. like vertical integration, do you want to own games at some point or where media is developed or social media platform? Or does that just make no sense for you?

Starting point is 02:44:17 Well, I mean, like, we bit on TikTok. So the industrial logic of we can monetize inventory better with our ads models than anyone else out there is sound, at least to us. But there's a couple things that can strain us. One, our culture is pretty unique. We've got very few people doing an exceptional large amount of output. Company is still under 1,000 people. So you just don't tend to see scale like ours at that.

Starting point is 02:44:38 That's crazy. Headcount. And so putting companies together into that is really tough. The other piece is we're really excited about the organic growth opportunities in front of us. We're at a point now, I mean, I put out a couple months ago that early in the year we were $11 billion of ad spend on our platform, plus now we've grown a couple quarters, so a bigger number today. And we only work with somewhere between 1,000 or 2,000 advertisers on the platform.

Starting point is 02:45:03 We've been completely managed. We haven't opened up the platform. So if you think about that, when Facebook opened up ads manager, you know, we're going to or they had $5 billion a year of revenue, and we've got more than double that today on a very large platform. We even even opened it up. We're going to open up our platform,

Starting point is 02:45:17 the Axon Advertising platform in two days. So once we take the advertiser account from the very low thousands to the tens, hundreds of thousands, and then subsequently millions, we think the opportunity in front of us as many quarters to possibly years. And so distracting, you know,

Starting point is 02:45:33 it would have been cool on TikTok. It's a solid asset, at least from observing from the outside. But the opportunity just immediately with the business is so massive that you may as well just focus on what you guys already do. Totally. And with a lean team, we don't have the resources to distract ourselves. So it's full speed ahead.

Starting point is 02:45:55 We want to give our solution to all the advertisers in the world. We know that we've made a lot of businesses inside gaming more successful. We want to do that across the board. We've seen it in a very small pilot in e-commerce over the last year. the few hundred like Ridge Wallet and others that you said, they're seeing a lot of spend on our platform comparable to what they're getting on meta. And it's on us to execute to get this in front of all the advertisers out there.

Starting point is 02:46:18 If we can do that and become another destination like meta where there's a ton of scale in its top of funnel, the bottom funnel revenue priced, all these businesses around us are going to grow. So we don't want to distract ourselves from that opportunity. How are you rallying the team around Q4? I'm assuming Axon is timed with the, the championship season for every retail brand, but like, what does it look? Is it going to be people

Starting point is 02:46:43 sleeping in the office more? I mean, there's been a bunch of that. It's around the clock right now, because we've been closed and managed, right? So opening up an ad platform at our scale is a pretty complex undertaking. You worry about like, I mean, fraud is an obvious case. You don't want to see fraud. You don't want to see scammy ads on your platform. That's the first thing that happens. And like, we're probably arguably the biggest ad platform that's ever opened up self-service in any sort of form at this at this moment and so when you've got that you got to do a bunch of work building tools and controls to make sure the quality of advertiser and all the checks you have in place are as automated as they possibly can be and as good as it would have been in a human managed

Starting point is 02:47:23 environment so now we've got that you've got user experience you need to make sure a customer can go live on their own all the way through to being able to pay you and so we'd have build a whole bunch of tools to get this right we've been doing that for the last year we've and tuning it with the customers on the platform. So the team is, no joke, working around the clock right now to get this out the door in a couple days. Yeah, you can't, you don't have the luxury of missing Q4 if you want to open up an ad, you know, in the ad business.

Starting point is 02:47:50 That's the most critical time. How are you thinking about compute infrastructure? I feel like we're seeing meta build these massive clusters and it feels like maybe that's because they're going to be generating so many ads or doing so much gen AI that they'll need a ton of compute. They still need a ton of core AI just to do ad matching. Have you had to do on-premise deployments, cloud stuff? How do you think about building out your infrastructure for the long-term and potentially like a very different just economic dynamic around how compute it like falls on the income statement? Yeah, I mean, we work with Google Cloud. So it runs through our income statement.

Starting point is 02:48:28 So it's easy for investors to understand because we're not capitalizing the cost. Because we're so lean and because we were bootstrapped maybe going back to the beginning we think about every dollar we spend anywhere across the organization as everyone's a founder you're spending money out of your own pocket and this holds true with compute as well i went into my team and said look we're going to go spend five billion dollars by a bunch of compute and in our scale we can't do 100 billion dollars in investment so let's say five billion we'd be wasting that money and it would make them uncomfortable we try to constrain it to to trail or maybe be on par with revenue growth so that it becomes a cost of your future revenue growth, but what does that mean? If our engineers create a more complex

Starting point is 02:49:08 model, they could consume more GPUs. If we had unlimited capacity, they just deploy the model. Well, in today's world, if they see more GPU consumption, and they go, holy crap, this is fantastic in result, but is going to consume too much GPUs, we just don't have it, they've got to go optimize their code. And you may get like, you get, okay, maybe we lose one point on accuracy, but the cost all of a sudden goes down 90%. So across the org, we're wired to 3%. And so across the org, we're wired to think about things that way. Optimize and automate and cut costs as much as possible so we can have the best business so we can fund our mission long term. Makes a ton of sense. Partnering with Google makes so much sense given where they are on nanobanana and where they are on Gemini and

Starting point is 02:49:47 like just the cost per performance, even if you go crazy into generating a bunch of ads, it feels like a great partner there. How did the Google will kill you like narrative? It clearly didn't happen, but walk me through the various, it seems like you were getting that criticism of venture capitalists early days. How did that actually play out? How do you sit together in the market of ad buying? Because you do compete, even though you work together

Starting point is 02:50:12 on the cloud side. How is that relationship evolved? Why did the market play out the way it did? And was anything, were there any decisions where you think, like, we made a really key decision? If you talk to advertisers, let's use the Ridge Guys, as an example again, you want performance and you want to constantly be experimenting

Starting point is 02:50:32 and you don't want platform dependency. In a perfect world, you'd be spending like 10% of your budget here, 10% of your budget here, 10% so that one, as prices fluctuate, because they're all markets, right? Your business isn't, you know, and we saw the businesses that were, you know, in the DDC era that were dependent purely on, you know, a meta and how bad things could get. Totally.

Starting point is 02:50:55 Going bankrupt based on, you know, small changes and price. But anyways. Yeah, I mean, Google is a is sort of a bottom funnel channel for most dollars spent, right? If you're selling a wallet, you go to Google, you search for a wallet, rich wallet appears there. Well, the consumer already knows they want to find a wallet. We were always going to be additive to the Google advertising story, so they did a really good job finding us early and bending on us across multiple business slides. Cool. And it became a really, really good partnership. If you think about the business model we deploy, it's top of funnel, the bottom of funnel priced on revenue.

Starting point is 02:51:28 Facebook does that. And if Facebook caught everyone's attention for every single ad, for every single product in the world, where there would be no other market, Facebook would capture all the top funnel attention and Google would be bottom funnel. Turns out not everyone in the world notices an ad on Facebook. So along we come with these really immersive ads,

Starting point is 02:51:46 capture users' attention. When we were able to go partner with a wallet company and sell more wallets. So in a way, when you do it this way, they're getting an arbitrage. They're getting a customer that pays them immediately. they ship the product after the fact, and they make a spread on that sale. If they can do that, there's no budgetary constraint.

Starting point is 02:52:04 So we don't have a sales force at all, really, at the company, because we're not going and asking for budget. We're going and saying, look, if you can sell more wallets and make money on us, you'll scale to infinite, as many wallets as you can go get in inventory. And so with that business model, we became additive to these other platforms. I think really that's the only way a small brand or niche company like us was able to have the large-scale success. If you're advising, I don't know, public company CEO, $100 million in market cap,

Starting point is 02:52:32 but they have a platform and they're not monetizing with advertising. And they're considering rolling their own versus partnering with you or partnering with someone else. Like, what are the tradeoffs? We've seen Uber's added a bunch of ads, Netflix added ads. There's a chat jeep. He's going to have ads soon. What are the different levers that a CEO should think about pulling? Have you done partnerships?

Starting point is 02:52:54 how do you think about partnering with like those large scale new platforms? Yeah, we haven't yet. I think at some point in our narrative, we're going to get to a point where we're going to want to take our demand out. But we don't have access demand. Sure. So few advertisers today. If we got 100,000 advertisers, we'd probably be able to take demand out.

Starting point is 02:53:10 Sure. But I think a lot of times people forget, one, the technology is really complex. You've only had, in terms of a model like ours is scalable, us and then obviously met a very large scale that's been able to execute on this discovery type of product price on revenue. And so if there's only a couple of cases, it's not trivial, and people over trivialize it a lot. The other thing is the ad format has to be similar to the ad format that we're really good at. Otherwise, it takes a lot of work.

Starting point is 02:53:36 And because we've got such an immersive ad format, you don't tend to have that in other places. Now, in television, you do, and that's an area that's been interesting to us. And specifically, like, streaming or? Anywhere where you can see 15 to 30 second clip. Yeah. So streaming is a great. great example. Pre-roll video is a good example, but a native ad that someone's scrolling by and probably misses or a little banner ad, those aren't going to create a lot of intent. So if the ad format isn't going to create intent, that's not a great matchup for us. It's something we can execute on over time, and I think that'll benefit the publishers, because really these third-party publishers, they shouldn't build their own ad tech for the most part. And there's not another solution out there that's really that focus on them with Google maybe deprioritizing their third-party ad stuff.

Starting point is 02:54:24 Yeah. Yeah. What about Apple? I feel like there was a Rumble years ago about, you know, a developer kid in SDK for adding ads into mobile games. And it certainly hasn't stopped your growth. So what, yeah, how did that play out? Yeah, I mean, they have ads on search. They can monetize their own properties, presumably. I always say, look, if there's more ad inventory and more discovery, the P&L of the businesses and the market swells. So let's say Apple rolls out a great ads product. And you go to Apple Music, you discover it. game. Well, you weren't going to discover that game before. So now that developer, who will almost always be an arbitrage marketer, got a better arbitrage. Their business, their P&L grows. What happens when that happens? Well, they have more money to reinvest in us. They have more money to reinvest in meta. And so if performance is there, that's fantastic for the ecosystem. If it's non-performant, then the customers are getting ripped off. And these customers in these categories, they don't have the business model or the brand to get ripped off. I had seen a few pitches over the last couple years for businesses that we're trying to effectively create

Starting point is 02:55:28 app lovin specifically for LLM type products. Why or why not is that an interesting opportunity? Because I would assume the app lovin for LLM is app lovin. You probably never thought you'd be saying that statement. The day people started saying, I want to be the app lovin of this was when we knew we made it, I guess. Like LLMs, they don't have a lot of space. So you don't want to interrupt the user experience. You can't alter the output. The referral program you guys were talking about a few minutes ago. Yeah, one thing that was like a deep, like if somebody put in like a deep research query somewhere. It's waiting 10 minutes, wait 20 minutes. They're waiting maybe and you could accelerate it. You really sit in front of that arrow. It's veiling. Come on. Yeah. That's a non-viewed ad.

Starting point is 02:56:11 Yeah. Yeah. Yeah. I think it's going to be a lot more referral and embedded into the output and automated transactions, which again, for us is good. If you think about dollars spent, a lot of dollars today go to search when the user already knows what they want. If they go to the LLM and they get an output that's something that they want and they can click purchase, all is going to happen is those dollars are going to shift around from search to these LLM outputs. And that's good because that's already bottom funnel. The user knows that they want it. And in theory, it's good because it's just going to be cheaper to the advertiser.

Starting point is 02:56:41 So again, if it's effectively cheaper, that wallet company now has more money to go spend on a platform like ours, where they're truly getting discovery from the customer. Yeah. Talk to me about the company culture, the type of, a person that succeeds. Do you have economists on staff that are understanding the nature of the market? Do you have a lot of data scientists? Do you expect everyone to be able to think in statistical ways? Yeah, so it's engineering first for sure. Research science runs the models. Those are typically like very intelligent math people alongside they're also engineers. So they can implement

Starting point is 02:57:15 their own research. And then on the business team, most of the business people are math, econ, They know how to do Python, SQL queries. So we're mostly more technical across the board. Now, core advertising business, our business and engineering teams are around 300 to 350. So if you just slimmed it down to what drives are the 98% of the value of the company, it's a very, very small team. I always liked when I listen to Steve Jobs talking about the original Mac team. It's A players like to work with A players, and then they despise that, like, everything dilutes when you go to B and C's. And we've got a really cutthroat ecosystem where you're either an exceptionally strong IC

Starting point is 02:57:57 where you can output a ton or you're not going to make it. And so we end up churning through a lot of people who can't cut it. But the core team that remains loves working with each other because everyone's an A player. Yeah. How do you think about what's your head count plan look like over the next few years? Yeah. I'm assuming you want to multiply revenue by a lot. Without adding heads.

Starting point is 02:58:21 Yeah. I mean, look, I'd love a world or we could run with 20 people, but we're probably not getting back to that world. But like, I remember. Good old days. Yeah, I mean, up until 100, I knew the story and the name and every, every one that was at the business. And so we've surpassed that. But being run this lien allows us to maintain a really high quality. So even our go-to-market team on e-commerce, which every investor looks at is a huge opportunity.

Starting point is 02:58:43 It's 30 people today. It's really hard to add A players if you maintain a high-quality bar. and you don't need to go into the state of, look, we've just got hiring quotas. Second, you do that, I think you blow up your culture and your organization. So we're slow and steady. Now, the company's built on automate first, hire, second. And so LLMs let us do a ton. Most of the coding of the company is done by LLMs now.

Starting point is 02:59:07 A lot of the business functions are done by LLM. So we now are in a world where that culture is perfectly paired with this technology that's evolving so quickly. So we should be able to do more with less. And if we can do that, we're going to be able to scale the revenue without having to add heads. And then it's nice that the product's also so good. It sells itself. Well, thank you so much for coming by the TBPN Ultrodome. Nobody loves ads more than maybe than us than you maybe.

Starting point is 02:59:32 Yeah, the chat says because you're a gong respecter, a gong ally, we have to hit the gong. There we go. There we go. We'll talk about Axon later this week. Excited. And before we bring it up. our next guest, let me tell you about fin.AI, the number one AI agent for customer service, number one in performance benchmarks, number one in competitive bakeoffs, number one ranking on G2.

Starting point is 02:59:57 And let me also tell you about Adio, customer relationship magic. Adio is the AI need of CRM that builds scales and grows your company to the next level. We have David Senra live in the TBPN Ultradome. We have a video that we're going to play. Bring him on down. Come on in. Let's play the video of his newest addition to the collection, to the history, to the world of business and podcasting. Welcome to the show. Great to see you. 24 hours post launch.

Starting point is 03:00:27 Yes, sir. Yeah, what are we watching here? Oh, it's a drill. What is this? Do you have the video? The text? Yes, yes, I played it. Okay.

Starting point is 03:00:36 That's the video, but do you have the text that Rob put above it? What did he say? He said, the right way is the hard way. Jerry Seinfeld. Intro for it. Yes, they bought a Rolodex. This is for your new show. Break it down for us.

Starting point is 03:00:47 So this is why I was excited to partner with Rob Moore and Andrew Heumman on this because I've been friends with them for like three years. Yeah. And I know he said, I gesticulate wildly, so don't break your new microphone. So I knew how seriously they take their work. The first time I ever met Rob was in, they invited me to Malibu to check out their new studio. Yeah. We talked podcasting for nine straight hours.

Starting point is 03:01:08 Yeah. That's not an exaggeration. We stayed the same restaurant. We sat there for lunch. And then we started talking podcasting. And like four hours later, the waitress was like, serving dinner now. Do you want to stay? And we're like, yes. We had lunch and dinner. So I didn't even know they were doing this. They know that I was like kind of anti-intro for a podcast. I just like,

Starting point is 03:01:26 want to get right to it. They're like, we're going to make a bumper for you. I was like, I don't even know what a bumper is. So that's what they call this. And it's like, I was like, it's got to be short. We, that's the, oh, that's from the video. Exactly. Oh, it's all makes sense now. I remember seeing that and I thought, oh, it's just CGI or something. No, it's not CGI. That's the, man made. They bought a drill. They bought a Rolodex. And they printed out a bunch of companies and the founder names of people I've covered on founders. Of course. And then they did all the editing.

Starting point is 03:01:53 And so the whole thing is like the idea behind it was, okay, you turn on Netflix, you hear that sound, right? Or HBO you hear that sound. I'm cool with that. I don't want a super long intro. I want a very wealthy audience and we got to get to the goddamn point. So yeah, I just love that. It's like an indication of like how serious we're taking the new show. No AI.

Starting point is 03:02:10 We made it with a drill. No, I'm a human AI. Yeah, the first episode dropped Saturday, Sunday? Sunday. Sunday. They're coming out every other Sunday. Every other Sunday. How's the response been?

Starting point is 03:02:21 I'm like wired. I couldn't sleep last night. You couldn't sleep Saturday night too. I saw David, David, I called David yesterday. Even though you have an eight sleep, eight sleep.com. Yes. Get a pod five, five year warranty through an air street drop. I actually enjoyed your eight sleep ad integrated into the Daniel Eck episode.

Starting point is 03:02:38 I guess I was in solidarity with you last night, David, because I put up a 55. Oh, Ralph. I didn't. See, I'm thinking. in a friend's house in Malibu right now and I don't even have my eight sleep. We got to correct that. But yeah, it was so funny yesterday. I called David.

Starting point is 03:02:54 We're just catching up and we realized we were like 30 seconds on the same road. We're like 500 feet from each other. So we're like, let's just pull over and hang out. So I got to get the immediate reaction. But no, it really is there's, to me, one of the most satisfying things is watching ultra-talented people do a new thing and just demonstrate the level that they're on. And you and Rob and Andrew have done that. I appreciate that.

Starting point is 03:03:24 On the note of like you got busy people, you got to get right into it. Talk to me about the decision for how you think about the first question. I was listening to Dorcasch over the weekend. He kicks it off with like a pretty quick intro, but he does it. He does introduce the guest. My guest is Richard Sutton. He's very, D'Arcash is insanely talented. insanely talented.

Starting point is 03:03:46 Especially for a young kid. And you know who else is a talented podcaster, Chris Williamson, friend of us. Oh, he texts me today. And Chris kicks off with no introduction. He had a thousand episodes. A thousand episodes? He came out today with Matthew. Wow.

Starting point is 03:03:59 That's amazing. Incredible. What a run. But they do introductions very differently. They do, Chris kicks off with one short, punchy, like, ungoogulable question, basically. Like, there's no right answer to the question. he'll just say, you know, what is the value of hard work or what is your life's goal?

Starting point is 03:04:19 And then the person just tees off and it gets you right in the episode. Dorcasch gives you a little bit more of an intro and they go back and forth and then he hits them with a question. How do you think about opening an episode? Do you like the way you're doing it? Do you think it'll evolve?

Starting point is 03:04:30 We obviously open with one of the most over-the-top openings with your watch a TVPN and it's like this crazy thing that just evolved. How do you think about it? All right. I think the important thing is like I have, I'm not interviewing anybody. Yeah.

Starting point is 03:04:41 And so like the tagline is like conversations with the greatest living founders. Sure. I have zero energy to interview anybody. I have unlimited energy to have long-form, deep conversations. You guys have been on the receiving ends of this. Like when we have dinner, it's not like, oh, we're going to talk for 30 minutes. No, we talk for like four goddamn hours.

Starting point is 03:04:59 And so that's why I think a lot of some people came out and they're like, you're talking too much in the Daniel Lack episode. It's like, because it's a conversation. I like that. Well, it's important to, I think that your opportunity is to not be, it's tough if you're a podcast and people only watch for the guests. Yeah.

Starting point is 03:05:17 It's not a good position to be in, right? People watch Joe Rogan. There's a cohort of people that watch Joe Rogan just when Donald Trump's on. But a huge amount of people are watching it because Joe Rogan, they're like, who is he talking to today? That's what we're doing.

Starting point is 03:05:31 So I think somebody I saw, they did analysis on like 2,000 of Joe's episodes and he spoke like 45 or something percent of the time. Yep. It's just like, the whole genesis of this, I think I talked about it on the Daniel Eck episode. It's because, you know, Patrick O'Shaughnessy, who, again, I think is the best business interviewer in the world.

Starting point is 03:05:51 Like, if I'm going to start an interview show and compete with him, like, that's just stupid. Like, the main message of founders, the founders podcast is like the importance of differentiation. Yep. So like there's been a million, let's interview, you know, business people shows. I'm not, I'm going to lose that game. I only want to play a game that I can be the best in the world at. And so what Patrick, you know, we've been really close friends. we're brothers, we talk her every day.

Starting point is 03:06:12 We've have a ton, he's in these dinners with me. He's in these conversations, and it happens over and over again. And he's mentioned to me for a few years, he said it before I was on Colossus. He's just like, man, I really think, have you ever thought about doing a show where you talk to other people, not just you sitting in a room fucking reading books and doing the crazy shit that you do?

Starting point is 03:06:29 And he's like, I've been around this. Like no one can have these conversations. And so what really pushed me, and I said it on the Danielic episode, was me, him and Daniel had this super intense four-hour dinner. I don't meet with anybody. And the first time I met Michael Dell, I flew to Austin, five hours. We talked for five hours.

Starting point is 03:06:46 You think I sat silent? I can't sit silent for anything. And they don't want me to sit silent. The whole point is like you're like this weird, like human AI with all this knowledge. And they want to like prompt and like get me going. So we get in the car because I'm leaving New York and going back to Greenwich. I'm staying at Patrick's house at night. And the first thing he says to me when we get in with his driver, he's like, you have to start recording these.

Starting point is 03:07:09 He's just like, I, he goes, I've known Daniel for four years. You got more out of him four hours than I did him in four years. He goes, I talked 2% of time. He talked 49% of time. You talk 49% of time. He's just like, no one can speak to the soul of the founder like you. You have to start recording these. Yeah, it's fascinating.

Starting point is 03:07:25 There's that story about- That's the whole fucking idea. There's no interview. It's just like, I'm going to talk to the greatest living founders and we're going to have insane conversations that are not predictable. Yeah, there's that story about Mark Zuckerberg that he has an extremely high question to talk ratio. he lets other people talk a lot.

Starting point is 03:07:41 And that was seen as like, you know, bullish on his ability to like hoover up information. At the same time, in the content world, I feel like there is a natural urge. Like I noticed somebody sent me a short that we didn't post, that someone else posted, of me and Jordi talking to Mark Andresen. And it was clear that it was like Mark had a funny take. And so that was what would get the views. And so we were there kind of as like accoutreement to the main event. and there's a, like window dressing.

Starting point is 03:08:15 Like window dressing. And so there's just like, there is a natural pull in the algorithm for if you're starting an interview show. I should just get someone really, no, I know, that's the point. But I'm saying, I'm saying. No, no, no. What I'm saying is that is that if you're just purely view maxing, it's like go get a viral clip machine, point a camera on them and say, go and then just film.

Starting point is 03:08:35 And then say, what's next? What's on your, what's on your mind? And those two things, you have to actively resist the view capture, the algorithm capture, and that's what you're doing. Before, I didn't even know, you know, I don't view Max. I didn't legitimately, before we did the giant ramp deal for founders, right? I did not know how many people listened to founders. You didn't look at the analytics. That's great.

Starting point is 03:08:58 What does I tell you? I don't give a shit. I want the best audience. Yeah, it's the best. Not the biggest. The reason the ramp deal came about because somebody we cannot name who, if you know, if you're elite and tech, you know who this person is. Whatever the opposite of a public figure is, this person is.

Starting point is 03:09:13 It's interesting that probably the smartest and best person in tech behind the scenes, like no one knows, like, isn't she invisible? Literally came because she's like, I was at Michael Dell's house. He wouldn't stop talking about your podcast. He didn't know we were friends. I know 10 to 15 billionaires personally that listen to your show. We have to find a way to work together. That's the genesis of all this.

Starting point is 03:09:34 It wasn't how many downloads you get. It's you have the attention and the trust. of the most elite people in the world. Look at the guest list on the new David Senner show. David Senator by David Center. Yeah, that's hilarious. When you pull it up on Spotify in your car, it says, Daniel, David, David, Senator, David, David,

Starting point is 03:09:52 Senator, David, it said it three times. Are you ready? Are you ready? Somebody in the chat says, hey, David, I saw you at Malibu Country Mart yesterday. You're quite tall and handsome. Let's hear it. Are you ready to be, are you ready to be a household live chat?

Starting point is 03:10:05 This is a live chat. You know, Joe? Okay, this happened twice. So first of, I land. Because I came out here just for you guys, just so you know. Amazing. Like, legitimately, I was like, I don't want to, you guys are really important to me. I wanted, like, do this in person.

Starting point is 03:10:16 We got to talk about how much has changed since last month's here. It's here. But, yeah, this has had, it's kind of, I don't like to think about this, because I walk in some, I don't know which case this is. One happened at the Tesla Supercharger, and another at, I walk into Sun Life and get a smoothie. And, like, I walk in, there's only three people in there, and two guys turn it like, David Senra in the eye.

Starting point is 03:10:36 Like, what the fuck, bro? And then another guy was charging. You used to only get recognized that Amman properties. Now it's about to be... Everywhere. I can't tell you. The new show is going to be the first one that's associate with Yomon. That's very excited.

Starting point is 03:10:53 Yeah. I shouldn't say anything yet, but we're 99% sure that this deal is done. Your personal brands already associated with Amman. How much variation do you think they'll be in episode length? A ton. A ton. Because, again, I'm not... There's no formulaic.

Starting point is 03:11:09 There's no formula here. So, like, the way I think about it's, like, founders is just the books that give me energy that I'm excited to learn about. And in many cases, there are even books. Like, I think the, the Colossus profiles that Patrick and his team were making are so remarkable. I didn't know who Thomas Perfetti, Pfefferty is.

Starting point is 03:11:25 I read that, the article just came out. Do you guys know about this guy? He's worth $80 billion. Wow. He's 83 years old. I think he owns 80%. This is the AI will not make you rich article? No.

Starting point is 03:11:35 No, no. No, this is a guy. I think he started. interactive brokers. Oh, wow. Yeah. 80% of it. It's legit.

Starting point is 03:11:41 Okay. Yeah. He's a Hungarian. Size chat. Yeah. He's a Hungarian immigrant. But you read this profile, I'm like, I couldn't put it down. I was, I literally got him with the phone.

Starting point is 03:11:52 I was like, these things are incredible. And I was like, okay, I'm so excited that I read this. Even though it's not a book, I have to make an episode. So that's the episode I'm working on right now. Right? So that's how we had thing with my founders. Yeah. David Senra by David Senra by David Senor by David Senor.

Starting point is 03:12:03 Hosted. Hosted. Yeah. Why did you pick David Senora to host the David Center to host the David send a show. You could have, you could have taken anyone. You're leaving a lot of room to slot other hosts in. The funny, easy. The funny, the easiest response

Starting point is 03:12:15 was by Christian Keel. I think he used to be an astronaut. He's like, where'd you come up with the name? Yeah, yeah, yeah, that's great. But with the founders' podcast, I feel like you have, you're thinking on episode length, how you divide up the episodes, has changed

Starting point is 03:12:31 and you have a philosophy around it. And you've kind of narrowed the aperture a little bit as you've become more laser-focused on what it takes to deliver a quality episode. Do you think that will happen over time? No, because this is the thing, okay? I am a control freak, right? As you both know.

Starting point is 03:12:46 So, like, founders is a one-person thing. Sure. Like, to the point where I'm now hand-editing the transcripts. Yeah, yeah. Everybody's, like, outsource it. This is why I wanted to play this thing, because Jerry Seinfeld way, the hard way is the right way, right?

Starting point is 03:12:57 Right? It's like, I'm complete, I have complete control. It's, like, permissionless. I feel like I'm, like, you know, creating a statue or, like, some kind of art. It's just like a very intense love affair that I have,

Starting point is 03:13:08 an irrational love affair I have with that show. This one, now I have a partner. I don't even mean I have an entire team now, but you don't have a team on founders. What I mean is like there's some, I don't have control over what that person is going to say whether they're going to be interesting or not. And so what we've noticed is every single person

Starting point is 03:13:25 that we've recorded with, and we have a crazy guest list, they all came from the audience, founders' audience, because that way at least they have, like, they know what I'm into, who like what I like to talk about. They're more prone to say yes for an unreleased show.

Starting point is 03:13:39 But there's wide variance between like some, in many cases, in some cases, like, they're, we've recorded some with incredible people. And they halfway through, I'm like, oh, they want me to do a founder's episode for them. They just want me, literally, they like, I'll ask them a question. Like, hey, what about this?

Starting point is 03:13:56 And then they'll look at me and like, want me to talk. I mean, that has to be the same case for Joe Rogan. It's like, if you're getting invited to the mothership, you're like, yeah, I want to hear, I want to hang out with Joe Rogan today. Yeah. So, no, I have to be, like, what I'm learning, and, you know, this is new to me. It's just, like, loosen up a little bit.

Starting point is 03:14:14 Like, obviously, the quality stuff has to be good. Like, we have to edit ruthlessly. We're not going to, there's a bunch of different ideas. Most of them came from Daniel Eck that I can talk about, like, one example he gave me. And again, like, the guy is the most powerful person in podcasting. He's willing to just fucking counsel you. You should take his advice. So he's like, listen, you need to build up the original feed, right?

Starting point is 03:14:34 and then maybe a year in, whatever the time frame is, he's like, you need to have a bunch of separate feeds. So he's like, have the two-hour conversation that we had. And then edit it, have another feed where you edit those conversations down to like 30 minutes. And then he's like, then you need a third feed where it's just, hey, Daniel had 10 fucking clips that were incredible, just that.

Starting point is 03:14:52 Isn't that something that he should be solving on the product side? Like when I went to your show, I saw that I could do in Spotify. I watched it on Spotify. Thank you. respect to Daniac. But I noticed that I could switch between, as the artist intended, as the host and the guest intended. And I also left a comment and I had fun in the comment section.

Starting point is 03:15:13 But you could watch video, switch to audio. But then there were also short form video clips that were all linked to the same profile. I didn't have to go find David Senor Shorts. Why will there be different feeds when that's something that they could solve on their side? That's a really interesting. I don't know. I'd ask them. I'll ask them.

Starting point is 03:15:30 I'm not sure. But this was, we were in Stockholm. together. He told me this in January, so a few months. Because the original idea, I guess I should back up too. The original idea was to launch in April. Because again, like you guys have a deep relationship with RAMP, just like I do. And, you know, I had this idea, give credit to Eric and Kareem for the level of trust they put in me, where I was like, I want to do this where I have long-form conversations with, starting with the people that are in my audience. And like, I think it'll be good. But like, I think I would love if you guys are

Starting point is 03:16:03 percent of sponsor like done like immediately right and the original idea was to launch in april of next year no this past april but me and daniel couldn't sync schedules and i'm like i'm not launching until yeah he's got to be number one yeah because of the pod father no yeah it's just like the respect to have form uh that like i just knew like out of every i'm friends with a bunch of people have recorded but not to the level of like i've spent hours and hours talking to him i just knew he's just so wise and he's not, I just knew I could get stuff out of him that, you know, I hadn't seen him talk about anywhere else. And so we recorded that like a few months ago, but it was like the beginning of summer. And then I talked to the head of business at Spotify and the head

Starting point is 03:16:45 of product. And they're both like, you don't launch a show in the summer. And so that's why I got pushback to end of September. Yeah. Because everybody's like, they're just distracted. It's just like, you want maximum impact. Yeah. And so we waited for a while. Yeah. I mean, the launch video was... Do you expect to record and scrap episodes entirely? You think that'll be frequent, or do you think that if you bring the right energy, you can always get...

Starting point is 03:17:05 The unfair advantage we have is that, like, I do a monologue podcast and maybe better than anybody else in the world. And so, like, if they're going to sit there and just listen to be talked, I can go forever, like, forever. No, I don't think I'll scrap it. I just think you have to be...

Starting point is 03:17:19 Again, you have to respect the time of the audience. And so... But this is different. So you mentioned the founders, the formula for founders. I don't like doing them longer in an hour. So that way you're listening on 1.5. So it's like a 40 year career, 40 hours of reading, right,

Starting point is 03:17:36 that you can listen to in 45 minutes. That's the value proper founders. And that will give you a very high-end audience if you just idea after idea after idea after idea. This other one, like, if I can have an interesting four-hour conversation, then it can be four hours long. I don't think I can. I think we could talk for four hours

Starting point is 03:17:52 and you should edit it down to probably like two. You're going to do Gaston Glock's daughter? We should do the first. You see here? No. She's on X. No. Yeah, she took over the company.

Starting point is 03:18:01 I believe it's his daughter or maybe granddaughter. Did you listen to that? Her last? Oh, yeah, it's one of my favorites. We could do, the three of us could do a 24-hour podcast on business podcasting. For sure. We could do that. It would be boring as hell for the audience.

Starting point is 03:18:13 But you guys have invited me before. You like, you want to be here for the whole show. And I was like, I don't think that's a good idea. I can't live stream for three hours. I will get in trouble. It's easy. It's easy. Watch.

Starting point is 03:18:23 There's a woman in San Francisco charging $30,000 to name your baby. What do you think? You're using that? Were you using that? No, you shouldn't have a kid if you need somebody else to name it. Good take. See? It's easier than you think. No, but I will get in trouble.

Starting point is 03:18:36 I will literally get canceled because I don't have a filter. I can't do live. Remember, I can edit the shit. I don't like... I think the only thing you do is swear. You're very, you're very, very, very tame other than that. You just always divert. You're just always any... You naturally divert to history and talking about...

Starting point is 03:18:51 Okay, so that's... That's the key, where it's like if you go back to differentiation, right? This is... Somebody said, the funniest thing where there's like a Red Bull futurist. Yeah. Okay. He's like somebody ripped, literally said we're going to be the TPBN

Starting point is 03:19:05 of Europe. And then he quote treated them and no disrespect to them. I don't know them. He's like, of course, like the TV in Europe works two days. Two days away. To their credit, they leaned into it and they were laughing about it, which I really appreciate it. No, but that's the point. It's just like, God bless your soul.

Starting point is 03:19:21 It's just like, then you want to avoid direct comparison. And I can tell you right now, they haven't for podcast advice but there's I can't think of other people in the podcast industry that I'd want to avoid direct comparison than you two it's like you're gonna lose that because everybody's gonna be like let's compare this to TPBN what is this thing the old the old to the wide go to the wide go to the main go to the main wide we're here I don't know where the camera should be looking into there's like fucking

Starting point is 03:19:47 10 of them in here you know are they stand really serious because someone just copied our last set and we completely changed this set you don't understand how serious these two guys taking It's just stupid. It is a great line in zero to one, right? Where Peter Thiel says, if you're copying Mark Zuckerberg, you're not learning from him. Yeah. You copy, you don't copy the what, you copy the how.

Starting point is 03:20:08 Yeah. It drives me insane. So my point is like, I'm not going to start another interview show. I'm going to say, hey, I'm going to have conversations. Conversation show. No, not even that, but there's other conversation shows like Joe Rogan. What you just said, I can't help myself. Everything I hear, read, anything.

Starting point is 03:20:25 It all ties back to the shit that I learned from. founders. So it's like, oh, like the first time I met Michael Dell when I flew the first time, you know, we were talking, we went through Zach Dell's base power. We're going over and they were showing like me these dashboards, everything else. Within like the first five minutes, I'd make references to like Jim Casey, Fred Smith, Brad Jacobs. So this is like, this just comes naturally. Yeah. So it's like the important thing about podcasting or I think any kind of business has to be like natural to you. So in our private conversations, we pull ideas from Oguvie all the time or anybody. So yeah, that's exactly what I want to do.

Starting point is 03:20:57 And so people are like, I liked it, well, you should do this. I don't know what to tell you because it's like, this is just who I am. I can't, I'm not acting. Like, I can only be who I am. Yeah. I think the thing that's so, I think, I think the conversations that, uh, giving people the experience of you being at dinner with Daniel or Ovitz, these people, like, I, I feel like incredibly blessed that I get to go to dinner with you and hang out with our friends

Starting point is 03:21:25 and get, get the David Center experience. of like having a normal conversation, but then always, you're always pulling it back and like contextualizing it with history. It's like incredibly addicting. I could just, I could do,

Starting point is 03:21:36 you know, go to dinner for six hours because it's still going to just be wildly interesting. But now creating that product, I saw somebody, somebody in the chat was looking earlier. They were looking at the first few guests and they were like,

Starting point is 03:21:46 I might need to come back to this podcast later after I've made a few billion. But it's actually the exact opposite of that. It's like, no, just giving this product to something that's completely free. Anybody can just go,

Starting point is 03:21:57 listen to it and you get to be get the experience of being at dinner with you and and somebody that's created a 150 billion dollar company a 10 billion dollar company so there's two things on that so i did this like uh my very first viral episode of founders because i obviously don't try to go viral is when i did that i had dinner with charlie munger that i remember that and it like it went so viral fit founders the top of the charts over all for all podcasts the funniest thing about that was that uh didn't you get some inbound from journalists who were like he's he didn't want people to write about it and you got scoops like because you're basically like reporting on it or so it was like new information berkshire so again the reason why people love munger and people

Starting point is 03:22:36 people i admire it's like they're chaotic uncontrollable people i am one of those people right it's like rob is not going to rob is not going to be like we i'm going to try to box this guy just like there's fucking no way so berkshire was they're they're very like they know understand the power of story they control everything and charlie was uncontrollable and so It was not like I, they knew I was a podcaster. They knew like I was going to make a podcast about it. I wasn't going to record anything. Yeah.

Starting point is 03:23:04 And so Berkshire's PR was super mad because they kept, not because of the episode, they actually, the episode was very positive. Yeah, it was amazing. But what they were upset about is then you had all these journalists hitting them up. They're like, I thought. What did you mean by this?

Starting point is 03:23:18 I thought Charlie didn't give interviews. This one. And I didn't. Yeah, no. So he's like, he's not, I thought he doesn't give interviews. I'm like, I'm not a job. journalist. I'm an enthusiast. I fucking this guy's my idol. I have a

Starting point is 03:23:30 bust. I have a bronze bust of Charlie Munger in my house. I'm not a journalist. I'm an enthusiast. I'm a business enthusiast. That's true. That's true. It's true. And so then I feedback goes like, listen, if Charlie says delete the episode, I will delete the episode. There you go. But some fucking lady at Birchon,

Starting point is 03:23:47 no, that's not going to happen. Like, I'm very proud of that. Have him call me. Have him call me. Here's my number. And we found out Charlie didn't give a shit because he doesn't he didn't give a fuck. He's almost 100 years old. to answer your question, or to pick back on your comment, so I was doing all these, like, I had dinner with episodes where I'd have dinner,

Starting point is 03:24:03 and then I'd summarize in an episode, but you didn't hear anybody else. And so the original idea I had for David Center by David Center by hosting David Center is record, there's, there's, Jay-Z is one of the most fascinating people to me. Sure. Episode 238 of Founders is his autobiography. And he doesn't ever do podcasts, and he did a podcast that was a title exclusive

Starting point is 03:24:23 when he was trying to build up title. and the podcast was fascinating it's called Rap Radar. They rented, they were in a private dining room Beverly Hills. So it's the two hosts and then Jay-Z

Starting point is 03:24:34 and they just have lunch, right? You don't, he's not like chewing. No, I'm like, so you edit that stuff out. And I'm like, that's exactly what I want. Yeah, that's great.

Starting point is 03:24:44 Because what you just said, how much would people pay to have dinner with Brad Jacobs or Daniel Eck or Mike Ovitz? I'm going to New York again in like two weeks to have dinner with Ovitz again. Ovid's become like a crazy friend.

Starting point is 03:24:55 Like he texts me all the time. He just booked a guest. I can't tell you who. Let me tell you the funny Ovid's experience I had this this week. He goes, hey, I just had dinner with X. Okay, I'm not good, I can't say who. Elon Musk's son. No.

Starting point is 03:25:07 And he goes, he's unbelievable. You should get him for the new show. And I was like, do you think you could help convince him? He goes, that's why I'm fucking texting you. Two hours. No, but give him credit. He's like, what's your email? Like, the email to the person that can book this for me.

Starting point is 03:25:22 Yeah. Two hours later, date done. Like, he's a. He's a shark. There we go. That's fantastic. He's crazy. So I do want to say one other thing about being free.

Starting point is 03:25:31 I just retweeted my own tweet. There we go. Because I think this is a pat on the back. No, because I think this is important. What you just said about the fact that it's like a miracle of itself is free. And I was feeling this today because like the message I've been getting for the Daniel Rock episode. And it's like, podcasting is a miracle.

Starting point is 03:25:48 You can receive a world class education on any subject you want on demand while your eyes are busy for free. And I understand that I've been doing it for almost 10 years. It's still a fucking miracle to me. I still cannot believe that I can listen to these conversations. I can, it just, I never. I still, I still, I still feel that podcasts were much more important to my education than my traditional education. I thought back of, I was lucky to discover, like, Tim Ferriss when I was 18, right? and being able to, yeah, you just did the show with Tim.

Starting point is 03:26:28 And just like being able to sit on those conversations and learn about different perspectives on life and different ways to live and how to work and how to be healthy. All these different things were incredibly formative. And I just thought I was never, growing up, I was never in an environment where I was in any of those rooms at all, right? I had a, you know, there's some people that are that benefited from that. And I just feel like it is the greatest public,

Starting point is 03:26:53 public good. You nailed it. Like, okay, if you went to Harvard or Princeton or Ivy League, like my parents didn't even fucking graduate high school, that was just never going to be an option. I had to work full time. Like, maybe you get,

Starting point is 03:27:04 there are a small percentage of people that get world-class education and they're surrounded by the crazy network. Think about, like, the majority of my friends were all at Harvard. They're all, like, Jewish. These Jewish genius Harvard people, they were all at school at the same time.

Starting point is 03:27:18 And then you fast forward, they've been out of school for like 15 years, and they're all completely, like, absolutely dominating. Like I couldn't imagine what that experience would have been like to have that when you were 18. Like life-changing. And they're still doing deals together. You go to your friend's house for the weekend and the dad, just the conversations at the dinner table, right?

Starting point is 03:27:35 But now those conversations are actually widely accessible to everyone. Exactly. And so, like, I've learned more. I'm looking for this. You know, you were one of the first advertisers on founders years ago. And I'm actually looking for this text now, text that you sent me probably like 2023. way back. Yeah, but you were just like,

Starting point is 03:27:55 founders is essentially like the only show that I'm listening to these days. Yep. And I want the response, though, because my response was kind of arrogant. Well, no, I tried to buy it should be. I asked if you were open to being acquired, I think, and you were, yeah, and then you wanted,

Starting point is 03:28:12 you wanted to buy all, oh, no, you're like, the only podcast I listened to religiously. This is, hold on, this is going to be funny. Do you think this is an act? So this is back in March 22, 2003, is when this is happening. Okay, so you text me only a podcast I listen to religiously. I go, no days off, I'm coming for everyone. Fact check, true. Narrator, he did come for everyone. How do you explain your approach to work? Because I know you, in my experience, you're working around the

Starting point is 03:28:45 clock. Like, you're always on, but it doesn't always look like being in front of a, uh, you know, a computer, you know, editing an episode or whatever, but what, what, and what have you kind of pulled from the greats to kind of inform your own process? So I think like, man, sometimes, so I just did Tim Ferriss and sometimes like, I think I said this on Tim Ferriss, like, I say stuff and I'm like, I should not be admitting to this. But then every time I do, people are like, man, I'm so glad you said X, Y, or Z because like I feel that exact same way. What, do you procrastinate sometimes? No, no, no. So my issue was,

Starting point is 03:29:25 I think for years, episode 222 of founders, Ed Thorpe, and if you look at the subtitle of that episode, it's like my personal blueprint. And what I thought was interesting about Ed Thorpe was like, oh, this guy, like, mastered life more than almost anybody ever come across out of the 400 episodes of done. And what I mean by that is, like,

Starting point is 03:29:40 he actually was, like, balanced. Like, he, you know, he has a, he looked a crazy, adventurous life, he invented the system for counting cards and blackjack, built the world's first wearable computer Claude Shannon, created the world's first quantitative hedge fund, made more money you could ever spend. But he, he, like, took care of his health.

Starting point is 03:29:59 He was a good family man. And so that's what I meant. I was like, oh, he's, like, all well balanced. And I was like, this is my blueprint. And the closest thing I have to, like, a mentor, I would say, like an explicit mentor and kind of like an older brother is Sam Hinky. And I was on the phone with him recently. I mean, recently it was probably like nine months ago.

Starting point is 03:30:17 Because I kind of understood this about myself for a while. And I'm like, man, I think I'm fucking lying. Like, I think I'm lying when I say, like, that's my blueprint. Like, I don't like, I don't like, I don't like, I don't like, I like, I like, I think I'm much more like Enzo Ferrari than I am, like, Ed Thorpe. And, you know, I think Sam's point, he's just like, he's a way better person than I am, like, legitimately better human being than I am. And his point is, like, trying to pull me back in that direction.

Starting point is 03:30:42 And I just think that's, like, not going to happen. So the answer your question is, like, if my eyes are open, then I'm thinking about work. and I'm only thinking about work. And it doesn't feel like work because I just have this irrational obsession. Dude, I don't want to talk about this. I literally, like, this sounds so terrible. I literally, like, had tears in my eyes

Starting point is 03:31:02 when I woke up yesterday because this is so fucking important to me. And I knew that yesterday it was like a big turning point in my life, but I woke up from a private message from Daniel. And it's like, I read it and, like, immediately started fucking crying. Like, not like sobbing,

Starting point is 03:31:18 but like tears in my eyes because to have somebody that you respect so much, I'm really proud of you. Like, what you're putting out is excellent, really great. And to have that and also love it the way I do, it's just like the reason I said, Inzer Ferrari, because he said, a man dominated by passion such as mine,

Starting point is 03:31:38 like can't be cut in half. His whole point was just like, he wasn't good at anything else, but thinking about Ferrari. And so my work thing is just like, I'm infinitely curious. And so, like, today, you know, I worked out and then every other thing was just reading.

Starting point is 03:31:52 And I'm reading two books at the same time. I'm working on Founders episodes. I'm trying to respond to messages about the Daniel episode. I don't know. And then I just come over here and like, we're friends, but this is kind of work too. And then after this, like, I'm going to go back and work some more. And like, that's all I want to do. I have a follow-up question.

Starting point is 03:32:08 But first I'm going to tell you about adquick.com. Out-of-home advertising, easy and measurable. Say goodbye to the headaches of out-of-home advertising. Only Ad-Quick combines technology out-of-home expertise and data to enable efficient, seamless ad buying across the globe. I don't mind the ad. I love the ads. And I love how they're weaved in.

Starting point is 03:32:23 But if I'm sitting here, it's got to be a ramp out. Okay. Run it back. Run it back. Time is money. Say both. There we go. Corporate cards, bill payments, accounting, and a whole lot more all in one place.

Starting point is 03:32:34 Go to ramp.com. My question, do you think that there is a material difference between retired and live player, active CEO, post-FU money, post-economic, in terms of the shape of an interview, what you get out. If you took a random listener and they had no idea who Daniel Eck and Brad Jacobs and Michael Dell and all these different folks who you're going to interview were. I'm not interviewing them. What?

Starting point is 03:33:03 Yeah. Conversing. Talking to. Do you think that they would be able to tell, okay, this person's still in the arena. This person is post-economic, checked out, no bags. It's not a money thing. It's not a retirement thing. What is it?

Starting point is 03:33:17 it's like a personality thing. Wait till you see the Brad... I cannot believe some of the stuff we got on film that Brad Jacobs does. There we go. Because that guy just can't help himself. He is who he is. And obviously, like, he's a public company.

Starting point is 03:33:32 I don't know if he's a CEO. He's definitely chairman. But a lot of these happen to be public company CEO. So, like, I have to be, you know, they have comms people and they have all this crazy stuff and I'm not going to try... I'm not intentionally trying to get in trouble with anybody.

Starting point is 03:33:43 You know, again, I'm an enthusiast. I'm not a journalist. I just think, hey, you're brilliant. can I pull something out of you that other people, but I benefit a lot from talking to you. And so it's kind of selfish that I get all these lessons and this inspiration and this education from you. Are you cool us sitting down

Starting point is 03:33:56 and just recording the conversations if you have anyways? And so person like Brad doesn't give a shit. He is 68, he's got the most energy of anybody else that's been around. He also has nothing to be insecure about because he started eight separate billion-dollar companies. And he's a wild... He wrote the book on starting billion-dollar companies. And he's a wild boy.

Starting point is 03:34:13 And he's a wild boy. asking for, you know, like, are you bullish on this particular market that you may or may not have bags in? That's not relevant. I don't. We don't talk at all about current affairs. Sure. Current events. Which is why we're beautiful Yinn and Yang because obviously we're focused on what's happening in markets and the news.

Starting point is 03:34:30 Another huge point of differentiation against like a Rogan is like you go on Rogan, it's going to be like, well, let's react to the news right now. It's very different. No, and my point, like, I'm not, you know, I'm not interested. We talked about this. John was like, well, you do a podcast with me. I was like, what are we going to do? He's like, talk about tech news. I'm like, I'm talking about tech news.

Starting point is 03:34:47 No, I'm not going to talk about tech news. Would you, Raghav is asking, would you interview David Senra on David Senra by David Senor hosted by David Senra? Okay. Would you interview yourself? Okay. So the reclusive genius that we all know and then we can't name his name, literally called me yesterday. And had a long conversation about this. And he had a, I'm not good, I can't say, I'll tell you guys off there.

Starting point is 03:35:11 But he had a great idea. And he's like, once we should do this for episode 100. And I'm like, yeah. Well, there's another great idea. Go to getbezzle.com. Your bezel concierge is available. The source of you. Any Watch on the planet?

Starting point is 03:35:22 Seriously any watch? Get a look. Last question I have on my side, then we should grab lunch. I'm sorry. Do you think we're exiting the hobbyist era of podcasting? Era? I mean, it was. We had this conversation yesterday.

Starting point is 03:35:40 It was. I mean, there was a, there was a beautiful, there was a, there was about a, a, a, A decade and a half where you could start a podcast and have a, you know, I was doing media part-time. I was one of the worst defender. Yeah, but we had this conversation, no. You convinced me. I had this conversation when it came in because it was just obvious.

Starting point is 03:36:01 We'll talk about this later. The first fucking video I saw of yours. I was like, this guy, we had to be friends. He's generational talent. We had a guy who stopped John while we were at breakfast this morning, had no idea about TBPN, but was obsessed. with John's videos. And John's like, I haven't published a video.

Starting point is 03:36:17 I published a video and like over a year. Yes, I was, I wouldn't think it was that early, but I came in really early years early. So anyways, we, we and Joy just randomly talk about this because we can't help ourselves yesterday when we randomly ran into each other. This is where I like, I don't mean any disrespect to other people.

Starting point is 03:36:38 So we are alive. I know, this is, I have to be very careful. And that's a great place to end. No, no. So, um, okay, no decision. There's no way this guy's watching. No, no, no, no, no. It all gets back.

Starting point is 03:36:52 Okay, okay, right. Okay. I was induced into a state of rage. Because I was, I was at a dinner, and I was introduced to another podcaster. And they mentioned, like, again, they're from, like, the old school. Yeah. When you can, like, come in and, like, it's just different. And basically his point was just like, yeah, there's, like, I have to figure out, like, a new form.

Starting point is 03:37:15 There were a lot of people that saw podcasting as just like a marketing side hobby. But you can't say I need to fix my show. Yeah. And I need a different, like I needed like a new like take on my existing show. And also tell me that I'm going to take three months off to do something else that's not the podcast. Yeah. And I told him, I go, you're going to get fucked up. Like, because what's happening now, this is exactly what I said though.

Starting point is 03:37:39 But this is what's happening now is people used to come in and they're like, oh, I'm just going to like, I can do three other things. Podcasts could be your third, fourth, fifth most important thing in your life. Yeah, it's just 90 minutes once a week. Yeah, but now you have people that act like founders and entrepreneurs coming in and the conversation we had. So I think of like, obviously I approached this as if an Enzo Ferrari was doing what I was doing, if a monger was doing what I was doing, if a smogne was doing what I was doing, if Steve Jobs was doing or Edwin Lamb was doing what I was doing, what would they do? That's exactly how I think. And then I come into this and I'm like, oh, and I knew you guys think like this anyways, but what I was just telling them, because I was like, where did the rest of it? You had like 15 people last time I was here.

Starting point is 03:38:14 The interns went back to college. I'm like, no, they're going to learn more. I go, these two. Or one of our interns didn't go back to college. But I go, this is what I said. I go, these two are going to take this as far as it could possibly go. There's no way, if you have a front row seat to that, skip a fucking year of college. They're going to take this as far as possible.

Starting point is 03:38:33 You guys are not at, you're not podcasters, but you're entrepreneurs. And you just realize like this is the product. Look what it's got. Your fucking one year anniversary is what, four days from now? I think so. Yeah. We need to figure out when we... Episode 365, probably.

Starting point is 03:38:47 But look at what you guys have done. Well, no. We didn't start... Yeah, we didn't start doing daily. Until I cornered you at the Faena in Miami and started yelling out of love. That's the real that's the real founding date. Yeah. But my point being is like we were just looking at episode one, right, before I came on.

Starting point is 03:39:06 And look what you guys did, you're entrepreneurs. You've iterated tremendously in a year. Yeah, yeah. And so, yeah, good luck competing with John and Jordy. You're not going to do that. Yeah, it's also, I mean, the function of, you know, John spending years building up, figuring out how to create great content online, me starting my career, basically selling ads on the internet.

Starting point is 03:39:30 There's a bunch of different. And then, I don't know, it would not have been possible if we hadn't each spent a decade doing what we did prior to. But then there's also something that is not copyable, which is the chemistry between you two, which is why I was so emphatic about you need to take this more seriously. Not that you weren't, but like you have, there's just, you can't buy that chemistry. What do you mean chemistry? We're mortal enemies.

Starting point is 03:39:52 No. You just play characters. Yeah, we're just great actors. We're here in Hollywood. Yeah. There's so many times where I remember when I went to casting and they were like, you're going to play this guy. This is a lie. Don't lie to your audience.

Starting point is 03:40:01 I was like, oh, okay. plays the character, a character that's friends with the other guy with the other. There's so many, the audience, they're lying. There's so many times when I'm with one. And the other one calls the other one. And it's just like they love each other. Like I can't hang out anybody as much as you guys hang out and then still like them. So like I'm like nobody.

Starting point is 03:40:22 I just, nobody likes talking about. 225 is not going to bench itself. Yeah, yeah, yeah. We can't. Who else will I get to shotily spot me while they're on their phone and I'm dying? The sex threat. I've, I've gotten better at spotting.

Starting point is 03:40:37 But when in the morning when we're working out, It's a couple of times John almost lost his head because I was locked in on the news when he was benching. Dude, the joint workouts are working. I saw the pictures of you at the beach. Your back's getting, it's getting way wider. You guys are on the juice. No, no juice. I do, I do.

Starting point is 03:40:57 You know the secret? I eat a double smash burger every morning at 8 a.m. There's secrets. Oh, the secrets. This is a lot of fun. Let me talk about Wander. You want to sing with us? Find your happy place

Starting point is 03:41:12 Find your happy place Book of Wander with inspiring views Hotel Great amenities Rumi beds top tier cleaning and 24 7 concierge service It's a vacationer but better Corey Hayes I just have to say

Starting point is 03:41:20 I'm so happy for you I'm so proud of you Congratulations I feel lucky to have met you before you became a household name and nobody in my career has done more for me

Starting point is 03:41:35 than you with zero financial incentive, anything, anything along those lines. So it's... One, I thank you very much. The second part, I think, is really important. Because at the end of the day, if you're insanely talented,

Starting point is 03:41:49 it's like, people chase money when you should really be chasing relationships. Like, that is all, like, outside of work, I care deeply about the friendships that I build. And, like, I don't purposely. You know, I talk to Kareem about this. He's like, man, you're selling all these ads for other podcasters. You're doing all sorts of stuff. Cream from Ramp, one of the, maybe the most brilliant mind

Starting point is 03:42:08 working in finance today. technical mind working in finance today, listen to the Daniel Ruk episode, you'll see how he's similar to, how Kareem's working style is very similar to Daniels. And he's like, why don't you start like a, you know, business doing this? It's like, I do it because I love the people,

Starting point is 03:42:21 I love podcasters, and I love super talented people that take their work very seriously. And it's like, I like seeing my friends win. And no side hustles. Yeah, and that's another reason. Never side hustle. Well, let's ring the gong for the first episode.

Starting point is 03:42:33 Thank you. There we go. Great hit. Thank you for tuning in, folks. We love you. We'll see you tomorrow. See you tomorrow. Thank you.

TBPN - Claude Sonnet 4.5 Reactions, David Senra Live in The Ultradome | Dylan Field, Adam Foroughi, Mike Krieger, Jeff Weinstein, Adam Draper, James Hawkins, Erik Bernhardsson

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.